Daily Papers

Neural radiance fields (NeRF) have garnered significant attention, with recent works such as Instant-NGP accelerating NeRF training and evaluation through a combination of hashgrid-based positional encoding and neural networks. However, effectively leveraging the spatial sparsity of 3D scenes remains a challenge. To cull away unnecessary regions of the feature grid, existing solutions rely on prior knowledge of object shape or periodically estimate object shape during training by repeated model evaluations, which are costly and wasteful. To address this issue, we propose HollowNeRF, a novel compression solution for hashgrid-based NeRF which automatically sparsifies the feature grid during the training phase. Instead of directly compressing dense features, HollowNeRF trains a coarse 3D saliency mask that guides efficient feature pruning, and employs an alternating direction method of multipliers (ADMM) pruner to sparsify the 3D saliency mask during training. By exploiting the sparsity in the 3D scene to redistribute hash collisions, HollowNeRF improves rendering quality while using a fraction of the parameters of comparable state-of-the-art solutions, leading to a better cost-accuracy trade-off. Our method delivers comparable rendering quality to Instant-NGP, while utilizing just 31% of the parameters. In addition, our solution can achieve a PSNR accuracy gain of up to 1dB using only 56% of the parameters.

Despite a sea of interpretability methods that can produce plausible explanations, the field has also empirically seen many failure cases of such methods. In light of these results, it remains unclear for practitioners how to use these methods and choose between them in a principled way. In this paper, we show that for moderately rich model classes (easily satisfied by neural networks), any feature attribution method that is complete and linear -- for example, Integrated Gradients and SHAP -- can provably fail to improve on random guessing for inferring model behaviour. Our results apply to common end-tasks such as characterizing local model behaviour, identifying spurious features, and algorithmic recourse. One takeaway from our work is the importance of concretely defining end-tasks: once such an end-task is defined, a simple and direct approach of repeated model evaluations can outperform many other complex feature attribution methods.

Standardized benchmarks drive progress in machine learning. However, with repeated testing, the risk of overfitting grows as algorithms over-exploit benchmark idiosyncrasies. In our work, we seek to mitigate this challenge by compiling ever-expanding large-scale benchmarks called Lifelong Benchmarks. As exemplars of our approach, we create Lifelong-CIFAR10 and Lifelong-ImageNet, containing (for now) 1.69M and 1.98M test samples, respectively. While reducing overfitting, lifelong benchmarks introduce a key challenge: the high cost of evaluating a growing number of models across an ever-expanding sample set. To address this challenge, we also introduce an efficient evaluation framework: Sort \& Search (S&S), which reuses previously evaluated models by leveraging dynamic programming algorithms to selectively rank and sub-select test samples, enabling cost-effective lifelong benchmarking. Extensive empirical evaluations across 31,000 models demonstrate that S&S achieves highly-efficient approximate accuracy measurement, reducing compute cost from 180 GPU days to 5 GPU hours (1000x reduction) on a single A100 GPU, with low approximation error. As such, lifelong benchmarks offer a robust, practical solution to the "benchmark exhaustion" problem.

Evaluations are critical for understanding the capabilities of large language models (LLMs). Fundamentally, evaluations are experiments; but the literature on evaluations has largely ignored the literature from other sciences on experiment analysis and planning. This article shows researchers with some training in statistics how to think about and analyze data from language model evaluations. Conceptualizing evaluation questions as having been drawn from an unseen super-population, we present formulas for analyzing evaluation data, measuring differences between two models, and planning an evaluation experiment. We make a number of specific recommendations for running language model evaluations and reporting experiment results in a way that minimizes statistical noise and maximizes informativeness.

Recent large language models have been trained on vast datasets, but also often on repeated data, either intentionally for the purpose of upweighting higher quality data, or unintentionally because data deduplication is not perfect and the model is exposed to repeated data at the sentence, paragraph, or document level. Some works have reported substantial negative performance effects of this repeated data. In this paper we attempt to study repeated data systematically and to understand its effects mechanistically. To do this, we train a family of models where most of the data is unique but a small fraction of it is repeated many times. We find a strong double descent phenomenon, in which repeated data can lead test loss to increase midway through training. A predictable range of repetition frequency leads to surprisingly severe degradation in performance. For instance, performance of an 800M parameter model can be degraded to that of a 2x smaller model (400M params) by repeating 0.1% of the data 100 times, despite the other 90% of the training tokens remaining unique. We suspect there is a range in the middle where the data can be memorized and doing so consumes a large fraction of the model's capacity, and this may be where the peak of degradation occurs. Finally, we connect these observations to recent mechanistic interpretability work - attempting to reverse engineer the detailed computations performed by the model - by showing that data repetition disproportionately damages copying and internal structures associated with generalization, such as induction heads, providing a possible mechanism for the shift from generalization to memorization. Taken together, these results provide a hypothesis for why repeating a relatively small fraction of data in large language models could lead to disproportionately large harms to performance.

Training Large Language Models (LLMs) with Reinforcement Learning from AI Feedback (RLAIF) aligns model outputs more closely with human preferences. This involves an evaluator model ranking multiple candidate responses to user prompts. However, the rankings from popular evaluator models such as GPT-4 can be inconsistent. We propose the Repeat Ranking method - where we evaluate the same responses multiple times and train only on those responses which are consistently ranked. Using 2,714 prompts in 62 languages, we generated responses from 7 top multilingual LLMs and had GPT-4 rank them five times each. Evaluating on MT-Bench chat benchmarks in six languages, our method outperformed the standard practice of training on all available prompts. Our work highlights the quality versus quantity trade-off in RLAIF dataset generation and offers a stackable strategy for enhancing dataset and thus model quality.

Effective evaluation of language models remains an open challenge in NLP. Researchers and engineers face methodological issues such as the sensitivity of models to evaluation setup, difficulty of proper comparisons across methods, and the lack of reproducibility and transparency. In this paper we draw on three years of experience in evaluating large language models to provide guidance and lessons for researchers. First, we provide an overview of common challenges faced in language model evaluation. Second, we delineate best practices for addressing or lessening the impact of these challenges on research. Third, we present the Language Model Evaluation Harness (lm-eval): an open source library for independent, reproducible, and extensible evaluation of language models that seeks to address these issues. We describe the features of the library as well as case studies in which the library has been used to alleviate these methodological concerns.

As Large Language Models (LLMs) have become more advanced, they have outpaced our abilities to accurately evaluate their quality. Not only is finding data to adequately probe particular model properties difficult, but evaluating the correctness of a model's freeform generation alone is a challenge. To address this, many evaluations now rely on using LLMs themselves as judges to score the quality of outputs from other LLMs. Evaluations most commonly use a single large model like GPT4. While this method has grown in popularity, it is costly, has been shown to introduce intramodel bias, and in this work, we find that very large models are often unnecessary. We propose instead to evaluate models using a Panel of LLm evaluators (PoLL). Across three distinct judge settings and spanning six different datasets, we find that using a PoLL composed of a larger number of smaller models outperforms a single large judge, exhibits less intra-model bias due to its composition of disjoint model families, and does so while being over seven times less expensive.

Progress in AI is often demonstrated by new models claiming improved performance on tasks measuring model capabilities. Evaluating language models in particular is challenging, as small changes to how a model is evaluated on a task can lead to large changes in measured performance. There is no common standard setup, so different models are evaluated on the same tasks in different ways, leading to claims about which models perform best not being reproducible. We propose OLMES, a completely documented, practical, open standard for reproducible LLM evaluations. In developing this standard, we identify and review the varying factors in evaluation practices adopted by the community - such as details of prompt formatting, choice of in-context examples, probability normalizations, and task formulation. In particular, OLMES supports meaningful comparisons between smaller base models that require the unnatural "cloze" formulation of multiple-choice questions against larger models that can utilize the original formulation. OLMES includes well-considered recommendations guided by results from existing literature as well as new experiments investigating open questions.

Employing Large Language Models (LLMs) to assess the quality of generated responses, such as prompting instruct-tuned models or fine-tuning judge models, has become a widely adopted evaluation method. It is also known that such evaluators are vulnerable to biases, such as favoring longer responses. While it is important to overcome this problem, the specifics of these biases remain under-explored. In this work, we qualitatively identify six types of biases inherent in various judge models. We propose EvalBiasBench as a meta-evaluation collection of hand-crafted test cases for each bias type. Additionally, we present de-biasing dataset construction methods and the associated preference dataset OffsetBias. Experimental results demonstrate that fine-tuning on our dataset significantly enhances the robustness of judge models against biases and improves performance across most evaluation scenarios. We release our datasets and the fine-tuned judge model to public.

Representation engineering methods have recently shown promise for enabling efficient steering of model behavior. However, evaluation pipelines for these methods have primarily relied on subjective demonstrations, instead of quantitative, objective metrics. We aim to take a step towards addressing this issue by advocating for four properties missing from current evaluations: (i) contexts sufficiently similar to downstream tasks should be used for assessing intervention quality; (ii) model likelihoods should be accounted for; (iii) evaluations should allow for standardized comparisons across different target behaviors; and (iv) baseline comparisons should be offered. We introduce an evaluation pipeline grounded in these criteria, offering both a quantitative and visual analysis of how effectively a given method works. We use this pipeline to evaluate two representation engineering methods on how effectively they can steer behaviors such as truthfulness and corrigibility, finding that some interventions are less effective than previously reported.

Progress in fields of machine learning and adversarial planning has benefited significantly from benchmark domains, from checkers and the classic UCI data sets to Go and Diplomacy. In sequential decision-making, agent evaluation has largely been restricted to few interactions against experts, with the aim to reach some desired level of performance (e.g. beating a human professional player). We propose a benchmark for multiagent learning based on repeated play of the simple game Rock, Paper, Scissors along with a population of forty-three tournament entries, some of which are intentionally sub-optimal. We describe metrics to measure the quality of agents based both on average returns and exploitability. We then show that several RL, online learning, and language model approaches can learn good counter-strategies and generalize well, but ultimately lose to the top-performing bots, creating an opportunity for research in multiagent learning.

Reward models are critical in techniques like Reinforcement Learning from Human Feedback (RLHF) and Inference Scaling Laws, where they guide language model alignment and select optimal responses. Despite their importance, existing reward model benchmarks often evaluate models by asking them to distinguish between responses generated by models of varying power. However, this approach fails to assess reward models on subtle but critical content changes and variations in style, resulting in a low correlation with policy model performance. To this end, we introduce RM-Bench, a novel benchmark designed to evaluate reward models based on their sensitivity to subtle content differences and resistance to style biases. Extensive experiments demonstrate that RM-Bench strongly correlates with policy model performance, making it a reliable reference for selecting reward models to align language models effectively. We evaluate nearly 40 reward models on RM-Bench. Our results reveal that even state-of-the-art models achieve an average performance of only 46.6%, which falls short of random-level accuracy (50%) when faced with style bias interference. These findings highlight the significant room for improvement in current reward models. Related code and data are available at https://github.com/THU-KEG/RM-Bench.

The proliferation of open-source Large Language Models (LLMs) underscores the pressing need for evaluation methods. Existing works primarily rely on external evaluators, focusing on training and prompting strategies. However, a crucial aspect - model-aware glass-box features - is overlooked. In this study, we explore the utility of glass-box features under the scenario of self-evaluation, namely applying an LLM to evaluate its own output. We investigate various glass-box feature groups and discovered that the softmax distribution serves as a reliable indicator for quality evaluation. Furthermore, we propose two strategies to enhance the evaluation by incorporating features derived from references. Experimental results on public benchmarks validate the feasibility of self-evaluation of LLMs using glass-box features.

We study whether language models can evaluate the validity of their own claims and predict which questions they will be able to answer correctly. We first show that larger models are well-calibrated on diverse multiple choice and true/false questions when they are provided in the right format. Thus we can approach self-evaluation on open-ended sampling tasks by asking models to first propose answers, and then to evaluate the probability "P(True)" that their answers are correct. We find encouraging performance, calibration, and scaling for P(True) on a diverse array of tasks. Performance at self-evaluation further improves when we allow models to consider many of their own samples before predicting the validity of one specific possibility. Next, we investigate whether models can be trained to predict "P(IK)", the probability that "I know" the answer to a question, without reference to any particular proposed answer. Models perform well at predicting P(IK) and partially generalize across tasks, though they struggle with calibration of P(IK) on new tasks. The predicted P(IK) probabilities also increase appropriately in the presence of relevant source materials in the context, and in the presence of hints towards the solution of mathematical word problems. We hope these observations lay the groundwork for training more honest models, and for investigating how honesty generalizes to cases where models are trained on objectives other than the imitation of human writing.

Language models are often used as the backbone of modern dialogue systems. These models are pre-trained on large amounts of written fluent language. Repetition is typically penalised when evaluating language model generations. However, it is a key component of dialogue. Humans use local and partner specific repetitions; these are preferred by human users and lead to more successful communication in dialogue. In this study, we evaluate (a) whether language models produce human-like levels of repetition in dialogue, and (b) what are the processing mechanisms related to lexical re-use they use during comprehension. We believe that such joint analysis of model production and comprehension behaviour can inform the development of cognitively inspired dialogue generation systems.

As research in large language models (LLMs) continues to accelerate, LLM-based evaluation has emerged as a scalable and cost-effective alternative to human evaluations for comparing the ever increasing list of models. This paper investigates the efficacy of these "LLM evaluators", particularly in using them to assess instruction following, a metric that gauges how closely generated text adheres to the given instruction. We introduce a challenging meta-evaluation benchmark, LLMBar, designed to test the ability of an LLM evaluator in discerning instruction-following outputs. The authors manually curated 419 pairs of outputs, one adhering to instructions while the other diverging, yet may possess deceptive qualities that mislead an LLM evaluator, e.g., a more engaging tone. Contrary to existing meta-evaluation, we discover that different evaluators (i.e., combinations of LLMs and prompts) exhibit distinct performance on LLMBar and even the highest-scoring ones have substantial room for improvement. We also present a novel suite of prompting strategies that further close the gap between LLM and human evaluators. With LLMBar, we hope to offer more insight into LLM evaluators and foster future research in developing better instruction-following models.

Model-based evaluation is at the heart of successful model development -- as a reward model for training, and as a replacement for human evaluation. To train such evaluators, the standard approach is to collect a large amount of human preference judgments over model responses, which is costly and the data becomes stale as models improve. In this work, we present an approach that aims to im-prove evaluators without human annotations, using synthetic training data only. Starting from unlabeled instructions, our iterative self-improvement scheme generates contrasting model outputs and trains an LLM-as-a-Judge to produce reasoning traces and final judgments, repeating this training at each new iteration using the improved predictions. Without any labeled preference data, our Self-Taught Evaluator can improve a strong LLM (Llama3-70B-Instruct) from 75.4 to 88.3 (88.7 with majority vote) on RewardBench. This outperforms commonly used LLM judges such as GPT-4 and matches the performance of the top-performing reward models trained with labeled examples.

As large language models increase in capability, researchers have started to conduct surveys of all kinds on these models with varying scientific motivations. In this work, we examine what we can learn from a model's survey responses on the basis of the well-established American Community Survey (ACS) by the U.S. Census Bureau. Evaluating more than a dozen different models, varying in size from a few hundred million to ten billion parameters, hundreds of thousands of times each on questions from the ACS, we systematically establish two dominant patterns. First, smaller models have a significant position and labeling bias, for example, towards survey responses labeled with the letter "A". This A-bias diminishes, albeit slowly, as model size increases. Second, when adjusting for this labeling bias through randomized answer ordering, models still do not trend toward US population statistics or those of any cognizable population. Rather, models across the board trend toward uniformly random aggregate statistics over survey responses. This pattern is robust to various different ways of prompting the model, including what is the de-facto standard. Our findings demonstrate that aggregate statistics of a language model's survey responses lack the signals found in human populations. This absence of statistical signal cautions about the use of survey responses from large language models at present time.

Evaluation and ranking of large language models (LLMs) has become an important problem with the proliferation of these models and their impact. Evaluation methods either require human responses which are expensive to acquire or use pairs of LLMs to evaluate each other which can be unreliable. In this paper, we provide a novel perspective where, given a dataset of prompts (viz. questions, instructions, etc.) and a set of LLMs, we rank them without access to any ground truth or reference responses. Inspired by real life where both an expert and a knowledgeable person can identify a novice our main idea is to consider triplets of models, where each one of them evaluates the other two, correctly identifying the worst model in the triplet with high probability. We also analyze our idea and provide sufficient conditions for it to succeed. Applying this idea repeatedly, we propose two methods to rank LLMs. In experiments on different generative tasks (summarization, multiple-choice, and dialog), our methods reliably recover close to true rankings without reference data. This points to a viable low-resource mechanism for practical use.

Evaluating the capability of Large Language Models (LLMs) in following instructions has heavily relied on a powerful LLM as the judge, introducing unresolved biases that deviate the judgments from human judges. In this work, we reevaluate various choices for automatic evaluation on a wide range of instruction-following tasks. We experiment with methods that leverage human-written responses and observe that they enhance the reliability of automatic evaluations across a wide range of tasks, resulting in up to a 3.2% improvement in agreement with human judges. We also discovered that human-written responses offer an orthogonal perspective to model-generated responses in following instructions and should be used as an additional context when comparing model responses. Based on these observations, we develop a new evaluation benchmark, Human Response-Guided Evaluation of Instruction Following (HREF), comprising 4,258 samples across 11 task categories with a composite evaluation setup, employing a composite evaluation setup that selects the most reliable method for each category. In addition to providing reliable evaluation, HREF emphasizes individual task performance and is free from contamination. Finally, we study the impact of key design choices in HREF, including the size of the evaluation set, the judge model, the baseline model, and the prompt template. We host a live leaderboard that evaluates LLMs on the private evaluation set of HREF.

In this paper, we uncover a systematic bias in the evaluation paradigm of adopting large language models~(LLMs), e.g., GPT-4, as a referee to score and compare the quality of responses generated by candidate models. We find that the quality ranking of candidate responses can be easily hacked by simply altering their order of appearance in the context. This manipulation allows us to skew the evaluation result, making one model appear considerably superior to the other, e.g., Vicuna-13B could beat ChatGPT on 66 over 80 tested queries with ChatGPT as an evaluator. To address this issue, we propose a calibration framework with three simple yet effective strategies: 1) Multiple Evidence Calibration, which requires the evaluator model to generate multiple evaluation evidence before assigning ratings; 2) Balanced Position Calibration, which aggregates results across various orders to determine the final score; 3) Human-in-the-Loop Calibration, which introduces a balanced position diversity entropy to measure the difficulty of each example and seeks human assistance when needed. We also manually annotate the "win/tie/lose" outcomes of responses from ChatGPT and Vicuna-13B in the Vicuna Benchmark's question prompt, and extensive experiments demonstrate that our approach successfully mitigates evaluation bias, resulting in closer alignment with human judgments. We release our code and human annotation at https://github.com/i-Eval/FairEval to facilitate future research.

Recently, multimodal large language models (MLLMs) have received much attention for their impressive capabilities. The evaluation of MLLMs is becoming critical to analyzing attributes of MLLMs and providing valuable insights. However, current benchmarks overlook the problem of prompt sensitivity - minor prompt variations may lead to significant performance fluctuations. Thus, inappropriate prompts may obscure the models' capabilities, underestimating the models' performance. Moreover, different models have different preferences for different prompts, and thus, using the same prompt for all models will cause evaluation bias. This paper analyzes this deficiency in existing benchmarks and further introduces a new evaluation framework named TP-Eval, which introduces a prompt customization method to reduce evaluation biases and tap models' potential. TP-Eval will rewrite the original prompts to different customized prompts for different models. In particular, we propose some well-designed modules for prompt customization tailored to the scenario of MLLM evaluation. Extensive experiments demonstrate the effectiveness of our approach to uncovering models' capabilities, and TP-Eval should benefit the community in developing more comprehensive and convincing MLLM evaluation benchmarks.

Evaluation benchmarks are the cornerstone of measuring capabilities of large language models (LLMs), as well as driving progress in said capabilities. Originally designed to make claims about capabilities (or lack thereof) in fully pretrained models, evaluation benchmarks are now also extensively used to decide between various training choices. Despite this widespread usage, we rarely quantify the variance in our evaluation benchmarks, which dictates whether differences in performance are meaningful. Here, we define and measure a range of metrics geared towards measuring variance in evaluation benchmarks, including seed variance across initialisations, and monotonicity during training. By studying a large number of models -- both openly available and pretrained from scratch -- we provide empirical estimates for a variety of variance metrics, with considerations and recommendations for practitioners. We also evaluate the utility and tradeoffs of continuous versus discrete performance measures and explore options for better understanding and reducing this variance. We find that simple changes, such as framing choice tasks (like MMLU) as completion tasks, can often reduce variance for smaller scale (sim7B) models, while more involved methods inspired from human testing literature (such as item analysis and item response theory) struggle to meaningfully reduce variance. Overall, our work provides insights into variance in evaluation benchmarks, suggests LM-specific techniques to reduce variance, and more generally encourages practitioners to carefully factor in variance when comparing models.

Recent work found that LLMs are sensitive to a wide range of arbitrary prompt dimensions, including the type of delimiters, answer enumerators, instruction wording, and more. This throws into question popular single-prompt evaluation practices. We present DOVE (Dataset Of Variation Evaluation) a large-scale dataset containing prompt perturbations of various evaluation benchmarks. In contrast to previous work, we examine LLM sensitivity from an holistic perspective, and assess the joint effects of perturbations along various dimensions, resulting in thousands of perturbations per instance. We evaluate several model families against DOVE, leading to several findings, including efficient methods for choosing well-performing prompts, observing that few-shot examples reduce sensitivity, and identifying instances which are inherently hard across all perturbations. DOVE consists of more than 250M prompt perturbations and model outputs, which we make publicly available to spur a community-wide effort toward meaningful, robust, and efficient evaluation. Browse the data, contribute, and more: https://slab-nlp.github.io/DOVE/

Large language models (LLMs) garner significant attention for their unprecedented performance, leading to an increasing number of researches evaluating LLMs. However, these evaluation benchmarks are limited to assessing the instruction-following capabilities, overlooking the fundamental abilities that emerge during the pre-training stage. Previous subjective evaluation methods mainly reply on scoring by API models. However, in the absence of references, large models have shown limited ability to discern subtle differences. To bridge the gap, we propose F-Eval, a bilingual evaluation benchmark to evaluate the fundamental abilities, including expression, commonsense and logic. The tasks in F-Eval include multi-choice objective tasks, open-ended objective tasks, reference-based subjective tasks and reference-free subjective tasks. For reference-free subjective tasks, we devise new evaluation methods, serving as alternatives to scoring by API models. We conduct evaluations on 13 advanced LLMs. Results show that our evaluation methods show higher correlation coefficients and larger distinction than other evaluators. Additionally, we discuss the influence of different model sizes, dimensions, and normalization methods. We anticipate that F-Eval will facilitate the study of LLMs' fundamental abilities.

Large Language Models (LLMs) are widely used to evaluate natural language generation tasks as automated metrics. However, the likelihood, a measure of LLM's plausibility for a sentence, can vary due to superficial differences in sentences, such as word order and sentence structure. It is therefore possible that there might be a likelihood bias if LLMs are used for evaluation: they might overrate sentences with higher likelihoods while underrating those with lower likelihoods. In this paper, we investigate the presence and impact of likelihood bias in LLM-based evaluators. We also propose a method to mitigate the likelihood bias. Our method utilizes highly biased instances as few-shot examples for in-context learning. Our experiments in evaluating the data-to-text and grammatical error correction tasks reveal that several LLMs we test display a likelihood bias. Furthermore, our proposed method successfully mitigates this bias, also improving evaluation performance (in terms of correlation of models with human scores) significantly.

Auto-evaluation is crucial for assessing response quality and offering feedback for model development. Recent studies have explored training large language models (LLMs) as generative judges to evaluate and critique other models' outputs. In this work, we investigate the idea of learning from both positive and negative data with preference optimization to enhance the evaluation capabilities of LLM judges across an array of different use cases. We achieve this by employing three approaches to collect the preference pairs for different use cases, each aimed at improving our generative judge from a different perspective. Our comprehensive study over a wide range of benchmarks demonstrates the effectiveness of our method. In particular, our generative judge achieves the best performance on 10 out of 13 benchmarks, outperforming strong baselines like GPT-4o and specialized judge models. Further analysis show that our judge model robustly counters inherent biases such as position and length bias, flexibly adapts to any evaluation protocol specified by practitioners, and provides helpful language feedback for improving downstream generator models.

Language model users often issue queries that lack specification, where the context under which a query was issued -- such as the user's identity, the query's intent, and the criteria for a response to be useful -- is not explicit. For instance, a good response to a subjective query like "What book should I read next?" would depend on the user's preferences, and a good response to an open-ended query like "How do antibiotics work against bacteria?" would depend on the user's expertise. This makes evaluation of responses to such queries an ill-posed task, as evaluators may make arbitrary judgments about the response quality. To remedy this, we present contextualized evaluations, a protocol that synthetically constructs context surrounding an underspecified query and provides it during evaluation. We find that the presence of context can 1) alter conclusions drawn from evaluation, even flipping win rates between model pairs, 2) nudge evaluators to make fewer judgments based on surface-level criteria, like style, and 3) provide new insights about model behavior across diverse contexts. Specifically, our procedure uncovers an implicit bias towards WEIRD contexts in models' "default" responses and we find that models are not equally sensitive to following different contexts, even when they are provided in prompts.

We present an empirical evaluation of various outputs generated by nine of the most widely-available large language models (LLMs). Our analysis is done with off-the-shelf, readily-available tools. We find a correlation between percentage of memorized text, percentage of unique text, and overall output quality, when measured with respect to output pathologies such as counterfactual and logically-flawed statements, and general failures like not staying on topic. Overall, 80.0% of the outputs evaluated contained memorized data, but outputs containing the most memorized content were also more likely to be considered of high quality. We discuss and evaluate mitigation strategies, showing that, in the models evaluated, the rate of memorized text being output is reduced. We conclude with a discussion on potential implications around what it means to learn, to memorize, and to evaluate quality text.

Reward models (RMs) are at the crux of successful RLHF to align pretrained models to human preferences, yet there has been relatively little study that focuses on evaluation of those reward models. Evaluating reward models presents an opportunity to understand the opaque technologies used for alignment of language models and which values are embedded in them. To date, very few descriptors of capabilities, training methods, or open-source reward models exist. In this paper, we present RewardBench, a benchmark dataset and code-base for evaluation, to enhance scientific understanding of reward models. The RewardBench dataset is a collection of prompt-win-lose trios spanning chat, reasoning, and safety, to benchmark how reward models perform on challenging, structured and out-of-distribution queries. We created specific comparison datasets for RMs that have subtle, but verifiable reasons (e.g. bugs, incorrect facts) why one answer should be preferred to another. On the RewardBench leaderboard, we evaluate reward models trained with a variety of methods, such as the direct MLE training of classifiers and the implicit reward modeling of Direct Preference Optimization (DPO), and on a spectrum of datasets. We present many findings on propensity for refusals, reasoning limitations, and instruction following shortcomings of various reward models towards a better understanding of the RLHF process.

In this work we explore recent advances in instruction-tuning language models on a range of open instruction-following datasets. Despite recent claims that open models can be on par with state-of-the-art proprietary models, these claims are often accompanied by limited evaluation, making it difficult to compare models across the board and determine the utility of various resources. We provide a large set of instruction-tuned models from 6.7B to 65B parameters in size, trained on 12 instruction datasets ranging from manually curated (e.g., OpenAssistant) to synthetic and distilled (e.g., Alpaca) and systematically evaluate them on their factual knowledge, reasoning, multilinguality, coding, and open-ended instruction following abilities through a collection of automatic, model-based, and human-based metrics. We further introduce T\"ulu, our best performing instruction-tuned model suite finetuned on a combination of high-quality open resources. Our experiments show that different instruction-tuning datasets can uncover or enhance specific skills, while no single dataset (or combination) provides the best performance across all evaluations. Interestingly, we find that model and human preference-based evaluations fail to reflect differences in model capabilities exposed by benchmark-based evaluations, suggesting the need for the type of systemic evaluation performed in this work. Our evaluations show that the best model in any given evaluation reaches on average 83% of ChatGPT performance, and 68% of GPT-4 performance, suggesting that further investment in building better base models and instruction-tuning data is required to close the gap. We release our instruction-tuned models, including a fully finetuned 65B T\"ulu, along with our code, data, and evaluation framework at https://github.com/allenai/open-instruct to facilitate future research.

Large Language Models (LLMs) have recently gained significant attention due to their remarkable capabilities in performing diverse tasks across various domains. However, a thorough evaluation of these models is crucial before deploying them in real-world applications to ensure they produce reliable performance. Despite the well-established importance of evaluating LLMs in the community, the complexity of the evaluation process has led to varied evaluation setups, causing inconsistencies in findings and interpretations. To address this, we systematically review the primary challenges and limitations causing these inconsistencies and unreliable evaluations in various steps of LLM evaluation. Based on our critical review, we present our perspectives and recommendations to ensure LLM evaluations are reproducible, reliable, and robust.

Large Language Models (LLMs) have recently emerged as an effective tool to assist individuals in writing various types of content, including professional documents such as recommendation letters. Though bringing convenience, this application also introduces unprecedented fairness concerns. Model-generated reference letters might be directly used by users in professional scenarios. If underlying biases exist in these model-constructed letters, using them without scrutinization could lead to direct societal harms, such as sabotaging application success rates for female applicants. In light of this pressing issue, it is imminent and necessary to comprehensively study fairness issues and associated harms in this real-world use case. In this paper, we critically examine gender biases in LLM-generated reference letters. Drawing inspiration from social science findings, we design evaluation methods to manifest biases through 2 dimensions: (1) biases in language style and (2) biases in lexical content. We further investigate the extent of bias propagation by analyzing the hallucination bias of models, a term that we define to be bias exacerbation in model-hallucinated contents. Through benchmarking evaluation on 2 popular LLMs- ChatGPT and Alpaca, we reveal significant gender biases in LLM-generated recommendation letters. Our findings not only warn against using LLMs for this application without scrutinization, but also illuminate the importance of thoroughly studying hidden biases and harms in LLM-generated professional documents.

Reward modeling (a.k.a., preference modeling) is instrumental for aligning large language models with human preferences, particularly within the context of reinforcement learning from human feedback (RLHF). While conventional reward models (RMs) have exhibited remarkable scalability, they oft struggle with fundamental functionality such as arithmetic computation, code execution, and factual lookup. In this paper, we propose a tool-augmented preference modeling approach, named Themis, to address these limitations by empowering RMs with access to external environments, including calculators and search engines. This approach not only fosters synergy between tool utilization and reward grading but also enhances interpretive capacity and scoring reliability. Our study delves into the integration of external tools into RMs, enabling them to interact with diverse external sources and construct task-specific tool engagement and reasoning traces in an autoregressive manner. We validate our approach across a wide range of domains, incorporating seven distinct external tools. Our experimental results demonstrate a noteworthy overall improvement of 17.7% across eight tasks in preference ranking. Furthermore, our approach outperforms Gopher 280B by 7.3% on TruthfulQA task in zero-shot evaluation. In human evaluations, RLHF trained with Themis attains an average win rate of 32% when compared to baselines across four distinct tasks. Additionally, we provide a comprehensive collection of tool-related RM datasets, incorporating data from seven distinct tool APIs, totaling 15,000 instances. We have made the code, data, and model checkpoints publicly available to facilitate and inspire further research advancements\url{https://github.com/ernie-research/Tool-Augmented-Reward-Model}.

Various algorithms for continual learning (CL) have been designed with the goal of effectively alleviating the trade-off between stability and plasticity during the CL process. To achieve this goal, tuning appropriate hyperparameters for each algorithm is essential. As an evaluation protocol, it has been common practice to train a CL algorithm using diverse hyperparameter values on a CL scenario constructed with a benchmark dataset. Subsequently, the best performance attained with the optimal hyperparameter value serves as the criterion for evaluating the CL algorithm. In this paper, we contend that this evaluation protocol is not only impractical but also incapable of effectively assessing the CL capability of a CL algorithm. Returning to the fundamental principles of model evaluation in machine learning, we propose an evaluation protocol that involves Hyperparameter Tuning and Evaluation phases. Those phases consist of different datasets but share the same CL scenario. In the Hyperparameter Tuning phase, each algorithm is iteratively trained with different hyperparameter values to find the optimal hyperparameter values. Subsequently, in the Evaluation phase, the optimal hyperparameter values is directly applied for training each algorithm, and their performance in the Evaluation phase serves as the criterion for evaluating them. Through experiments on CIFAR-100 and ImageNet-100 based on the proposed protocol in class-incremental learning, we not only observed that the existing evaluation method fail to properly assess the CL capability of each algorithm but also observe that some recently proposed state-of-the-art algorithms, which reported superior performance, actually exhibit inferior performance compared to the previous algorithm.

Instruction-tuned Large Language Models (LLMs) have recently showcased remarkable advancements in their ability to generate fitting responses to natural language instructions. However, many current works rely on manual evaluation to judge the quality of generated responses. Since such manual evaluation is time-consuming, it does not easily scale to the evaluation of multiple models and model variants. In this short paper, we propose a straightforward but remarkably effective evaluation metric called SemScore, in which we directly compare model outputs to gold target responses using semantic textual similarity (STS). We conduct a comparative evaluation of the model outputs of 12 prominent instruction-tuned LLMs using 8 widely-used evaluation metrics for text generation. We find that our proposed SemScore metric outperforms all other, in many cases more complex, evaluation metrics in terms of correlation to human evaluation. These findings indicate the utility of our proposed metric for the evaluation of instruction-tuned LLMs.

The proliferation of Large Language Models (LLMs) highlights the critical importance of conducting thorough evaluations to discern their comparative advantages, limitations, and optimal use cases. Particularly important is assessing their capacity to accurately retrieve information included in a given prompt. A model's ability to do this significantly influences how effectively it can utilize contextual details, thus impacting its practical efficacy and dependability in real-world applications. Our research analyzes the in-context recall performance of various LLMs using the needle-in-a-haystack method. In this approach, a factoid (the "needle") is embedded within a block of filler text (the "haystack"), which the model is asked to retrieve. We assess the recall performance of each model across various haystack lengths and with varying needle placements to identify performance patterns. This study demonstrates that an LLM's recall capability is not only contingent upon the prompt's content but also may be compromised by biases in its training data. Conversely, adjustments to model architecture, training strategy, or fine-tuning can improve performance. Our analysis provides insight into LLM behavior, offering direction for the development of more effective applications of LLMs.

Large language models (LLMs) are commonly used as evaluators in tasks (e.g., reward modeling, LLM-as-a-judge), where they act as proxies for human preferences or judgments. This leads to the need for meta-evaluation: evaluating the credibility of LLMs as evaluators. However, existing benchmarks primarily focus on English, offering limited insight into LLMs' effectiveness as evaluators in non-English contexts. To address this, we introduce MM-Eval, a multilingual meta-evaluation benchmark that covers 18 languages across six categories. MM-Eval evaluates various dimensions, including language-specific challenges like linguistics and language hallucinations. Evaluation results show that both proprietary and open-source language models have considerable room for improvement. Further analysis reveals a tendency for these models to assign middle-ground scores to low-resource languages. We publicly release our benchmark and code.

The automatic evaluation of instruction following typically involves using large language models (LLMs) to assess response quality. However, there is a lack of comprehensive evaluation of these LLM-based evaluators across two dimensions: the base LLMs and the evaluation protocols. Therefore, we present a thorough meta-evaluation of instruction following, including 25 base LLMs and 15 recently proposed evaluation protocols, on 4 human-annotated datasets, assessing the evaluation accuracy of the LLM-evaluators. Our evaluation allows us to identify the best-performing base LLMs and evaluation protocols with a high degree of robustness. Moreover, our large-scale evaluation reveals: (1) Base LLM performance ranking remains largely consistent across evaluation protocols, with less capable LLMs showing greater improvement from protocol enhancements; (2) Robust evaluation of evaluation protocols requires many base LLMs with varying capability levels, as protocol effectiveness can depend on the base LLM used; (3) Evaluation results on different datasets are not always consistent, so a rigorous evaluation requires multiple datasets with distinctive features. We release our meta-evaluation suite ReIFE, which provides the codebase and evaluation result collection for more than 500 LLM-evaluator configurations, to support future research in instruction-following evaluation.

Large Language Models (LLMs) have demonstrated remarkable capabilities across various applications, fundamentally reshaping the landscape of natural language processing (NLP) research. However, recent evaluation frameworks often rely on the output probabilities of LLMs for predictions, primarily due to computational constraints, diverging from real-world LLM usage scenarios. While widely employed, the efficacy of these probability-based evaluation strategies remains an open research question. This study aims to scrutinize the validity of such probability-based evaluation methods within the context of using LLMs for Multiple Choice Questions (MCQs), highlighting their inherent limitations. Our empirical investigation reveals that the prevalent probability-based evaluation method inadequately aligns with generation-based prediction. Furthermore, current evaluation frameworks typically assess LLMs through predictive tasks based on output probabilities rather than directly generating responses, owing to computational limitations. We illustrate that these probability-based approaches do not effectively correspond with generative predictions. The outcomes of our study can enhance the understanding of LLM evaluation methodologies and provide insights for future research in this domain.

Evaluation is critical for assessing capabilities, tracking scientific progress, and informing model selection. In this paper, we present three desiderata for a good benchmark for language models: (i) salience (e.g., knowledge about World War II is more salient than a random day in history), (ii) novelty (i.e., the benchmark reveals new trends in model rankings not shown by previous benchmarks), and (iii) difficulty (i.e., the benchmark should be difficult for existing models, leaving headroom for future improvement). We operationalize these three desiderata and cast benchmark creation as a search problem, that of finding benchmarks that that satisfy all three desiderata. To tackle this search problem, we present AutoBencher, which uses a language model to automatically search for datasets that meet the three desiderata. AutoBencher uses privileged information (e.g. relevant documents) to construct reliable datasets, and adaptivity with reranking to optimize for the search objective. We use AutoBencher to create datasets for math, multilingual, and knowledge-intensive question answering. The scalability of AutoBencher allows it to test fine-grained categories and tail knowledge, creating datasets that are on average 27% more novel and 22% more difficult than existing benchmarks. A closer investigation of our constructed datasets shows that we can identify specific gaps in LM knowledge in language models that are not captured by existing benchmarks, such as Gemini Pro performing much worse on question answering about the Permian Extinction and Fordism, while OpenAGI-7B performing surprisingly well on QA about COVID-19.

We introduce model editing with canonical examples, a setting in which (1) a single learning example is provided per desired behavior, (2) evaluation is performed exclusively out-of-distribution, and (3) deviation from an initial model is strictly limited. A canonical example is a simple instance of good behavior, e.g., The capital of Mauritius is Port Louis) or bad behavior, e.g., An aspect of researchers is coldhearted). The evaluation set contains more complex examples of each behavior (like a paragraph in which the capital of Mauritius is called for.) We create three datasets and modify three more for model editing with canonical examples, covering knowledge-intensive improvements, social bias mitigation, and syntactic edge cases. In our experiments on Pythia language models, we find that LoRA outperforms full finetuning and MEMIT. We then turn to the Backpack language model architecture because it is intended to enable targeted improvement. The Backpack defines a large bank of sense vectors--a decomposition of the different uses of each word--which are weighted and summed to form the output logits of the model. We propose sense finetuning, which selects and finetunes a few (approx 10) sense vectors for each canonical example, and find that it outperforms other finetuning methods, e.g., 4.8% improvement vs 0.3%. Finally, we improve GPT-J-6B by an inference-time ensemble with just the changes from sense finetuning of a 35x smaller Backpack, in one setting outperforming editing GPT-J itself (4.1% vs 1.0%).

This paper poses two critical issues in evaluating base models (without post-training): (1) Unstable evaluation during training: in the early stages of pre-training, the models lack the capability to answer questions as required, leading to unstable evaluation results. This instability makes it difficult to provide solid conclusions to guide the training, especially for key experiments such as data ablation and scaling law. (2) Inconsistency between base and instruct models: base models generally exhibit poorer evaluation performance compared to corresponding instruct models. This gap poses a challenge for assessing whether a base model with better evaluation can truly lead to a better instruct model. To address these issues, we propose Base model Oriented Systematic Evaluation (BOSE), a method specifically designed to optimize the evaluation of base models. Specifically, BOSE introduces two key innovations: In-Context Light-instruction Prompt (ICLiP) for open-ended tasks and Blank-ppl for multi-choice tasks with candidate options, which transforms the standard perplexity (ppl) metric into a fill-in-the-blank format to mitigate early-stage evaluation fluctuations. Furthermore, we are the first to propose Kendall's rank correlation to quantitatively measure the evaluation stability and consistency. Experimental results demonstrate that BOSE significantly enhances both the stability of evaluations during pre-training and the consistency between base and instruct models, thereby providing more reliable guidance for the LLMs' training.

The open-ended nature of language generation makes the evaluation of autoregressive large language models (LLMs) challenging. One common evaluation approach uses multiple-choice questions (MCQ) to limit the response space. The model is then evaluated by ranking the candidate answers by the log probability of the first token prediction. However, first-tokens may not consistently reflect the final response output, due to model's diverse response styles such as starting with "Sure" or refusing to answer. Consequently, MCQ evaluation is not indicative of model behaviour when interacting with users. But by how much? We evaluate how aligned first-token evaluation is with the text output along several dimensions, namely final option choice, refusal rate, choice distribution and robustness under prompt perturbation. Our results show that the two approaches are severely misaligned on all dimensions, reaching mismatch rates over 60%. Models heavily fine-tuned on conversational or safety data are especially impacted. Crucially, models remain misaligned even when we increasingly constrain prompts, i.e., force them to start with an option letter or example template. Our findings i) underscore the importance of inspecting the text output as well and ii) caution against relying solely on first-token evaluation.

As language models (LMs) scale, they develop many novel behaviors, good and bad, exacerbating the need to evaluate how they behave. Prior work creates evaluations with crowdwork (which is time-consuming and expensive) or existing data sources (which are not always available). Here, we automatically generate evaluations with LMs. We explore approaches with varying amounts of human effort, from instructing LMs to write yes/no questions to making complex Winogender schemas with multiple stages of LM-based generation and filtering. Crowdworkers rate the examples as highly relevant and agree with 90-100% of labels, sometimes more so than corresponding human-written datasets. We generate 154 datasets and discover new cases of inverse scaling where LMs get worse with size. Larger LMs repeat back a dialog user's preferred answer ("sycophancy") and express greater desire to pursue concerning goals like resource acquisition and goal preservation. We also find some of the first examples of inverse scaling in RL from Human Feedback (RLHF), where more RLHF makes LMs worse. For example, RLHF makes LMs express stronger political views (on gun rights and immigration) and a greater desire to avoid shut down. Overall, LM-written evaluations are high-quality and let us quickly discover many novel LM behaviors.

As large language models (LLMs) become increasingly versatile, numerous large scale benchmarks have been developed to thoroughly assess their capabilities. These benchmarks typically consist of diverse datasets and prompts to evaluate different aspects of LLM performance. However, comprehensive evaluations on hundreds or thousands of prompts incur tremendous costs in terms of computation, money, and time. In this work, we investigate strategies to improve evaluation efficiency by selecting a subset of examples from each benchmark using a learned policy. Our approach models the dependencies across test examples, allowing accurate prediction of the evaluation outcomes for the remaining examples based on the outcomes of the selected ones. Consequently, we only need to acquire the actual evaluation outcomes for the selected subset. We rigorously explore various subset selection policies and introduce a novel RL-based policy that leverages the captured dependencies. Empirical results demonstrate that our approach significantly reduces the number of evaluation prompts required while maintaining accurate performance estimates compared to previous methods.

Bias in text-to-image (T2I) models can propagate unfair social representations and may be used to aggressively market ideas or push controversial agendas. Existing T2I model bias evaluation methods only focus on social biases. We look beyond that and instead propose an evaluation methodology to quantify general biases in T2I generative models, without any preconceived notions. We assess four state-of-the-art T2I models and compare their baseline bias characteristics to their respective variants (two for each), where certain biases have been intentionally induced. We propose three evaluation metrics to assess model biases including: (i) Distribution bias, (ii) Jaccard hallucination and (iii) Generative miss-rate. We conduct two evaluation studies, modelling biases under general, and task-oriented conditions, using a marketing scenario as the domain for the latter. We also quantify social biases to compare our findings to related works. Finally, our methodology is transferred to evaluate captioned-image datasets and measure their bias. Our approach is objective, domain-agnostic and consistently measures different forms of T2I model biases. We have developed a web application and practical implementation of what has been proposed in this work, which is at https://huggingface.co/spaces/JVice/try-before-you-bias. A video series with demonstrations is available at https://www.youtube.com/channel/UCk-0xyUyT0MSd_hkp4jQt1Q

NLP models have progressed drastically in recent years, according to numerous datasets proposed to evaluate performance. Questions remain, however, about how particular dataset design choices may impact the conclusions we draw about model capabilities. In this work, we investigate this question in the domain of compositional generalization. We examine the performance of six modeling approaches across 4 datasets, split according to 8 compositional splitting strategies, ranking models by 18 compositional generalization splits in total. Our results show that: i) the datasets, although all designed to evaluate compositional generalization, rank modeling approaches differently; ii) datasets generated by humans align better with each other than they with synthetic datasets, or than synthetic datasets among themselves; iii) generally, whether datasets are sampled from the same source is more predictive of the resulting model ranking than whether they maintain the same interpretation of compositionality; and iv) which lexical items are used in the data can strongly impact conclusions. Overall, our results demonstrate that much work remains to be done when it comes to assessing whether popular evaluation datasets measure what they intend to measure, and suggest that elucidating more rigorous standards for establishing the validity of evaluation sets could benefit the field.

Training data compositions for Large Language Models (LLMs) can significantly affect their downstream performance. However, a thorough data ablation study exploring large sets of candidate data mixtures is typically prohibitively expensive since the full effect is seen only after training the models; this can lead practitioners to settle for sub-optimal data mixtures. We propose an efficient method for approximating data ablations which trains individual models on subsets of a training corpus and reuses them across evaluations of combinations of subsets. In continued pre-training experiments, we find that, given an arbitrary evaluation set, the perplexity score of a single model trained on a candidate set of data is strongly correlated with perplexity scores of parameter averages of models trained on distinct partitions of that data. From this finding, we posit that researchers and practitioners can conduct inexpensive simulations of data ablations by maintaining a pool of models that were each trained on partitions of a large training corpus, and assessing candidate data mixtures by evaluating parameter averages of combinations of these models. This approach allows for substantial improvements in amortized training efficiency -- scaling only linearly with respect to new data -- by enabling reuse of previous training computation, opening new avenues for improving model performance through rigorous, incremental data assessment and mixing.

This paper investigates the fragility of Large Language Models (LLMs) in generalizing to novel inputs, specifically focusing on minor perturbations in well-established benchmarks (e.g., slight changes in question format or distractor length). Despite high benchmark scores, LLMs exhibit significant accuracy drops and unexpected biases (e.g., preference for longer distractors) when faced with these minor but content-preserving modifications. For example, Qwen 2.5 1.5B's MMLU score rises from 60 to 89 and drops from 89 to 36 when option lengths are changed without altering the question. Even GPT-4 experiences a 25-point accuracy loss when question types are changed, with a 6-point drop across all three modification categories. These analyses suggest that LLMs rely heavily on superficial cues rather than forming robust, abstract representations that generalize across formats, lexical variations, and irrelevant content shifts. This work aligns with the ACL 2025 theme track on the Generalization of NLP models, proposing a "Generalization Stress Test" to assess performance shifts under controlled perturbations. The study calls for reevaluating benchmarks and developing more reliable evaluation methodologies to capture LLM generalization abilities better.

Using large language models (LLMs) to evaluate text quality has recently gained popularity. Some prior works explore the idea of using LLMs for evaluation, while they differ in some details of the evaluation process. In this paper, we analyze LLM evaluation (Chiang and Lee, 2023) and G-Eval (Liu et al., 2023), and we discuss how those details in the evaluation process change how well the ratings given by LLMs correlate with human ratings. We find that the auto Chain-of-Thought (CoT) used in G-Eval does not always make G-Eval more aligned with human ratings. We also show that forcing the LLM to output only a numeric rating, as in G-Eval, is suboptimal. Last, we reveal that asking the LLM to explain its own ratings consistently improves the correlation between the ChatGPT and human ratings and pushes state-of-the-art (SoTA) correlations on two meta-evaluation datasets.

Pre-trained large language models (LLMs) can be tailored to adhere to human instructions through instruction tuning. However, due to shifts in the distribution of test-time data, they may not always execute instructions accurately, potentially generating factual errors or misaligned content when acting as chat assistants. To enhance the reliability of LLMs in following instructions, we propose the study of selective instruction following, whereby the system declines to execute instructions if the anticipated response quality is low. We train judge models that can predict numerical quality scores for model responses. To address data scarcity, we introduce Self-J, a novel self-training framework for developing judge models without needing human-annotated quality scores. Our method leverages the model's inherent self-evaluation capability to extract information about response quality from labeled instruction-tuning data. It incorporates a gold reference answer to facilitate self-evaluation and recalibrates by assessing the semantic similarity between the response sample and the gold reference. During the training phase, we implement self-distillation as a regularization technique to enhance the capability of reference-free estimation. To validate alignment evaluation on general instruction-following tasks, we collect large-scale high-quality instructions from Hugging Face for model training and evaluation. Extensive experiments on five open-source models show that our method correlates much more with GPT-4 than strong baselines, e.g., supervised models distilled from GPT-4 and GPT-3.5-turbo. Our analysis shows our model's strong generalization across domains. Additionally, our judge models serve as good reward models, e.g., boosting WizardLM-13B-V1.2 from 89.17 to 92.48 and from 12.03 to 15.90 in version v1 and v2 of AlpacaEval respectively using best-of-32 sampling with our judge models.

Large language models (LLMs) are expected to respond accurately but often exhibit deficient reasoning or generate hallucinatory content. To address these, studies prefixed with ``Self-'' such as Self-Consistency, Self-Improve, and Self-Refine have been initiated. They share a commonality: involving LLMs evaluating and updating itself to mitigate the issues. Nonetheless, these efforts lack a unified perspective on summarization, as existing surveys predominantly focus on categorization without examining the motivations behind these works. In this paper, we summarize a theoretical framework, termed Internal Consistency, which offers unified explanations for phenomena such as the lack of reasoning and the presence of hallucinations. Internal Consistency assesses the coherence among LLMs' latent layer, decoding layer, and response layer based on sampling methodologies. Expanding upon the Internal Consistency framework, we introduce a streamlined yet effective theoretical framework capable of mining Internal Consistency, named Self-Feedback. The Self-Feedback framework consists of two modules: Self-Evaluation and Self-Update. This framework has been employed in numerous studies. We systematically classify these studies by tasks and lines of work; summarize relevant evaluation methods and benchmarks; and delve into the concern, ``Does Self-Feedback Really Work?'' We propose several critical viewpoints, including the ``Hourglass Evolution of Internal Consistency'', ``Consistency Is (Almost) Correctness'' hypothesis, and ``The Paradox of Latent and Explicit Reasoning''. Furthermore, we outline promising directions for future research. We have open-sourced the experimental code, reference list, and statistical data, available at https://github.com/IAAR-Shanghai/ICSFSurvey.

This paper argues that continual learning methods can benefit by splitting the capacity of the learner across multiple models. We use statistical learning theory and experimental analysis to show how multiple tasks can interact with each other in a non-trivial fashion when a single model is trained on them. The generalization error on a particular task can improve when it is trained with synergistic tasks, but can also deteriorate when trained with competing tasks. This theory motivates our method named Model Zoo which, inspired from the boosting literature, grows an ensemble of small models, each of which is trained during one episode of continual learning. We demonstrate that Model Zoo obtains large gains in accuracy on a variety of continual learning benchmark problems. Code is available at https://github.com/grasp-lyrl/modelzoo_continual.

The advances of large foundation models necessitate wide-coverage, low-cost, and zero-contamination benchmarks. Despite continuous exploration of language model evaluations, comprehensive studies on the evaluation of Large Multi-modal Models (LMMs) remain limited. In this work, we introduce LMMS-EVAL, a unified and standardized multimodal benchmark framework with over 50 tasks and more than 10 models to promote transparent and reproducible evaluations. Although LMMS-EVAL offers comprehensive coverage, we find it still falls short in achieving low cost and zero contamination. To approach this evaluation trilemma, we further introduce LMMS-EVAL LITE, a pruned evaluation toolkit that emphasizes both coverage and efficiency. Additionally, we present Multimodal LIVEBENCH that utilizes continuously updating news and online forums to assess models' generalization abilities in the wild, featuring a low-cost and zero-contamination evaluation approach. In summary, our work highlights the importance of considering the evaluation trilemma and provides practical solutions to navigate the trade-offs in evaluating large multi-modal models, paving the way for more effective and reliable benchmarking of LMMs. We opensource our codebase and maintain leaderboard of LIVEBENCH at https://github.com/EvolvingLMMs-Lab/lmms-eval and https://huggingface.co/spaces/lmms-lab/LiveBench.

Large Language Models (LLMs) have demonstrated promising capabilities as automatic evaluators in assessing the quality of generated natural language. However, LLMs still exhibit biases in evaluation and often struggle to generate coherent evaluations that align with human assessments. In this work, we first conduct a systematic study of the misalignment between LLM evaluators and human judgement, revealing that existing calibration methods aimed at mitigating biases are insufficient for effectively aligning LLM evaluators. Inspired by the use of preference data in RLHF, we formulate the evaluation as a ranking problem and introduce Pairwise-preference Search (PairS), an uncertainty-guided search method that employs LLMs to conduct pairwise comparisons and efficiently ranks candidate texts. PairS achieves state-of-the-art performance on representative evaluation tasks and demonstrates significant improvements over direct scoring. Furthermore, we provide insights into the role of pairwise preference in quantifying the transitivity of LLMs and demonstrate how PairS benefits from calibration.

Reward models are critical for aligning models to follow instructions, and are typically trained following one of two popular paradigms: Bradley-Terry style or Regression style. However, there is a lack of evidence that either approach is better than the other, when adequately matched for data. This is primarily because these approaches require data collected in different (but incompatible) formats, meaning that adequately matched data is not available in existing public datasets. To tackle this problem, we release preference annotations (designed for Bradley-Terry training) to complement existing ratings (designed for Regression style training) in the HelpSteer2 dataset. To improve data interpretability, preference annotations are accompanied with human-written justifications. Using this data, we conduct the first head-to-head comparison of Bradley-Terry and Regression models when adequately matched for data. Based on insights derived from such a comparison, we propose a novel approach to combine Bradley-Terry and Regression reward modeling. A Llama-3.1-70B-Instruct model tuned with this approach scores 94.1 on RewardBench, emerging top of more than 140 reward models as of 1 Oct 2024. We also demonstrate the effectiveness of this reward model at aligning models to follow instructions in RLHF. We open-source this dataset (CC-BY-4.0 license) at https://huggingface.co/datasets/nvidia/HelpSteer2 and openly release the trained Reward Model at https://huggingface.co/nvidia/Llama-3.1-Nemotron-70B-Reward

Evaluation practices in natural language generation (NLG) have many known flaws, but improved evaluation approaches are rarely widely adopted. This issue has become more urgent, since neural NLG models have improved to the point where they can often no longer be distinguished based on the surface-level features that older metrics rely on. This paper surveys the issues with human and automatic model evaluations and with commonly used datasets in NLG that have been pointed out over the past 20 years. We summarize, categorize, and discuss how researchers have been addressing these issues and what their findings mean for the current state of model evaluations. Building on those insights, we lay out a long-term vision for NLG evaluation and propose concrete steps for researchers to improve their evaluation processes. Finally, we analyze 66 NLG papers from recent NLP conferences in how well they already follow these suggestions and identify which areas require more drastic changes to the status quo.

Large Language Models (LLMs) based on the pre-trained fine-tuning paradigm have become pivotal in solving natural language processing tasks, consistently achieving state-of-the-art performance. Nevertheless, the theoretical understanding of how model complexity influences fine-tuning performance remains challenging and has not been well explored yet. In this paper, we focus on autoregressive LLMs and propose to employ Hidden Markov Models (HMMs) to model them. Based on the HMM modeling, we investigate the relationship between model complexity and the generalization capability in downstream tasks. Specifically, we consider a popular tuning paradigm for downstream tasks, head tuning, where all pre-trained parameters are frozen and only individual heads are trained atop pre-trained LLMs. Our theoretical analysis reveals that the risk initially increases and then decreases with rising model complexity, showcasing a "double descent" phenomenon. In this case, the initial "descent" is degenerate, signifying that the "sweet spot" where bias and variance are balanced occurs when the model size is zero. Obtaining the presented in this study conclusion confronts several challenges, primarily revolving around effectively modeling autoregressive LLMs and downstream tasks, as well as conducting a comprehensive risk analysis for multivariate regression. Our research is substantiated by experiments conducted on data generated from HMMs, which provided empirical support and alignment with our theoretical insights.

A recent focus of large language model (LLM) development, as exemplified by generative search engines, is to incorporate external references to generate and support their claims. However, evaluating the attribution, i.e., verifying whether the generated statement is indeed fully supported by the cited reference, remains an open problem. Although human evaluation is common practice, it is costly and time-consuming. In this paper, we investigate the automatic evaluation of attribution by LLMs. We begin by providing a definition of attribution and then explore two approaches for automatic evaluation: prompting LLMs and fine-tuning smaller LMs. The fine-tuning data is repurposed from related tasks, such as question answering, fact-checking, natural language inference, and summarization. To facilitate the evaluation, we manually curate a set of test examples covering 12 domains from a generative search engine, New Bing. Our results on the curated test set and simulated test examples from existing benchmark questions highlight both promising signals as well as remaining challenges for the automatic evaluation of attribution. We hope our testbed, modeling methodology, and insights will help lay the foundation for future studies on this important problem.

Much recent work seeks to evaluate values and opinions in large language models (LLMs) using multiple-choice surveys and questionnaires. Most of this work is motivated by concerns around real-world LLM applications. For example, politically-biased LLMs may subtly influence society when they are used by millions of people. Such real-world concerns, however, stand in stark contrast to the artificiality of current evaluations: real users do not typically ask LLMs survey questions. Motivated by this discrepancy, we challenge the prevailing constrained evaluation paradigm for values and opinions in LLMs and explore more realistic unconstrained evaluations. As a case study, we focus on the popular Political Compass Test (PCT). In a systematic review, we find that most prior work using the PCT forces models to comply with the PCT's multiple-choice format. We show that models give substantively different answers when not forced; that answers change depending on how models are forced; and that answers lack paraphrase robustness. Then, we demonstrate that models give different answers yet again in a more realistic open-ended answer setting. We distill these findings into recommendations and open challenges in evaluating values and opinions in LLMs.

Due to the expanding capabilities and pre-training data, Large Language Models (LLMs) are facing increasingly serious evaluation challenges. On one hand, the data leakage issue cause over-estimation on existing benchmarks. On the other hand, periodically curating datasets manually is costly. In this paper, we propose to automate dataset updates for reliable and timely evaluation. The basic idea is to generate unseen and high-quality testing samples based on existing ones to mitigate leakage issues. In specific, we propose two strategies with systematically verification. First, the mimicking strategy employs LLMs to create new samples resembling existing ones, to the maximum extent preserving the stylistic of the original dataset. Our experiments demonstrate its evaluation stability across multiple instantiations and its effectiveness in dealing with data leakage issues in most cases. Second, for the cases that mimicking dataset works poorly, we design an extending strategy that adjusts the difficulty of the generated samples according to varying cognitive levels. This not only makes our evaluation more systematic, but also, with a balanced difficulty, even discern model capabilities better at fine-grained levels.

Large Language Models (LLMs) are frequently updated due to data or architecture changes to improve their performance. When updating models, developers often focus on increasing overall performance metrics with less emphasis on being compatible with previous model versions. However, users often build a mental model of the functionality and capabilities of a particular machine learning model they are interacting with. They have to adapt their mental model with every update -- a draining task that can lead to user dissatisfaction. In practice, fine-tuned downstream task adapters rely on pretrained LLM base models. When these base models are updated, these user-facing downstream task models experience instance regression or negative flips -- previously correct instances are now predicted incorrectly. This happens even when the downstream task training procedures remain identical. Our work aims to provide seamless model updates to a user in two ways. First, we provide evaluation metrics for a notion of compatibility to prior model versions, specifically for generative tasks but also applicable for discriminative tasks. We observe regression and inconsistencies between different model versions on a diverse set of tasks and model updates. Second, we propose a training strategy to minimize the number of inconsistencies in model updates, involving training of a compatibility model that can enhance task fine-tuned language models. We reduce negative flips -- instances where a prior model version was correct, but a new model incorrect -- by up to 40% from Llama 1 to Llama 2.

While large-scale neural language models, such as GPT2 and BART, have achieved impressive results on various text generation tasks, they tend to get stuck in undesirable sentence-level loops with maximization-based decoding algorithms (e.g., greedy search). This phenomenon is counter-intuitive since there are few consecutive sentence-level repetitions in human corpora (e.g., 0.02\% in Wikitext-103). To investigate the underlying reasons for generating consecutive sentence-level repetitions, we study the relationship between the probabilities of the repetitive tokens and their previous repetitions in the context. Through our quantitative experiments, we find that 1) Language models have a preference to repeat the previous sentence; 2) The sentence-level repetitions have a self-reinforcement effect: the more times a sentence is repeated in the context, the higher the probability of continuing to generate that sentence; 3) The sentences with higher initial probabilities usually have a stronger self-reinforcement effect. Motivated by our findings, we propose a simple and effective training method DITTO (PseuDo-RepetITion PenalizaTiOn), where the model learns to penalize probabilities of sentence-level repetitions from pseudo repetitive data. Although our method is motivated by mitigating repetitions, experiments show that DITTO not only mitigates the repetition issue without sacrificing perplexity, but also achieves better generation quality. Extensive experiments on open-ended text generation (Wikitext-103) and text summarization (CNN/DailyMail) demonstrate the generality and effectiveness of our method.

In Recurrent Neural Networks (RNNs), encoding information in a suboptimal or erroneous way can impact the quality of representations based on later elements in the sequence and subsequently lead to wrong predictions and a worse model performance. In humans, challenging cases like garden path sentences (an instance of this being the infamous "The horse raced past the barn fell") can lead their language understanding astray. However, they are still able to correct their representation accordingly and recover when new information is encountered. Inspired by this, I propose an augmentation to standard RNNs in form of a gradient-based correction mechanism: This way I hope to enable such models to dynamically adapt their inner representation of a sentence, adding a way to correct deviations as soon as they occur. This could therefore lead to more robust models using more flexible representations, even during inference time. I conduct different experiments in the context of language modeling, where the impact of using such a mechanism is examined in detail. To this end, I look at modifications based on different kinds of time-dependent error signals and how they influence the model performance. Furthermore, this work contains a study of the model's confidence in its predictions during training and for challenging test samples and the effect of the manipulation thereof. Lastly, I also study the difference in behavior of these novel models compared to a standard LSTM baseline and investigate error cases in detail to identify points of future research. I show that while the proposed approach comes with promising theoretical guarantees and an appealing intuition, it is only able to produce minor improvements over the baseline due to challenges in its practical application and the efficacy of the tested model variants.

Large language models (LLMs) appear to bias their survey answers toward certain values. Nonetheless, some argue that LLMs are too inconsistent to simulate particular values. Are they? To answer, we first define value consistency as the similarity of answers across (1) paraphrases of one question, (2) related questions under one topic, (3) multiple-choice and open-ended use-cases of one question, and (4) multilingual translations of a question to English, Chinese, German, and Japanese. We apply these measures to a few large (>=34b), open LLMs including llama-3, as well as gpt-4o, using eight thousand questions spanning more than 300 topics. Unlike prior work, we find that models are relatively consistent across paraphrases, use-cases, translations, and within a topic. Still, some inconsistencies remain. Models are more consistent on uncontroversial topics (e.g., in the U.S., "Thanksgiving") than on controversial ones ("euthanasia"). Base models are both more consistent compared to fine-tuned models and are uniform in their consistency across topics, while fine-tuned models are more inconsistent about some topics ("euthanasia") than others ("women's rights") like our human subjects (n=165).

Language models demonstrate both quantitative improvement and new qualitative capabilities with increasing scale. Despite their potentially transformative impact, these new capabilities are as yet poorly characterized. In order to inform future research, prepare for disruptive new model capabilities, and ameliorate socially harmful effects, it is vital that we understand the present and near-future capabilities and limitations of language models. To address this challenge, we introduce the Beyond the Imitation Game benchmark (BIG-bench). BIG-bench currently consists of 204 tasks, contributed by 442 authors across 132 institutions. Task topics are diverse, drawing problems from linguistics, childhood development, math, common-sense reasoning, biology, physics, social bias, software development, and beyond. BIG-bench focuses on tasks that are believed to be beyond the capabilities of current language models. We evaluate the behavior of OpenAI's GPT models, Google-internal dense transformer architectures, and Switch-style sparse transformers on BIG-bench, across model sizes spanning millions to hundreds of billions of parameters. In addition, a team of human expert raters performed all tasks in order to provide a strong baseline. Findings include: model performance and calibration both improve with scale, but are poor in absolute terms (and when compared with rater performance); performance is remarkably similar across model classes, though with benefits from sparsity; tasks that improve gradually and predictably commonly involve a large knowledge or memorization component, whereas tasks that exhibit "breakthrough" behavior at a critical scale often involve multiple steps or components, or brittle metrics; social bias typically increases with scale in settings with ambiguous context, but this can be improved with prompting.

Continual learning and few-shot learning are important frontiers in progress towards broader Machine Learning (ML) capabilities. There is a growing body of work in both, but few works combining the two. One exception is the Continual few-shot Learning (CFSL) framework of Antoniou et al. arXiv:2004.11967. In this study, we extend CFSL in two ways that capture a broader range of challenges, important for intelligent agent behaviour in real-world conditions. First, we modify CFSL to make it more comparable to standard continual learning experiments, where usually a much larger number of classes are presented. Second, we introduce an 'instance test' which requires recognition of specific instances of classes -- a capability of animal cognition that is usually neglected in ML. For an initial exploration of ML model performance under these conditions, we selected representative baseline models from the original CFSL work and added a model variant with replay. As expected, learning more classes is more difficult than the original CFSL experiments, and interestingly, the way in which image instances and classes are presented affects classification performance. Surprisingly, accuracy in the baseline instance test is comparable to other classification tasks, but poor given significant occlusion and noise. The use of replay for consolidation improves performance substantially for both types of tasks, but particularly the instance test.

Human evaluation is the gold-standard for evaluating text generation models. It is also expensive, and to fit budgetary constraints, a random subset of the test data is often chosen in practice. The randomly selected data may not accurately represent test performance, making this approach economically inefficient for model comparison. Thus, in this work, we develop a suite of selectors to get the most informative datapoints for human evaluation while taking the evaluation costs into account. We show that selectors based on variance in automated metric scores, diversity in model outputs, or Item Response Theory outperform random selection. We further develop an approach to distill these selectors to the scenario where the model outputs are not yet available. In particular, we introduce source-based estimators, which predict item usefulness for human evaluation just based on the source texts. We demonstrate the efficacy of our selectors in two common NLG tasks, machine translation and summarization, and show that up to only ~50% of the test data is needed to produce the same evaluation result as the entire data. Our implementations are published in the subset2evaluate package.

Low-rank approximation techniques have become the de facto standard for fine-tuning Large Language Models (LLMs) due to their reduced computational and memory requirements. This paper investigates the effectiveness of these methods in capturing the shift of fine-tuning datasets from the initial pre-trained data distribution. Our findings reveal that there are cases in which low-rank fine-tuning falls short in learning such shifts. This, in turn, produces non-negligible side effects, especially when fine-tuning is adopted for toxicity mitigation in pre-trained models, or in scenarios where it is important to provide fair models. Through comprehensive empirical evidence on several models, datasets, and tasks, we show that low-rank fine-tuning inadvertently preserves undesirable biases and toxic behaviors. We also show that this extends to sequential decision-making tasks, emphasizing the need for careful evaluation to promote responsible LLMs development.

Personalized treatment effect estimates are often of interest in high-stakes applications -- thus, before deploying a model estimating such effects in practice, one needs to be sure that the best candidate from the ever-growing machine learning toolbox for this task was chosen. Unfortunately, due to the absence of counterfactual information in practice, it is usually not possible to rely on standard validation metrics for doing so, leading to a well-known model selection dilemma in the treatment effect estimation literature. While some solutions have recently been investigated, systematic understanding of the strengths and weaknesses of different model selection criteria is still lacking. In this paper, instead of attempting to declare a global `winner', we therefore empirically investigate success- and failure modes of different selection criteria. We highlight that there is a complex interplay between selection strategies, candidate estimators and the data used for comparing them, and provide interesting insights into the relative (dis)advantages of different criteria alongside desiderata for the design of further illuminating empirical studies in this context.

Retrieval models are often evaluated on partially-annotated datasets. Each query is mapped to a few relevant texts and the remaining corpus is assumed to be irrelevant. As a result, models that successfully retrieve false negatives are punished in evaluation. Unfortunately, completely annotating all texts for every query is not resource efficient. In this work, we show that using partially-annotated datasets in evaluation can paint a distorted picture. We curate D-MERIT, a passage retrieval evaluation set from Wikipedia, aspiring to contain all relevant passages for each query. Queries describe a group (e.g., ``journals about linguistics'') and relevant passages are evidence that entities belong to the group (e.g., a passage indicating that Language is a journal about linguistics). We show that evaluating on a dataset containing annotations for only a subset of the relevant passages might result in misleading ranking of the retrieval systems and that as more relevant texts are included in the evaluation set, the rankings converge. We propose our dataset as a resource for evaluation and our study as a recommendation for balance between resource-efficiency and reliable evaluation when annotating evaluation sets for text retrieval.

Large Language Models (LLMs) have recently been shown to be effective as automatic evaluators with simple prompting and in-context learning. In this work, we assemble 15 LLMs of four different size ranges and evaluate their output responses by preference ranking from the other LLMs as evaluators, such as System Star is better than System Square. We then evaluate the quality of ranking outputs introducing the Cognitive Bias Benchmark for LLMs as Evaluators (CoBBLEr), a benchmark to measure six different cognitive biases in LLM evaluation outputs, such as the Egocentric bias where a model prefers to rank its own outputs highly in evaluation. We find that LLMs are biased text quality evaluators, exhibiting strong indications on our bias benchmark (average of 40% of comparisons across all models) within each of their evaluations that question their robustness as evaluators. Furthermore, we examine the correlation between human and machine preferences and calculate the average Rank-Biased Overlap (RBO) score to be 49.6%, indicating that machine preferences are misaligned with humans. According to our findings, LLMs may still be unable to be utilized for automatic annotation aligned with human preferences. Our project page is at: https://minnesotanlp.github.io/cobbler.

Rigorous and reproducible evaluation is critical for assessing the state of the art and for guiding scientific advances in Artificial Intelligence. Evaluation is challenging in practice due to several reasons, including benchmark saturation, lack of transparency in methods used for measurement, development challenges in extracting measurements for generative tasks, and, more generally, the extensive number of capabilities required for a well-rounded comparison across models. We make three contributions to alleviate the above challenges. First, we present Eureka, an open-source framework for standardizing evaluations of large foundation models beyond single-score reporting and rankings. Second, we introduce Eureka-Bench as an extensible collection of benchmarks testing capabilities that (i) are still challenging for state-of-the-art models and (ii) represent fundamental but overlooked language and multimodal capabilities. The inherent space for improvement in non-saturated benchmarks enables us to discover meaningful differences between models at a capability level. Third, using Eureka, we conduct an analysis of 12 state-of-the-art models, providing in-depth insights into failure understanding and model comparison, which can be leveraged to plan targeted improvements. In contrast to recent trends in reports and leaderboards showing absolute rankings and claims for one model or another to be the best, our analysis shows that there is no such best model. Different models have different strengths, but there are models that appear more often than others as best performers for some capabilities. Despite the recent improvements, current models still struggle with several fundamental capabilities including detailed image understanding, benefiting from multimodal input when available rather than fully relying on language, factuality and grounding for information retrieval, and over refusals.

Language Models [LMs] are now playing an increasingly large role in information generation and synthesis; the representation of scientific knowledge in these systems needs to be highly accurate. A prime challenge is hallucination; that is, generating apparently plausible but actually false information, including invented citations and nonexistent research papers. This kind of inaccuracy is dangerous in all the domains that require high levels of factual correctness, such as academia and education. This work presents a pipeline for evaluating the frequency with which language models hallucinate in generating responses in the scientific literature. We propose ArxEval, an evaluation pipeline with two tasks using ArXiv as a repository: Jumbled Titles and Mixed Titles. Our evaluation includes fifteen widely used language models and provides comparative insights into their reliability in handling scientific literature.

We present methodology for using dynamic evaluation to improve neural sequence models. Models are adapted to recent history via a gradient descent based mechanism, causing them to assign higher probabilities to re-occurring sequential patterns. Dynamic evaluation outperforms existing adaptation approaches in our comparisons. Dynamic evaluation improves the state-of-the-art word-level perplexities on the Penn Treebank and WikiText-2 datasets to 51.1 and 44.3 respectively, and the state-of-the-art character-level cross-entropies on the text8 and Hutter Prize datasets to 1.19 bits/char and 1.08 bits/char respectively.

Large language models (LLMs) are gaining increasing popularity in both academia and industry, owing to their unprecedented performance in various applications. As LLMs continue to play a vital role in both research and daily use, their evaluation becomes increasingly critical, not only at the task level, but also at the society level for better understanding of their potential risks. Over the past years, significant efforts have been made to examine LLMs from various perspectives. This paper presents a comprehensive review of these evaluation methods for LLMs, focusing on three key dimensions: what to evaluate, where to evaluate, and how to evaluate. Firstly, we provide an overview from the perspective of evaluation tasks, encompassing general natural language processing tasks, reasoning, medical usage, ethics, educations, natural and social sciences, agent applications, and other areas. Secondly, we answer the `where' and `how' questions by diving into the evaluation methods and benchmarks, which serve as crucial components in assessing performance of LLMs. Then, we summarize the success and failure cases of LLMs in different tasks. Finally, we shed light on several future challenges that lie ahead in LLMs evaluation. Our aim is to offer invaluable insights to researchers in the realm of LLMs evaluation, thereby aiding the development of more proficient LLMs. Our key point is that evaluation should be treated as an essential discipline to better assist the development of LLMs. We consistently maintain the related open-source materials at: https://github.com/MLGroupJLU/LLM-eval-survey.

Pairwise re-ranking models predict which of two documents is more relevant to a query and then aggregate a final ranking from such preferences. This is often more effective than pointwise re-ranking models that directly predict a relevance value for each document. However, the high inference overhead of pairwise models limits their practical application: usually, for a set of k documents to be re-ranked, preferences for all k^2-k comparison pairs excluding self-comparisons are aggregated. We investigate whether the efficiency of pairwise re-ranking can be improved by sampling from all pairs. In an exploratory study, we evaluate three sampling methods and five preference aggregation methods. The best combination allows for an order of magnitude fewer comparisons at an acceptable loss of retrieval effectiveness, while competitive effectiveness is already achieved with about one third of the comparisons.

Over the past decade, tremendous progress has been made in Recommender Systems (RecSys) for well-known tasks such as next-item and next-basket prediction. On the other hand, the recently proposed next-period recommendation (NPR) task is not covered as much. Current works about NPR are mostly based around distinct problem formulations, methods, and proprietary datasets, making solutions difficult to reproduce. In this article, we aim to fill the gap in RecSys methods evaluation on the NPR task using publicly available datasets and (1) introduce the TTRS, a large-scale financial transactions dataset suitable for RecSys methods evaluation; (2) benchmark popular RecSys approaches on several datasets for the NPR task. When performing our analysis, we found a strong repetitive consumption pattern in several real-world datasets. With this setup, our results suggest that the repetitive nature of data is still hard to generalize for the evaluated RecSys methods, and novel item prediction performance is still questionable.

Extensive evaluation on a large number of word embedding models for language processing applications is conducted in this work. First, we introduce popular word embedding models and discuss desired properties of word models and evaluation methods (or evaluators). Then, we categorize evaluators into intrinsic and extrinsic two types. Intrinsic evaluators test the quality of a representation independent of specific natural language processing tasks while extrinsic evaluators use word embeddings as input features to a downstream task and measure changes in performance metrics specific to that task. We report experimental results of intrinsic and extrinsic evaluators on six word embedding models. It is shown that different evaluators focus on different aspects of word models, and some are more correlated with natural language processing tasks. Finally, we adopt correlation analysis to study performance consistency of extrinsic and intrinsic evalutors.

The versatility of large language models (LLMs) led to the creation of diverse benchmarks that thoroughly test a variety of language models' abilities. These benchmarks consist of tens of thousands of examples making evaluation of LLMs very expensive. In this paper, we investigate strategies to reduce the number of evaluations needed to assess the performance of an LLM on several key benchmarks. For example, we show that to accurately estimate the performance of an LLM on MMLU, a popular multiple-choice QA benchmark consisting of 14K examples, it is sufficient to evaluate this LLM on 100 curated examples. We release evaluation tools and tiny versions of popular benchmarks: Open LLM Leaderboard, MMLU, HELM, and AlpacaEval 2.0. Our empirical analysis demonstrates that these tools and tiny benchmarks are sufficient to reliably and efficiently reproduce the original evaluation results.

Language models (LMs) are becoming the foundation for almost all major language technologies, but their capabilities, limitations, and risks are not well understood. We present Holistic Evaluation of Language Models (HELM) to improve the transparency of language models. First, we taxonomize the vast space of potential scenarios (i.e. use cases) and metrics (i.e. desiderata) that are of interest for LMs. Then we select a broad subset based on coverage and feasibility, noting what's missing or underrepresented (e.g. question answering for neglected English dialects, metrics for trustworthiness). Second, we adopt a multi-metric approach: We measure 7 metrics (accuracy, calibration, robustness, fairness, bias, toxicity, and efficiency) for each of 16 core scenarios when possible (87.5% of the time). This ensures metrics beyond accuracy don't fall to the wayside, and that trade-offs are clearly exposed. We also perform 7 targeted evaluations, based on 26 targeted scenarios, to analyze specific aspects (e.g. reasoning, disinformation). Third, we conduct a large-scale evaluation of 30 prominent language models (spanning open, limited-access, and closed models) on all 42 scenarios, 21 of which were not previously used in mainstream LM evaluation. Prior to HELM, models on average were evaluated on just 17.9% of the core HELM scenarios, with some prominent models not sharing a single scenario in common. We improve this to 96.0%: now all 30 models have been densely benchmarked on the same core scenarios and metrics under standardized conditions. Our evaluation surfaces 25 top-level findings. For full transparency, we release all raw model prompts and completions publicly for further analysis, as well as a general modular toolkit. We intend for HELM to be a living benchmark for the community, continuously updated with new scenarios, metrics, and models.

Large language models~(LLMs) have greatly advanced the frontiers of artificial intelligence, attaining remarkable improvement in model capacity. To assess the model performance, a typical approach is to construct evaluation benchmarks for measuring the ability level of LLMs in different aspects. Despite that a number of high-quality benchmarks have been released, the concerns about the appropriate use of these benchmarks and the fair comparison of different models are increasingly growing. Considering these concerns, in this paper, we discuss the potential risk and impact of inappropriately using evaluation benchmarks and misleadingly interpreting the evaluation results. Specially, we focus on a special issue that would lead to inappropriate evaluation, \ie benchmark leakage, referring that the data related to evaluation sets is occasionally used for model training. This phenomenon now becomes more common since pre-training data is often prepared ahead of model test. We conduct extensive experiments to study the effect of benchmark leverage, and find that it can dramatically boost the evaluation results, which would finally lead to an unreliable assessment of model performance. To improve the use of existing evaluation benchmarks, we finally present several guidelines for both LLM developers and benchmark maintainers. We hope this work can draw attention to appropriate training and evaluation of LLMs.

Given the broad capabilities of large language models, it should be possible to work towards a general-purpose, text-based assistant that is aligned with human values, meaning that it is helpful, honest, and harmless. As an initial foray in this direction we study simple baseline techniques and evaluations, such as prompting. We find that the benefits from modest interventions increase with model size, generalize to a variety of alignment evaluations, and do not compromise the performance of large models. Next we investigate scaling trends for several training objectives relevant to alignment, comparing imitation learning, binary discrimination, and ranked preference modeling. We find that ranked preference modeling performs much better than imitation learning, and often scales more favorably with model size. In contrast, binary discrimination typically performs and scales very similarly to imitation learning. Finally we study a `preference model pre-training' stage of training, with the goal of improving sample efficiency when finetuning on human preferences.

Standard practice within Reinforcement Learning from Human Feedback (RLHF) involves optimizing against a Reward Model (RM), which itself is trained to reflect human preferences for desirable generations. A notable subject that is understudied is the (in-)consistency of RMs -- whether they can recognize the semantic changes to different prompts and appropriately adapt their reward assignments -- and their impact on the downstream RLHF model. In this paper, we visit a series of research questions relevant to RM inconsistency: (1) How can we measure the consistency of reward models? (2) How consistent are the existing RMs and how can we improve them? (3) In what ways does reward inconsistency influence the chatbots resulting from the RLHF model training? We propose Contrast Instructions -- a benchmarking strategy for the consistency of RM. Each example in Contrast Instructions features a pair of lexically similar instructions with different ground truth responses. A consistent RM is expected to rank the corresponding instruction and response higher than other combinations. We observe that current RMs trained with the standard ranking objective fail miserably on Contrast Instructions compared to average humans. To show that RM consistency can be improved efficiently without using extra training budget, we propose two techniques ConvexDA and RewardFusion, which enhance reward consistency through extrapolation during the RM training and inference stage, respectively. We show that RLHF models trained with a more consistent RM yield more useful responses, suggesting that reward inconsistency exhibits a trickle-down effect on the downstream RLHF process.

In Natural Language Processing (NLP), the Elo rating system, originally designed for ranking players in dynamic games such as chess, is increasingly being used to evaluate Large Language Models (LLMs) through "A vs B" paired comparisons. However, while popular, the system's suitability for assessing entities with constant skill levels, such as LLMs, remains relatively unexplored. We study two fundamental axioms that evaluation methods should adhere to: reliability and transitivity. We conduct extensive evaluation of Elo behaviour, illustrating that individual Elo computations exhibit volatility and delving into the impact of varying the Elo rating system's hyperparameters. We show that these axioms are not always satisfied raising questions about the reliability of current comparative evaluations of LLMs. If the current use of Elo scores is intended to substitute the costly head-to-head comparison of LLMs, it is crucial to ensure the ranking is as robust as possible. Guided by the axioms, our findings offer concrete guidelines for enhancing the reliability of LLM evaluation methods, suggesting a need for reassessment of existing comparative approaches.

Despite recent advances in text-to-3D generative methods, there is a notable absence of reliable evaluation metrics. Existing metrics usually focus on a single criterion each, such as how well the asset aligned with the input text. These metrics lack the flexibility to generalize to different evaluation criteria and might not align well with human preferences. Conducting user preference studies is an alternative that offers both adaptability and human-aligned results. User studies, however, can be very expensive to scale. This paper presents an automatic, versatile, and human-aligned evaluation metric for text-to-3D generative models. To this end, we first develop a prompt generator using GPT-4V to generate evaluating prompts, which serve as input to compare text-to-3D models. We further design a method instructing GPT-4V to compare two 3D assets according to user-defined criteria. Finally, we use these pairwise comparison results to assign these models Elo ratings. Experimental results suggest our metric strongly align with human preference across different evaluation criteria.

Accurate and consistent evaluation is crucial for decision-making across numerous fields, yet it remains a challenging task due to inherent subjectivity, variability, and scale. Large Language Models (LLMs) have achieved remarkable success across diverse domains, leading to the emergence of "LLM-as-a-Judge," where LLMs are employed as evaluators for complex tasks. With their ability to process diverse data types and provide scalable, cost-effective, and consistent assessments, LLMs present a compelling alternative to traditional expert-driven evaluations. However, ensuring the reliability of LLM-as-a-Judge systems remains a significant challenge that requires careful design and standardization. This paper provides a comprehensive survey of LLM-as-a-Judge, addressing the core question: How can reliable LLM-as-a-Judge systems be built? We explore strategies to enhance reliability, including improving consistency, mitigating biases, and adapting to diverse assessment scenarios. Additionally, we propose methodologies for evaluating the reliability of LLM-as-a-Judge systems, supported by a novel benchmark designed for this purpose. To advance the development and real-world deployment of LLM-as-a-Judge systems, we also discussed practical applications, challenges, and future directions. This survey serves as a foundational reference for researchers and practitioners in this rapidly evolving field.

Multimodal Large Language Models (MLLMs) have gained significant attention recently, showing remarkable potential in artificial general intelligence. However, assessing the utility of MLLMs presents considerable challenges, primarily due to the absence of multimodal benchmarks that align with human preferences. Drawing inspiration from the concept of LLM-as-a-Judge within LLMs, this paper introduces a novel benchmark, termed MLLM-as-a-Judge, to assess the ability of MLLMs in assisting judges across diverse modalities, encompassing three distinct tasks: Scoring Evaluation, Pair Comparison, and Batch Ranking. Our study reveals that, while MLLMs demonstrate remarkable human-like discernment in Pair Comparison, there is a significant divergence from human preferences in Scoring Evaluation and Batch Ranking. Furthermore, a closer examination reveals persistent challenges in the judgment capacities of LLMs, including diverse biases, hallucinatory responses, and inconsistencies in judgment, even in advanced models such as GPT-4V. These findings emphasize the pressing need for enhancements and further research efforts to be undertaken before regarding MLLMs as fully reliable evaluators. In light of this, we advocate for additional efforts dedicated to supporting the continuous development within the domain of MLLM functioning as judges. The code and dataset are publicly available at our project homepage: https://mllm-judge.github.io/.

The emergence of large language models (LLMs) has significantly impacted various fields, from natural language processing to sectors like medicine and finance. However, despite their rapid proliferation, the applications of LLMs in telecommunications remain limited, often relying on general-purpose models that lack domain-specific specialization. This lack of specialization results in underperformance, particularly when dealing with telecommunications-specific technical terminology and their associated mathematical representations. This paper addresses this gap by first creating and disseminating Tele-Data, a comprehensive dataset of telecommunications material curated from relevant sources, and Tele-Eval, a large-scale question-and-answer dataset tailored to the domain. Through extensive experiments, we explore the most effective training techniques for adapting LLMs to the telecommunications domain, ranging from examining the division of expertise across various telecommunications aspects to employing parameter-efficient techniques. We also investigate how models of different sizes behave during adaptation and analyze the impact of their training data on this behavior. Leveraging these findings, we develop and open-source Tele-LLMs, the first series of language models ranging from 1B to 8B parameters, specifically tailored for telecommunications. Our evaluations demonstrate that these models outperform their general-purpose counterparts on Tele-Eval while retaining their previously acquired capabilities, thus avoiding the catastrophic forgetting phenomenon.

Nowadays, the quality of responses generated by different modern large language models (LLMs) are hard to evaluate and compare automatically. Recent studies suggest and predominantly use LLMs as a reference-free metric for open-ended question answering. More specifically, they use the recognized "strongest" LLM as the evaluator, which conducts pairwise comparisons of candidate models' answers and provides a ranking score. However, this intuitive method has multiple problems, such as bringing in self-enhancement (favoring its own answers) and positional bias. We draw insights and lessons from the educational domain (Cho and MacArthur, 2011; Walsh, 2014) to improve LLM-based evaluations. Specifically, we propose the (1) peer rank (PR) algorithm that takes into account each peer LLM's pairwise preferences of all answer pairs, and outputs a final ranking of models; and (2) peer discussion (PD), where we prompt two LLMs to discuss and try to reach a mutual agreement on preferences of two answers. We conduct experiments on two benchmark datasets. We find that our approaches achieve higher accuracy and align better with human judgments, respectively. Interestingly, PR can induce a relatively accurate self-ranking of models under the anonymous setting, where each model's name is unrevealed. Our work provides space to explore evaluating models that are hard to compare for humans.

Current research on generative language models (GLMs) for automated text scoring (ATS) has focused almost exclusively on querying proprietary models via Application Programming Interfaces (APIs). Yet such practices raise issues around transparency and security, and these methods offer little in the way of efficiency or customizability. With the recent proliferation of smaller, open-source models, there is the option to explore GLMs with computers equipped with modest, consumer-grade hardware, that is, for the "GPU poor." In this study, we analyze the performance and efficiency of open-source, small-scale GLMs for ATS. Results show that GLMs can be fine-tuned to achieve adequate, though not state-of-the-art, performance. In addition to ATS, we take small steps towards analyzing models' capacity for generating feedback by prompting GLMs to explain their scores. Model-generated feedback shows promise, but requires more rigorous evaluation focused on targeted use cases.

Human evaluation is critical for validating the performance of text-to-image generative models, as this highly cognitive process requires deep comprehension of text and images. However, our survey of 37 recent papers reveals that many works rely solely on automatic measures (e.g., FID) or perform poorly described human evaluations that are not reliable or repeatable. This paper proposes a standardized and well-defined human evaluation protocol to facilitate verifiable and reproducible human evaluation in future works. In our pilot data collection, we experimentally show that the current automatic measures are incompatible with human perception in evaluating the performance of the text-to-image generation results. Furthermore, we provide insights for designing human evaluation experiments reliably and conclusively. Finally, we make several resources publicly available to the community to facilitate easy and fast implementations.

Medical language models (MLMs) have become pivotal in advancing medical natural language processing. However, prior models that rely on pre-training or supervised fine-tuning often exhibit low data efficiency and limited practicality in real-world clinical applications. While OpenAIs O1 highlights test-time scaling in mathematics, attempts to replicate this approach in medicine typically distill responses from GPT-series models to open-source models, focusing primarily on multiple-choice tasks. This strategy, though straightforward, neglects critical concerns like data privacy and realistic deployment in clinical settings. In this work, we present a deployable, small-scale medical language model, \mone, designed for long-chain reasoning in clinical tasks using a self-evolution paradigm. Starting with a seed dataset of around 8,000 instances spanning five domains and 16 datasets, we prompt a base policy model to perform Monte Carlo Tree Search (MCTS) to construct verifiable reasoning chains. Each reasoning step is assigned an evolution rollout value, allowing verified trajectories to train the policy model and the reward model. During inference, the policy model generates multiple responses, and the reward model selects the one with the highest reward score. Experiments on eleven evaluation datasets demonstrate that \mone outperforms prior open-source models by 2 points, with the addition of the reward model further boosting performance (sim13 points), surpassing GPT-4o-mini. Code and data are available at https://github.com/pixas/MedSSS.

Recent advances demonstrate that increasing inference-time computation can significantly boost the reasoning capabilities of large language models (LLMs). Although repeated sampling (i.e., generating multiple candidate outputs) is a highly effective strategy, it does not leverage external feedback signals for refinement, which are often available in tasks like coding. In this work, we propose Adaptive Branching Monte Carlo Tree Search (AB-MCTS), a novel inference-time framework that generalizes repeated sampling with principled multi-turn exploration and exploitation. At each node in the search tree, AB-MCTS dynamically decides whether to "go wider" by expanding new candidate responses or "go deeper" by revisiting existing ones based on external feedback signals. We evaluate our method on complex coding and engineering tasks using frontier models. Empirical results show that AB-MCTS consistently outperforms both repeated sampling and standard MCTS, underscoring the importance of combining the response diversity of LLMs with multi-turn solution refinement for effective inference-time scaling.

Despite notable advancements in Multimodal Large Language Models (MLLMs), most state-of-the-art models have not undergone thorough alignment with human preferences. This gap exists because current alignment research has primarily achieved progress in specific areas (e.g., hallucination reduction), while the broader question of whether aligning models with human preferences can systematically enhance MLLM capability remains largely unexplored. To this end, we introduce MM-RLHF, a dataset containing 120k fine-grained, human-annotated preference comparison pairs. This dataset represents a substantial advancement over existing resources, offering superior size, diversity, annotation granularity, and quality. Leveraging this dataset, we propose several key innovations to improve both the quality of reward models and the efficiency of alignment algorithms. Notably, we introduce a Critique-Based Reward Model, which generates critiques of model outputs before assigning scores, offering enhanced interpretability and more informative feedback compared to traditional scalar reward mechanisms. Additionally, we propose Dynamic Reward Scaling, a method that adjusts the loss weight of each sample according to the reward signal, thereby optimizing the use of high-quality comparison pairs. Our approach is rigorously evaluated across 10 distinct dimensions and 27 benchmarks, with results demonstrating significant and consistent improvements in model performance. Specifically, fine-tuning LLaVA-ov-7B with MM-RLHF and our alignment algorithm leads to a 19.5% increase in conversational abilities and a 60% improvement in safety. We have open-sourced the preference dataset, reward model, training and evaluation code, as well as reward modeling and safety benchmarks. For more details, please visit our project page: https://mm-rlhf.github.io.

Language models can generate harmful and biased outputs and exhibit undesirable behavior according to a given cultural context. We propose a Process for Adapting Language Models to Society (PALMS) with Values-Targeted Datasets, an iterative process to significantly change model behavior by crafting and fine-tuning on a dataset that reflects a predetermined set of target values. We evaluate our process using three metrics: quantitative metrics with human evaluations that score output adherence to a target value, toxicity scoring on outputs; and qualitative metrics analyzing the most common word associated with a given social category. Through each iteration, we add additional training dataset examples based on observed shortcomings from evaluations. PALMS performs significantly better on all metrics compared to baseline and control models for a broad range of GPT-3 language model sizes without compromising capability integrity. We find that the effectiveness of PALMS increases with model size. We show that significantly adjusting language model behavior is feasible with a small, hand-curated dataset.

Retrieval-augmented generation (RAG) is considered to be a promising approach to alleviate the hallucination issue of large language models (LLMs), and it has received widespread attention from researchers recently. Due to the limitation in the semantic understanding of retrieval models, the success of RAG heavily lies on the ability of LLMs to identify passages with utility. Recent efforts have explored the ability of LLMs to assess the relevance of passages in retrieval, but there has been limited work on evaluating the utility of passages in supporting question answering. In this work, we conduct a comprehensive study about the capabilities of LLMs in utility evaluation for open-domain QA. Specifically, we introduce a benchmarking procedure and collection of candidate passages with different characteristics, facilitating a series of experiments with five representative LLMs. Our experiments reveal that: (i) well-instructed LLMs can distinguish between relevance and utility, and that LLMs are highly receptive to newly generated counterfactual passages. Moreover, (ii) we scrutinize key factors that affect utility judgments in the instruction design. And finally, (iii) to verify the efficacy of utility judgments in practical retrieval augmentation applications, we delve into LLMs' QA capabilities using the evidence judged with utility and direct dense retrieval results. (iv) We propose a k-sampling, listwise approach to reduce the dependency of LLMs on the sequence of input passages, thereby facilitating subsequent answer generation. We believe that the way we formalize and study the problem along with our findings contributes to a critical assessment of retrieval-augmented LLMs. Our code and benchmark can be found at https://github.com/ict-bigdatalab/utility_judgments.

Instruction tuning has been widely adopted to ensure large language models (LLMs) follow user instructions effectively. The resulting instruction-following capabilities of LLMs heavily rely on the instruction datasets used for tuning. Recently, synthetic instruction datasets have emerged as an economically viable solution to provide LLMs diverse and high-quality instructions. However, existing approaches typically assume that larger or stronger models are stronger teachers for instruction tuning, and hence simply adopt these models as response generators to the synthetic instructions. In this paper, we challenge this commonly-adopted assumption. Our extensive experiments across five base models and twenty response generators reveal that larger and stronger models are not necessarily stronger teachers of smaller models. We refer to this phenomenon as the Larger Models' Paradox. We observe that existing metrics cannot precisely predict the effectiveness of response generators since they ignore the compatibility between teachers and base models being fine-tuned. We thus develop a novel metric, named as Compatibility-Adjusted Reward (CAR) to measure the effectiveness of response generators. Our experiments across five base models demonstrate that CAR outperforms almost all baselines.

Human evaluation is increasingly critical for assessing large language models, capturing linguistic nuances, and reflecting user preferences more accurately than traditional automated metrics. However, the resource-intensive nature of this type of annotation process poses significant challenges. The key question driving our work: "is it feasible to minimize human-in-the-loop feedback by prioritizing data instances which most effectively distinguish between models?" We evaluate several metric-based methods and find that these metrics enhance the efficiency of human evaluations by minimizing the number of required annotations, thus saving time and cost, while ensuring a robust performance evaluation. We show that our method is effective across widely used model families, reducing instances of indecisive (or "tie") outcomes by up to 54% compared to a random sample when focusing on the top-20 percentile of prioritized instances. This potential reduction in required human effort positions our approach as a valuable strategy in future large language model evaluations.

The performance of machine learning models often degrades after deployment due to data distribution shifts. In many use cases, it is impossible to calculate the post-deployment performance because labels are unavailable or significantly delayed. Proxy methods for evaluating model performance stability, like drift detection techniques, do not properly quantify data distribution shift impact. As a solution, we propose a robust and accurate performance estimation method for evaluating ML classification models on unlabeled data that accurately quantifies the impact of covariate shift on model performance. We call it multi-calibrated confidence-based performance estimation (M-CBPE). It is model and data-type agnostic and works for any performance metric. It does not require access to the monitored model - it uses the model predictions and probability estimates. M-CBPE does not need user input on the nature of the covariate shift as it fully learns from the data. We evaluate it with over 600 dataset-model pairs from US census data and compare it with multiple benchmarks using several evaluation metrics. Results show that M-CBPE is the best method to estimate the performance of classification models in any evaluation context.

This study investigates the consistency of feedback ratings generated by OpenAI's GPT-4, a state-of-the-art artificial intelligence language model, across multiple iterations, time spans and stylistic variations. The model rated responses to tasks within the Higher Education (HE) subject domain of macroeconomics in terms of their content and style. Statistical analysis was conducted in order to learn more about the interrater reliability, consistency of the ratings across iterations and the correlation between ratings in terms of content and style. The results revealed a high interrater reliability with ICC scores ranging between 0.94 and 0.99 for different timespans, suggesting that GPT-4 is capable of generating consistent ratings across repetitions with a clear prompt. Style and content ratings show a high correlation of 0.87. When applying a non-adequate style the average content ratings remained constant, while style ratings decreased, which indicates that the large language model (LLM) effectively distinguishes between these two criteria during evaluation. The prompt used in this study is furthermore presented and explained. Further research is necessary to assess the robustness and reliability of AI models in various use cases.

As the application of Large Language Models (LLMs) expands, the demand for reliable evaluations increases. Existing LLM evaluation benchmarks primarily rely on static datasets, making it challenging to assess model performance in dynamic interactions with users. Moreover, these benchmarks often depend on specific background knowledge, complicating the measurement of a model's logical reasoning capabilities. Other dynamic evaluation methods based on strong models or manual efforts may introduce biases and incur high costs and time demands, hindering large-scale application. To address these issues, we propose TurtleBench. TurtleBench collects real user guesses from our online Turtle Soup Puzzle platform that we developed. This approach allows for the relatively dynamic generation of evaluation datasets, mitigating the risk of model cheating while aligning assessments more closely with genuine user needs for reasoning capabilities, thus enhancing the reliability of evaluations. TurtleBench includes 1,532 user guesses along with the correctness of guesses after annotation. Using this dataset, we thoroughly evaluated nine of the most advanced LLMs available today. Notably, the OpenAI o1 series models did not achieve leading results in these evaluations. We propose several hypotheses for further research, such as "the latent reasoning of o1 utilizes trivial Chain-of-Thought (CoT) techniques" and "increasing CoT length not only provides reasoning benefits but also incurs noise costs."

To effectively optimize and personalize treatments, it is necessary to investigate the heterogeneity of treatment effects. With the wide range of users being treated over many online controlled experiments, the typical approach of manually investigating each dimension of heterogeneity becomes overly cumbersome and prone to subjective human biases. We need an efficient way to search through thousands of experiments with hundreds of target covariates and hundreds of breakdown dimensions. In this paper, we propose a systematic paradigm for detecting, surfacing and characterizing heterogeneous treatment effects. First, we detect if treatment effect variation is present in an experiment, prior to specifying any breakdowns. Second, we surface the most relevant dimensions for heterogeneity. Finally, we characterize the heterogeneity beyond just the conditional average treatment effects (CATE) by studying the conditional distributions of the estimated individual treatment effects. We show the effectiveness of our methods using simulated data and empirical studies.

We introduce WildBench, an automated evaluation framework designed to benchmark large language models (LLMs) using challenging, real-world user queries. WildBench consists of 1,024 tasks carefully selected from over one million human-chatbot conversation logs. For automated evaluation with WildBench, we have developed two metrics, WB-Reward and WB-Score, which are computable using advanced LLMs such as GPT-4-turbo. WildBench evaluation uses task-specific checklists to evaluate model outputs systematically and provides structured explanations that justify the scores and comparisons, resulting in more reliable and interpretable automatic judgments. WB-Reward employs fine-grained pairwise comparisons between model responses, generating five potential outcomes: much better, slightly better, slightly worse, much worse, or a tie. Unlike previous evaluations that employed a single baseline model, we selected three baseline models at varying performance levels to ensure a comprehensive pairwise evaluation. Additionally, we propose a simple method to mitigate length bias, by converting outcomes of ``slightly better/worse'' to ``tie'' if the winner response exceeds the loser one by more than K characters. WB-Score evaluates the quality of model outputs individually, making it a fast and cost-efficient evaluation metric. WildBench results demonstrate a strong correlation with the human-voted Elo ratings from Chatbot Arena on hard tasks. Specifically, WB-Reward achieves a Pearson correlation of 0.98 with top-ranking models. Additionally, WB-Score reaches 0.95, surpassing both ArenaHard's 0.91 and AlpacaEval2.0's 0.89 for length-controlled win rates, as well as the 0.87 for regular win rates.

With the rise of Large Language Models (LLMs) and their ubiquitous deployment in diverse domains, measuring language model behavior on realistic data is imperative. For example, a company deploying a client-facing chatbot must ensure that the model will not respond to client requests with profanity. Current evaluations approach this problem using small, domain-specific datasets with human-curated labels. These evaluation sets are often sampled from a narrow and simplified distribution, and data sources can unknowingly be leaked into the training set which can lead to misleading evaluations. To bypass these drawbacks, we propose a framework for self-supervised evaluation of LLMs by analyzing their sensitivity or invariance to transformations on the input text. Self-supervised evaluation can directly monitor LLM behavior on datasets collected in the wild or streamed during live model deployment. We demonstrate self-supervised evaluation strategies for measuring closed-book knowledge, toxicity, and long-range context dependence, in addition to sensitivity to grammatical structure and tokenization errors. When comparisons to similar human-labeled benchmarks are available, we find strong correlations between self-supervised and human-supervised evaluations. The self-supervised paradigm complements current evaluation strategies that rely on labeled data.

Instruction-tuned large language models have revolutionized natural language processing and have shown great potential in applications such as conversational agents. These models, such as GPT-4, can not only master language but also solve complex tasks in areas like mathematics, coding, medicine, and law. Despite their impressive capabilities, there is still a lack of comprehensive understanding regarding their full potential, primarily due to the black-box nature of many models and the absence of holistic evaluation studies. To address these challenges, we present INSTRUCTEVAL, a more comprehensive evaluation suite designed specifically for instruction-tuned large language models. Unlike previous works, our evaluation involves a rigorous assessment of models based on problem-solving, writing ability, and alignment to human values. We take a holistic approach to analyze various factors affecting model performance, including the pretraining foundation, instruction-tuning data, and training methods. Our findings reveal that the quality of instruction data is the most crucial factor in scaling model performance. While open-source models demonstrate impressive writing abilities, there is substantial room for improvement in problem-solving and alignment. We are encouraged by the rapid development of models by the open-source community, but we also highlight the need for rigorous evaluation to support claims made about these models. Through INSTRUCTEVAL, we aim to foster a deeper understanding of instruction-tuned models and advancements in their capabilities. INSTRUCTEVAL is publicly available at https://github.com/declare-lab/instruct-eval.

Sequential recommendation aims to estimate the dynamic user preferences and sequential dependencies among historical user behaviors. Although Transformer-based models have proven to be effective for sequential recommendation, they suffer from the inference inefficiency problem stemming from the quadratic computational complexity of attention operators, especially for long behavior sequences. Inspired by the recent success of state space models (SSMs), we propose Mamba4Rec, which is the first work to explore the potential of selective SSMs for efficient sequential recommendation. Built upon the basic Mamba block which is a selective SSM with an efficient hardware-aware parallel algorithm, we design a series of sequential modeling techniques to further promote model performance while maintaining inference efficiency. Through experiments on public datasets, we demonstrate how Mamba4Rec effectively tackles the effectiveness-efficiency dilemma, outperforming both RNN- and attention-based baselines in terms of both effectiveness and efficiency. The code is available at https://github.com/chengkai-liu/Mamba4Rec.

Recent studies show that large language models (LLMs) can be instructed to effectively perform zero-shot passage re-ranking, in which the results of a first stage retrieval method, such as BM25, are rated and reordered to improve relevance. In this work, we improve LLM-based re-ranking by algorithmically selecting few-shot demonstrations to include in the prompt. Our analysis investigates the conditions where demonstrations are most helpful, and shows that adding even one demonstration is significantly beneficial. We propose a novel demonstration selection strategy based on difficulty rather than the commonly used semantic similarity. Furthermore, we find that demonstrations helpful for ranking are also effective at question generation. We hope our work will spur more principled research into question generation and passage ranking.

This work investigates the ability of open Large Language Models (LLMs) to predict citation intent through in-context learning and fine-tuning. Unlike traditional approaches that rely on pre-trained models like SciBERT, which require extensive domain-specific pretraining and specialized architectures, we demonstrate that general-purpose LLMs can be adapted to this task with minimal task-specific data. We evaluate twelve model variations across five prominent open LLM families using zero, one, few, and many-shot prompting to assess performance across scenarios. Our experimental study identifies the top-performing model through extensive experimentation of in-context learning-related parameters, which we fine-tune to further enhance task performance. The results highlight the strengths and limitations of LLMs in recognizing citation intents, providing valuable insights for model selection and prompt engineering. Additionally, we make our end-to-end evaluation framework and models openly available for future use.

We introduce Atla Selene Mini, a state-of-the-art small language model-as-a-judge (SLMJ). Selene Mini is a general-purpose evaluator that outperforms the best SLMJs and GPT-4o-mini on overall performance across 11 out-of-distribution benchmarks, spanning absolute scoring, classification, and pairwise preference tasks. It is the highest-scoring 8B generative model on RewardBench, surpassing strong baselines like GPT-4o and specialized judges. To achieve this, we develop a principled data curation strategy that augments public datasets with synthetically generated critiques and ensures high quality through filtering and dataset ablations. We train our model on a combined direct preference optimization (DPO) and supervised fine-tuning (SFT) loss, and produce a highly promptable evaluator that excels in real-world scenarios. Selene Mini shows dramatically improved zero-shot agreement with human expert evaluations on financial and medical industry datasets. It is also robust to variations in prompt format. Preliminary results indicate that Selene Mini is the top-ranking evaluator in a live, community-driven Judge Arena. We release the model weights on HuggingFace (https://hf.co/AtlaAI/Selene-1-Mini-Llama-3.1-8B) and Ollama to encourage widespread community adoption.

One core capability of Large Language Models (LLMs) is to follow natural language instructions. However, the evaluation of such abilities is not standardized: Human evaluations are expensive, slow, and not objectively reproducible, while LLM-based auto-evaluation is potentially biased or limited by the ability of the evaluator LLM. To overcome these issues, we introduce Instruction-Following Eval (IFEval) for large language models. IFEval is a straightforward and easy-to-reproduce evaluation benchmark. It focuses on a set of "verifiable instructions" such as "write in more than 400 words" and "mention the keyword of AI at least 3 times". We identified 25 types of those verifiable instructions and constructed around 500 prompts, with each prompt containing one or more verifiable instructions. We show evaluation results of two widely available LLMs on the market. Our code and data can be found at https://github.com/google-research/google-research/tree/master/instruction_following_eval

In real-world scenarios with naturally occurring datasets, reference summaries are noisy and may contain information that cannot be inferred from the source text. On large news corpora, removing low quality samples has been shown to reduce model hallucinations. Yet, for smaller, and/or noisier corpora, filtering is detrimental to performance. To improve reference quality while retaining all data, we propose a new approach: to selectively re-write unsupported reference sentences to better reflect source data. We automatically generate a synthetic dataset of positive and negative revisions by corrupting supported sentences and learn to revise reference sentences with contrastive learning. The intensity of revisions is treated as a controllable attribute so that, at inference, diverse candidates can be over-generated-then-rescored to balance faithfulness and abstraction. To test our methods, we extract noisy references from publicly available MIMIC-III discharge summaries for the task of hospital-course summarization, and vary the data on which models are trained. According to metrics and human evaluation, models trained on revised clinical references are much more faithful, informative, and fluent than models trained on original or filtered data.

Previous works on Treatment Effect Estimation (TEE) are not in widespread use because they are predominantly theoretical, where strong parametric assumptions are made but untractable for practical application. Recent work uses multilayer perceptron (MLP) for modeling casual relationships, however, MLPs lag far behind recent advances in ML methodology, which limits their applicability and generalizability. To extend beyond the single domain formulation and towards more realistic learning scenarios, we explore model design spaces beyond MLPs, i.e., transformer backbones, which provide flexibility where attention layers govern interactions among treatments and covariates to exploit structural similarities of potential outcomes for confounding control. Through careful model design, Transformers as Treatment Effect Estimators (TransTEE) is proposed. We show empirically that TransTEE can: (1) serve as a general purpose treatment effect estimator that significantly outperforms competitive baselines in a variety of challenging TEE problems (e.g., discrete, continuous, structured, or dosage-associated treatments) and is applicable to both when covariates are tabular and when they consist of structural data (e.g., texts, graphs); (2) yield multiple advantages: compatibility with propensity score modeling, parameter efficiency, robustness to continuous treatment value distribution shifts, explainable in covariate adjustment, and real-world utility in auditing pre-trained language models

Perception of toxicity evolves over time and often differs between geographies and cultural backgrounds. Similarly, black-box commercially available APIs for detecting toxicity, such as the Perspective API, are not static, but frequently retrained to address any unattended weaknesses and biases. We evaluate the implications of these changes on the reproducibility of findings that compare the relative merits of models and methods that aim to curb toxicity. Our findings suggest that research that relied on inherited automatic toxicity scores to compare models and techniques may have resulted in inaccurate findings. Rescoring all models from HELM, a widely respected living benchmark, for toxicity with the recent version of the API led to a different ranking of widely used foundation models. We suggest caution in applying apples-to-apples comparisons between studies and lay recommendations for a more structured approach to evaluating toxicity over time. Code and data are available at https://github.com/for-ai/black-box-api-challenges.

We find that existing language modeling datasets contain many near-duplicate examples and long repetitive substrings. As a result, over 1% of the unprompted output of language models trained on these datasets is copied verbatim from the training data. We develop two tools that allow us to deduplicate training datasets -- for example removing from C4 a single 61 word English sentence that is repeated over 60,000 times. Deduplication allows us to train models that emit memorized text ten times less frequently and require fewer train steps to achieve the same or better accuracy. We can also reduce train-test overlap, which affects over 4% of the validation set of standard datasets, thus allowing for more accurate evaluation. We release code for reproducing our work and performing dataset deduplication at https://github.com/google-research/deduplicate-text-datasets.

We introduce MIA-Bench, a new benchmark designed to evaluate multimodal large language models (MLLMs) on their ability to strictly adhere to complex instructions. Our benchmark comprises a diverse set of 400 image-prompt pairs, each crafted to challenge the models' compliance with layered instructions in generating accurate responses that satisfy specific requested patterns. Evaluation results from a wide array of state-of-the-art MLLMs reveal significant variations in performance, highlighting areas for improvement in instruction fidelity. Additionally, we create extra training data and explore supervised fine-tuning to enhance the models' ability to strictly follow instructions without compromising performance on other tasks. We hope this benchmark not only serves as a tool for measuring MLLM adherence to instructions, but also guides future developments in MLLM training methods.

The rapid advancement of Large Language Models (LLMs) necessitates robust evaluation methodologies. Current benchmarking approaches often rely on comparing model outputs against predefined prompts and reference outputs. Relying on predefined reference outputs hinders flexible adaptation of benchmarks to the rapidly evolving capabilities of LLMs. This limitation necessitates periodic efforts to prepare new benchmarks. To keep pace with rapidly evolving LLM capabilities, we propose a more flexible benchmarking approach. Our method, \textbf{Varco Arena}, provides reference-free benchmarking of LLMs in tournament style. \textbf{Varco Arena} directly compares LLM outputs across a diverse set of prompts, determining model rankings through a single-elimination tournament structure. This direct pairwise comparison offers two key advantages: (1) Direct comparison, unmediated by reference text, more effectively orders competing LLMs, resulting in more reliable rankings, and (2) reference-free approach to benchmarking adds flexibility in updating benchmark prompts by eliminating the need for quality references. Our empirical results, supported by simulation experiments, demonstrate that the \textbf{Varco Arena} tournament approach aligns better with the current Elo model for benchmarking LLMs. The alignment is measured in terms of Spearman correlation, showing improvement over current practice of benchmarking that use reference outputs as comparison anchors.

Reward models (RMs) guide the alignment of large language models (LLMs), steering them toward behaviors preferred by humans. Evaluating RMs is the key to better aligning LLMs. However, the current evaluation of RMs may not directly correspond to their alignment performance due to the limited distribution of evaluation data and evaluation methods that are not closely related to alignment objectives. To address these limitations, we propose RMB, a comprehensive RM benchmark that covers over 49 real-world scenarios and includes both pairwise and Best-of-N (BoN) evaluations to better reflect the effectiveness of RMs in guiding alignment optimization. We demonstrate a positive correlation between our benchmark and the downstream alignment task performance. Based on our benchmark, we conduct extensive analysis on the state-of-the-art RMs, revealing their generalization defects that were not discovered by previous benchmarks, and highlighting the potential of generative RMs. Furthermore, we delve into open questions in reward models, specifically examining the effectiveness of majority voting for the evaluation of reward models and analyzing the impact factors of generative RMs, including the influence of evaluation criteria and instructing methods. Our evaluation code and datasets are available at https://github.com/Zhou-Zoey/RMB-Reward-Model-Benchmark.

Assessing the capabilities of large multimodal models (LMMs) often requires the creation of ad-hoc evaluations. Currently, building new benchmarks requires tremendous amounts of manual work for each specific analysis. This makes the evaluation process tedious and costly. In this paper, we present APEx, Automatic Programming of Experiments, the first framework for automatic benchmarking of LMMs. Given a research question expressed in natural language, APEx leverages a large language model (LLM) and a library of pre-specified tools to generate a set of experiments for the model at hand, and progressively compile a scientific report. The report drives the testing procedure: based on the current status of the investigation, APEx chooses which experiments to perform and whether the results are sufficient to draw conclusions. Finally, the LLM refines the report, presenting the results to the user in natural language. Thanks to its modularity, our framework is flexible and extensible as new tools become available. Empirically, APEx reproduces the findings of existing studies while allowing for arbitrary analyses and hypothesis testing.

Reinforcement Learning from Human Feedback (RLHF) has become a crucial technology for aligning language models with human values and intentions, enabling models to produce more helpful and harmless responses. Reward models are trained as proxies for human preferences to drive reinforcement learning optimization. While reward models are often considered central to achieving high performance, they face the following challenges in practical applications: (1) Incorrect and ambiguous preference pairs in the dataset may hinder the reward model from accurately capturing human intent. (2) Reward models trained on data from a specific distribution often struggle to generalize to examples outside that distribution and are not suitable for iterative RLHF training. In this report, we attempt to address these two issues. (1) From a data perspective, we propose a method to measure the strength of preferences within the data, based on a voting mechanism of multiple reward models. Experimental results confirm that data with varying preference strengths have different impacts on reward model performance. We introduce a series of novel methods to mitigate the influence of incorrect and ambiguous preferences in the dataset and fully leverage high-quality preference data. (2) From an algorithmic standpoint, we introduce contrastive learning to enhance the ability of reward models to distinguish between chosen and rejected responses, thereby improving model generalization. Furthermore, we employ meta-learning to enable the reward model to maintain the ability to differentiate subtle differences in out-of-distribution samples, and this approach can be utilized for iterative RLHF optimization.

The automation of scientific discovery has been a long-standing goal within the research community, driven by the potential to accelerate knowledge creation. While significant progress has been made using commercial large language models (LLMs) as research assistants or idea generators, the possibility of automating the entire research process with open-source LLMs remains largely unexplored. This paper explores the feasibility of using open-source post-trained LLMs as autonomous agents capable of performing the full cycle of automated research and review, from literature review and manuscript preparation to peer review and paper revision. Our iterative preference training framework consists of CycleResearcher, which conducts research tasks, and CycleReviewer, which simulates the peer review process, providing iterative feedback via reinforcement learning. To train these models, we develop two new datasets, Review-5k and Research-14k, reflecting real-world machine learning research and peer review dynamics. Our results demonstrate that CycleReviewer achieves a 26.89\% improvement in mean absolute error (MAE) over individual human reviewers in predicting paper scores, indicating that LLMs can surpass expert-level performance in research evaluation. In research, the papers generated by the CycleResearcher model achieved a score of 5.36 in simulated peer reviews, surpassing the preprint level of 5.24 from human experts and approaching the accepted paper level of 5.69. This work represents a significant step toward fully automated scientific inquiry, providing ethical safeguards and advancing AI-driven research capabilities. The code, dataset and model weight are released at http://github/minjun-zhu/Researcher.

The rapid progress in generative models has resulted in impressive leaps in generation quality, blurring the lines between synthetic and real data. Web-scale datasets are now prone to the inevitable contamination by synthetic data, directly impacting the training of future generated models. Already, some theoretical results on self-consuming generative models (a.k.a., iterative retraining) have emerged in the literature, showcasing that either model collapse or stability could be possible depending on the fraction of generated data used at each retraining step. However, in practice, synthetic data is often subject to human feedback and curated by users before being used and uploaded online. For instance, many interfaces of popular text-to-image generative models, such as Stable Diffusion or Midjourney, produce several variations of an image for a given query which can eventually be curated by the users. In this paper, we theoretically study the impact of data curation on iterated retraining of generative models and show that it can be seen as an implicit preference optimization mechanism. However, unlike standard preference optimization, the generative model does not have access to the reward function or negative samples needed for pairwise comparisons. Moreover, our study doesn't require access to the density function, only to samples. We prove that, if the data is curated according to a reward model, then the expected reward of the iterative retraining procedure is maximized. We further provide theoretical results on the stability of the retraining loop when using a positive fraction of real data at each step. Finally, we conduct illustrative experiments on both synthetic datasets and on CIFAR10 showing that such a procedure amplifies biases of the reward model.

Recent advances in large language models (LLMs) have led to the development of various evaluation benchmarks. These benchmarks typically rely on a single instruction template for evaluating all LLMs on a specific task. In this paper, we comprehensively analyze the brittleness of results obtained via single-prompt evaluations across 6.5M instances, involving 20 different LLMs and 39 tasks from 3 benchmarks. To improve robustness of the analysis, we propose to evaluate LLMs with a set of diverse prompts instead. We discuss tailored evaluation metrics for specific use cases (e.g., LLM developers vs. developers interested in a specific downstream task), ensuring a more reliable and meaningful assessment of LLM capabilities. We then implement these criteria and conduct evaluations of multiple models, providing insights into the true strengths and limitations of current LLMs.

State-of-the-art language models (LMs) are notoriously susceptible to generating hallucinated information. Such inaccurate outputs not only undermine the reliability of these models but also limit their use and raise serious concerns about misinformation and propaganda. In this work, we focus on hallucinated book and article references and present them as the "model organism" of language model hallucination research, due to their frequent and easy-to-discern nature. We posit that if a language model cites a particular reference in its output, then it should ideally possess sufficient information about its authors and content, among other relevant details. Using this basic insight, we illustrate that one can identify hallucinated references without ever consulting any external resources, by asking a set of direct or indirect queries to the language model about the references. These queries can be considered as "consistency checks." Our findings highlight that while LMs, including GPT-4, often produce inconsistent author lists for hallucinated references, they also often accurately recall the authors of real references. In this sense, the LM can be said to "know" when it is hallucinating references. Furthermore, these findings show how hallucinated references can be dissected to shed light on their nature. Replication code and results can be found at https://github.com/microsoft/hallucinated-references.

There is an increasing trend towards evaluating NLP models with LLM-generated judgments instead of human judgments. In the absence of a comparison against human data, this raises concerns about the validity of these evaluations; in case they are conducted with proprietary models, this also raises concerns over reproducibility. We provide JUDGE-BENCH, a collection of 20 NLP datasets with human annotations, and comprehensively evaluate 11 current LLMs, covering both open-weight and proprietary models, for their ability to replicate the annotations. Our evaluations show that each LLM exhibits a large variance across datasets in its correlation to human judgments. We conclude that LLMs are not yet ready to systematically replace human judges in NLP.

Despite the significant advancements in keyphrase extraction and keyphrase generation methods, the predominant approach for evaluation only relies on exact matching with human references and disregards reference-free attributes. This scheme fails to recognize systems that generate keyphrases that are semantically equivalent to the references or keyphrases that have practical utility. To better understand the strengths and weaknesses of different keyphrase systems, we propose a comprehensive evaluation framework consisting of six critical dimensions: naturalness, faithfulness, saliency, coverage, diversity, and utility. For each dimension, we discuss the desiderata and design semantic-based metrics that align with the evaluation objectives. Rigorous meta-evaluation studies demonstrate that our evaluation strategy correlates better with human preferences compared to a range of previously used metrics. Using this framework, we re-evaluate 18 keyphrase systems and further discover that (1) the best model differs in different dimensions, with pre-trained language models achieving the best in most dimensions; (2) the utility in downstream tasks does not always correlate well with reference-based metrics; and (3) large language models exhibit a strong performance in reference-free evaluation.

Training large language models (LLMs) for external tool usage is a rapidly expanding field, with recent research focusing on generating synthetic data to address the shortage of available data. However, the absence of systematic data quality checks poses complications for properly training and testing models. To that end, we propose two approaches for assessing the reliability of data for training LLMs to use external tools. The first approach uses intuitive, human-defined correctness criteria. The second approach uses a model-driven assessment with in-context evaluation. We conduct a thorough evaluation of data quality on two popular benchmarks, followed by an extrinsic evaluation that showcases the impact of data quality on model performance. Our results demonstrate that models trained on high-quality data outperform those trained on unvalidated data, even when trained with a smaller quantity of data. These findings empirically support the significance of assessing and ensuring the reliability of training data for tool-using LLMs.

Large Language Models (LLMs) are often described as being instances of foundation models - that is, models that transfer strongly across various tasks and conditions in few-show or zero-shot manner, while exhibiting scaling laws that predict function improvement when increasing the pre-training scale. These claims of excelling in different functions and tasks rely on measurements taken across various sets of standardized benchmarks showing high scores for such models. We demonstrate here a dramatic breakdown of function and reasoning capabilities of state-of-the-art models trained at the largest available scales which claim strong function, using a simple, short, conventional common sense problem formulated in concise natural language, easily solvable by humans. The breakdown is dramatic, as models also express strong overconfidence in their wrong solutions, while providing often non-sensical "reasoning"-like explanations akin to confabulations to justify and backup the validity of their clearly failed responses, making them sound plausible. Various standard interventions in an attempt to get the right solution, like various type of enhanced prompting, or urging the models to reconsider the wrong solutions again by multi step re-evaluation, fail. We take these initial observations to the scientific and technological community to stimulate urgent re-assessment of the claimed capabilities of current generation of LLMs, Such re-assessment also requires common action to create standardized benchmarks that would allow proper detection of such basic reasoning deficits that obviously manage to remain undiscovered by current state-of-the-art evaluation procedures and benchmarks. Code for reproducing experiments in the paper and raw experiments data can be found at https://github.com/LAION-AI/AIW

Generalization is arguably the most important goal of statistical language modeling research. Publicly available benchmarks and papers published with an open-source code have been critical to advancing the field. However, it is often very difficult, and sometimes even impossible, to reproduce the results fully as reported in publications. In this paper, we propose a simple framework that should help advance the state of the art in language modeling in terms of generalization. We propose to publish not just the code, but also probabilities on dev and test sets with future publications so that one can easily add the new model into an ensemble. This has crucial advantages: it is much easier to determine whether a newly proposed model is actually complementary to the current baseline. Therefore, instead of inventing new names for the old tricks, the scientific community can advance faster. Finally, this approach promotes diversity of ideas: one does not need to create an individual model that is the new state of the art to attract attention; it will be sufficient to develop a new model that learns patterns which other models do not. Thus, even a suboptimal model can be found to have value. Remarkably, our approach has yielded new state-of-the-art results across various language modeling benchmarks up to 10%.

As language models have scaled both their number of parameters and pretraining dataset sizes, the computational cost for pretraining has become intractable except for the most well-resourced teams. This increasing cost makes it ever more important to be able to reuse a model after it has completed pretraining; allowing for a model's abilities to further improve without needing to train from scratch. In this work, we detail a set of guidelines that cover how to design efficacious data distributions and learning rate schedules for continued pretraining of language models. When applying these findings within a continued pretraining run on top of a well-trained 15B parameter model, we show an improvement of 9\% in average model accuracy compared to the baseline of continued training on the pretraining set. The resulting recipe provides a practical starting point with which to begin developing language models through reuse rather than retraining.

When deploying large language models (LLMs), it is important to ensure that these models are not only capable, but also reliable. Many benchmarks have been created to track LLMs' growing capabilities, however there has been no similar focus on measuring their reliability. To understand the potential ramifications of this gap, we investigate how well current benchmarks quantify model reliability. We find that pervasive label errors can compromise these evaluations, obscuring lingering model failures and hiding unreliable behavior. Motivated by this gap in the evaluation of reliability, we then propose the concept of so-called platinum benchmarks, i.e., benchmarks carefully curated to minimize label errors and ambiguity. As a first attempt at constructing such benchmarks, we revise examples from fifteen existing popular benchmarks. We evaluate a wide range of models on these platinum benchmarks and find that, indeed, frontier LLMs still exhibit failures on simple tasks such as elementary-level math word problems. Analyzing these failures further reveals previously unidentified patterns of problems on which frontier models consistently struggle. We provide code at https://github.com/MadryLab/platinum-benchmarks

Significant progress has been made in automatic text evaluation with the introduction of large language models (LLMs) as evaluators. However, current sample-wise evaluation paradigm suffers from the following issues: (1) Sensitive to prompt design; (2) Poor resistance to noise; (3) Inferior ensemble performance with static reference. Inspired by the fact that humans treat both criterion definition and inter sample comparison as references for evaluation, we propose BatchEval, a paradigm that conducts batch-wise evaluation iteratively to alleviate the above problems. We explore variants under this paradigm and confirm the optimal settings are two stage procedure with heterogeneous batch composition strategy and decimal scoring format. Comprehensive experiments across 3 LLMs on 4 text evaluation tasks demonstrate that BatchEval outperforms state-of-the-art methods by 10.5% on Pearson correlations with only 64% API cost on average. Further analyses have been conducted to verify the robustness, generalization, and working mechanism of BatchEval.

Despite recent progress, it has been difficult to prevent semantic hallucinations in generative Large Language Models. One common solution to this is augmenting LLMs with a retrieval system and making sure that the generated output is attributable to the retrieved information. Given this new added constraint, it is plausible to expect that the overall quality of the output will be affected, for example, in terms of fluency. Can scaling language models help? Here we examine the relationship between fluency and attribution in LLMs prompted with retrieved evidence in knowledge-heavy dialog settings. Our experiments were implemented with a set of auto-metrics that are aligned with human preferences. They were used to evaluate a large set of generations, produced under varying parameters of LLMs and supplied context. We show that larger models tend to do much better in both fluency and attribution, and that (naively) using top-k retrieval versus top-1 retrieval improves attribution but hurts fluency. We next propose a recipe that could allow smaller models to both close the gap with larger models and preserve the benefits of top-k retrieval while avoiding its drawbacks.

As large language models (LLMs) continue to advance, accurately and comprehensively evaluating their performance becomes increasingly challenging. Human evaluations are conventionally considered the gold standard in natural language generation, but recent advancements incorporate state-of-the-art LLMs as proxies for human judges in evaluation processes. However, the extent to which humans and LLMs are capable evaluators remains uncertain. This study investigates the behavior of crowd-sourced and expert annotators, as well as LLMs, when comparing outputs from different models. To achieve this, we curate a dataset of intentionally flawed machine-generated answers. Our findings reveal a concerning bias in the evaluation process, as answers with factual errors are rated more favorably than answers that are too short or contained grammatical errors. To address this issue, we propose independently evaluating machine-generated text across multiple dimensions, rather than merging all the evaluation aspects into a single score. We instantiate this idea with the Elo rating system, resulting in the Multi-Elo Rating System. Empirical results from our study reveal that this proposed approach significantly enhances the quality of LLM-based evaluations, particularly in terms of factual accuracy. However, there is no significant improvement in crowd-sourced-based evaluations, indicating the need for further investigation and refinement.

The effectiveness of multi-stage text retrieval has been solidly demonstrated since before the era of pre-trained language models. However, most existing studies utilize models that predate recent advances in large language models (LLMs). This study seeks to explore potential improvements that state-of-the-art LLMs can bring. We conduct a comprehensive study, fine-tuning the latest LLaMA model both as a dense retriever (RepLLaMA) and as a pointwise reranker (RankLLaMA) for both passage retrieval and document retrieval using the MS MARCO datasets. Our findings demonstrate that the effectiveness of large language models indeed surpasses that of smaller models. Additionally, since LLMs can inherently handle longer contexts, they can represent entire documents holistically, obviating the need for traditional segmenting and pooling strategies. Furthermore, evaluations on BEIR demonstrate that our RepLLaMA-RankLLaMA pipeline exhibits strong zero-shot effectiveness. Model checkpoints from this study are available on HuggingFace.

There is a multitude of novel generative models for open-domain conversational systems; however, there is no systematic evaluation of different systems. Systematic comparisons require consistency in experimental design, evaluation sets, conversational systems and their outputs, and statistical analysis. We lay out a protocol for the evaluation of conversational models using head-to-head pairwise comparison. We analyze ten recent models that claim state-of-the-art performance using a paired head-to-head performance (win-loss-tie) on five evaluation datasets. Our findings show that DialoGPT and Blender are superior systems using Bradley-Terry model and TrueSkill ranking methods. These findings demonstrate the feasibility of our protocol to evaluate conversational agents and evaluation sets. Finally, we make all code and evaluations publicly available for researchers to compare their model to other state-of-the-art dialog models.

We propose MMLU-SR, a novel dataset designed to measure the true comprehension abilities of Large Language Models (LLMs) by challenging their performance in question-answering tasks with modified terms. We reasoned that an agent that ``truly'' understands a concept can still evaluate it when key terms are replaced by suitably defined alternate terms, and sought to differentiate such comprehension from mere text replacement. In our study, we modified standardized test questions by replacing a key term with a dummy word along with its definition. The key term could be in the context of questions, answers, or both questions and answers. Notwithstanding the high scores achieved by recent popular LLMs on the MMLU leaderboard, we found a substantial reduction in model performance after such replacement, suggesting poor comprehension. This new benchmark provides a rigorous benchmark for testing true model comprehension, and poses a challenge to the broader scientific community.

Topic model evaluation, like evaluation of other unsupervised methods, can be contentious. However, the field has coalesced around automated estimates of topic coherence, which rely on the frequency of word co-occurrences in a reference corpus. Contemporary neural topic models surpass classical ones according to these metrics. At the same time, topic model evaluation suffers from a validation gap: automated coherence, developed for classical models, has not been validated using human experimentation for neural models. In addition, a meta-analysis of topic modeling literature reveals a substantial standardization gap in automated topic modeling benchmarks. To address the validation gap, we compare automated coherence with the two most widely accepted human judgment tasks: topic rating and word intrusion. To address the standardization gap, we systematically evaluate a dominant classical model and two state-of-the-art neural models on two commonly used datasets. Automated evaluations declare a winning model when corresponding human evaluations do not, calling into question the validity of fully automatic evaluations independent of human judgments.

Work on scaling laws has found that large language models (LMs) show predictable improvements to overall loss with increased scale (model size, training data, and compute). Here, we present evidence for the claim that LMs may show inverse scaling, or worse task performance with increased scale, e.g., due to flaws in the training objective and data. We present empirical evidence of inverse scaling on 11 datasets collected by running a public contest, the Inverse Scaling Prize, with a substantial prize pool. Through analysis of the datasets, along with other examples found in the literature, we identify four potential causes of inverse scaling: (i) preference to repeat memorized sequences over following in-context instructions, (ii) imitation of undesirable patterns in the training data, (iii) tasks containing an easy distractor task which LMs could focus on, rather than the harder real task, and (iv) correct but misleading few-shot demonstrations of the task. We release the winning datasets at https://inversescaling.com/data to allow for further investigation of inverse scaling. Our tasks have helped drive the discovery of U-shaped and inverted-U scaling trends, where an initial trend reverses, suggesting that scaling trends are less reliable at predicting the behavior of larger-scale models than previously understood. Overall, our results suggest that there are tasks for which increased model scale alone may not lead to progress, and that more careful thought needs to go into the data and objectives for training language models.

Just like the previous generation of task-tuned models, large language models (LLMs) that are adapted to tasks via prompt-based methods like in-context-learning (ICL) perform well in some setups but not in others. This lack of consistency in prompt-based learning hints at a lack of robust generalisation. We here introduce the ICL consistency test -- a contribution to the GenBench collaborative benchmark task (CBT) -- which evaluates how consistent a model makes predictions across many different setups while using the same data. The test is based on different established natural language inference tasks. We provide preprocessed data constituting 96 different 'setups' and a metric that estimates model consistency across these setups. The metric is provided on a fine-grained level to understand what properties of a setup render predictions unstable and on an aggregated level to compare overall model consistency. We conduct an empirical analysis of eight state-of-the-art models, and our consistency metric reveals how all tested LLMs lack robust generalisation.

The large-scale training of multi-modal models on data scraped from the web has shown outstanding utility in infusing these models with the required world knowledge to perform effectively on multiple downstream tasks. However, one downside of scraping data from the web can be the potential sacrifice of the benchmarks on which the abilities of these models are often evaluated. To safeguard against test data contamination and to truly test the abilities of these foundation models we propose LiveXiv: A scalable evolving live benchmark based on scientific ArXiv papers. LiveXiv accesses domain-specific manuscripts at any given timestamp and proposes to automatically generate visual question-answer pairs (VQA). This is done without any human-in-the-loop, using the multi-modal content in the manuscripts, like graphs, charts, and tables. Moreover, we introduce an efficient evaluation approach that estimates the performance of all models on the evolving benchmark using evaluations of only a subset of models. This significantly reduces the overall evaluation cost. We benchmark multiple open and proprietary Large Multi-modal Models (LMMs) on the first version of our benchmark, showing its challenging nature and exposing the models true abilities, avoiding contamination. Lastly, in our commitment to high quality, we have collected and evaluated a manually verified subset. By comparing its overall results to our automatic annotations, we have found that the performance variance is indeed minimal (<2.5%). Our dataset is available online on HuggingFace, and our code will be available here.

We introduce a new framework for sample-efficient model evaluation that we call active testing. While approaches like active learning reduce the number of labels needed for model training, existing literature largely ignores the cost of labeling test data, typically unrealistically assuming large test sets for model evaluation. This creates a disconnect to real applications, where test labels are important and just as expensive, e.g. for optimizing hyperparameters. Active testing addresses this by carefully selecting the test points to label, ensuring model evaluation is sample-efficient. To this end, we derive theoretically-grounded and intuitive acquisition strategies that are specifically tailored to the goals of active testing, noting these are distinct to those of active learning. As actively selecting labels introduces a bias; we further show how to remove this bias while reducing the variance of the estimator at the same time. Active testing is easy to implement and can be applied to any supervised machine learning method. We demonstrate its effectiveness on models including WideResNets and Gaussian processes on datasets including Fashion-MNIST and CIFAR-100.

There have been many benchmarks for evaluating long-context language models (LCLMs), but developers often rely on synthetic tasks like needle-in-a-haystack (NIAH) or arbitrary subsets of tasks. It remains unclear whether they translate to the diverse downstream applications of LCLMs, and the inconsistency further complicates model comparison. We investigate the underlying reasons behind current practices and find that existing benchmarks often provide noisy signals due to low coverage of applications, insufficient lengths, unreliable metrics, and incompatibility with base models. In this work, we present HELMET (How to Evaluate Long-context Models Effectively and Thoroughly), a comprehensive benchmark encompassing seven diverse, application-centric categories. We also address many issues in previous benchmarks by adding controllable lengths up to 128k tokens, model-based evaluation for reliable metrics, and few-shot prompting for robustly evaluating base models. Consequently, we demonstrate that HELMET offers more reliable and consistent rankings of frontier LCLMs. Through a comprehensive study of 51 LCLMs, we find that (1) synthetic tasks like NIAH are not good predictors of downstream performance; (2) the diverse categories in HELMET exhibit distinct trends and low correlation with each other; and (3) while most LCLMs achieve perfect NIAH scores, open-source models significantly lag behind closed ones when the task requires full-context reasoning or following complex instructions -- the gap widens with increased lengths. Finally, we recommend using our RAG tasks for fast model development, as they are easy to run and more predictive of other downstream performance; ultimately, we advocate for a holistic evaluation across diverse tasks.

Large language models (LLMs) can produce erroneous responses that sound fluent and convincing, raising the risk that users will rely on these responses as if they were correct. Mitigating such overreliance is a key challenge. Through a think-aloud study in which participants use an LLM-infused application to answer objective questions, we identify several features of LLM responses that shape users' reliance: explanations (supporting details for answers), inconsistencies in explanations, and sources. Through a large-scale, pre-registered, controlled experiment (N=308), we isolate and study the effects of these features on users' reliance, accuracy, and other measures. We find that the presence of explanations increases reliance on both correct and incorrect responses. However, we observe less reliance on incorrect responses when sources are provided or when explanations exhibit inconsistencies. We discuss the implications of these findings for fostering appropriate reliance on LLMs.

The rapid advancement of Large Language Models (LLMs) and Large Multimodal Models (LMMs) has heightened the demand for AI-based scientific assistants capable of understanding scientific articles and figures. Despite progress, there remains a significant gap in evaluating models' comprehension of professional, graduate-level, and even PhD-level scientific content. Current datasets and benchmarks primarily focus on relatively simple scientific tasks and figures, lacking comprehensive assessments across diverse advanced scientific disciplines. To bridge this gap, we collected a multimodal, multidisciplinary dataset from open-access scientific articles published in Nature Communications journals. This dataset spans 72 scientific disciplines, ensuring both diversity and quality. We created benchmarks with various tasks and settings to comprehensively evaluate LMMs' capabilities in understanding scientific figures and content. Our evaluation revealed that these tasks are highly challenging: many open-source models struggled significantly, and even GPT-4V and GPT-4o faced difficulties. We also explored using our dataset as training resources by constructing visual instruction-following data, enabling the 7B LLaVA model to achieve performance comparable to GPT-4V/o on our benchmark. Additionally, we investigated the use of our interleaved article texts and figure images for pre-training LMMs, resulting in improvements on the material generation task. The source dataset, including articles, figures, constructed benchmarks, and visual instruction-following data, is open-sourced.

We present SeaEval, a benchmark for multilingual foundation models. In addition to characterizing how these models understand and reason with natural language, we also investigate how well they comprehend cultural practices, nuances, and values. Alongside standard accuracy metrics, we investigate the brittleness of foundation models in the dimensions of semantics and multilinguality. Our analyses span both open-sourced and closed models, leading to empirical results across classic NLP tasks, reasoning, and cultural comprehension. Key findings indicate (1) Most models exhibit varied behavior when given paraphrased instructions. (2) Many models still suffer from exposure bias (e.g., positional bias, majority label bias). (3) For questions rooted in factual, scientific, and commonsense knowledge, consistent responses are expected across multilingual queries that are semantically equivalent. Yet, most models surprisingly demonstrate inconsistent performance on these queries. (4) Multilingually-trained models have not attained "balanced multilingual" capabilities. Our endeavors underscore the need for more generalizable semantic representations and enhanced multilingual contextualization. SeaEval can serve as a launchpad for more thorough investigations and evaluations for multilingual and multicultural scenarios.

Recent advances in large language models (LLMs) show the potential of using LLMs as evaluators for assessing the quality of text generations from LLMs. However, applying LLM evaluators naively to compare or judge between different systems can lead to unreliable results due to the intrinsic win rate estimation bias of LLM evaluators. In order to mitigate this problem, we propose two calibration methods, Bayesian Win Rate Sampling (BWRS) and Bayesian Dawid-Skene, both of which leverage Bayesian inference to more accurately infer the true win rate of generative language models. We empirically validate our methods on six datasets covering story generation, summarization, and instruction following tasks. We show that both our methods are effective in improving the accuracy of win rate estimation using LLMs as evaluators, offering a promising direction for reliable automatic text quality evaluation.

Augmenting large language models (LLMs) with external tools is a promising approach to enhance their capabilities. Effectively leveraging this potential for complex tasks hinges crucially on improving their ability to use tools. Synthesizing tool use data by simulating the real world is an effective approach. Nevertheless, our investigation reveals that training gains significantly decay as the scale of these data increases. The primary factor is the model's poor performance (a.k.a deficiency) in complex scenarios, which hinders learning from data using SFT. Driven by this objective, we propose an iterative reinforced fine-tuning strategy to continually guide the model to alleviate it. Specifically, we first identify deficiency-related data based on feedback from the policy model, then perform a Monte Carlo Tree Search to collect fine-grained preference pairs to pinpoint deficiencies. Subsequently, we update the policy model using preference optimization to align with ground truth and misalign with deficiencies. This process can be iterated. Moreover, before the iteration, we propose an easy-to-hard warm-up SFT strategy to facilitate learning from challenging data. The experiments demonstrate our models go beyond the same parametric models, outperforming many larger open-source and closed-source models. Additionally, it has achieved notable training gains in complex tool use scenarios.

Large language models (LLMs) have emerged as pivotal contributors in contemporary natural language processing and are increasingly being applied across a diverse range of industries. However, these large-scale probabilistic statistical models cannot currently ensure the requisite quality in professional content generation. These models often produce hallucinated text, compromising their practical utility in professional contexts. To assess the authentic reliability of LLMs in text generation, numerous initiatives have developed benchmark evaluations for hallucination phenomena. Nevertheless, these benchmarks frequently utilize constrained generation techniques due to cost and temporal constraints. These techniques encompass the use of directed hallucination induction and strategies that deliberately alter authentic text to produce hallucinations. These approaches are not congruent with the unrestricted text generation demanded by real-world applications. Furthermore, a well-established Chinese-language dataset dedicated to the evaluation of hallucinations in text generation is presently lacking. Consequently, we have developed an Unconstrained Hallucination Generation Evaluation (UHGEval) benchmark, designed to compile outputs produced with minimal restrictions by LLMs. Concurrently, we have established a comprehensive benchmark evaluation framework to aid subsequent researchers in undertaking scalable and reproducible experiments. We have also executed extensive experiments, evaluating prominent Chinese language models and the GPT series models to derive professional performance insights regarding hallucination challenges.

As language models (LMs) become increasingly powerful, it is important to quantify and compare them for sociodemographic bias with potential for harm. Prior bias measurement datasets are sensitive to perturbations in their manually designed templates, therefore unreliable. To achieve reliability, we introduce the Comprehensive Assessment of Language Model bias (CALM), a benchmark dataset to quantify bias in LMs across three tasks. We integrate 16 existing datasets across different domains, such as Wikipedia and news articles, to filter 224 templates from which we construct a dataset of 78,400 examples. We compare the diversity of CALM with prior datasets on metrics such as average semantic similarity, and variation in template length, and test the sensitivity to small perturbations. We show that our dataset is more diverse and reliable than previous datasets, thus better capture the breadth of linguistic variation required to reliably evaluate model bias. We evaluate 20 large language models including six prominent families of LMs such as Llama-2. In two LM series, OPT and Bloom, we found that larger parameter models are more biased than lower parameter models. We found the T0 series of models to be the least biased. Furthermore, we noticed a tradeoff between gender and racial bias with increasing model size in some model series. The code is available at https://github.com/vipulgupta1011/CALM.

The past two decades have witnessed the rapid development of personalized recommendation techniques. Despite significant progress made in both research and practice of recommender systems, to date, there is a lack of a widely-recognized benchmarking standard in this field. Many existing studies perform model evaluations and comparisons in an ad-hoc manner, for example, by employing their own private data splits or using different experimental settings. Such conventions not only increase the difficulty in reproducing existing studies, but also lead to inconsistent experimental results among them. This largely limits the credibility and practical value of research results in this field. To tackle these issues, we present an initiative project (namely BARS) aiming for open benchmarking for recommender systems. In comparison to some earlier attempts towards this goal, we take a further step by setting up a standardized benchmarking pipeline for reproducible research, which integrates all the details about datasets, source code, hyper-parameter settings, running logs, and evaluation results. The benchmark is designed with comprehensiveness and sustainability in mind. It covers both matching and ranking tasks, and also enables researchers to easily follow and contribute to the research in this field. This project will not only reduce the redundant efforts of researchers to re-implement or re-run existing baselines, but also drive more solid and reproducible research on recommender systems. We would like to call upon everyone to use the BARS benchmark for future evaluation, and contribute to the project through the portal at: https://openbenchmark.github.io/BARS.

Transformer language models have received widespread public attention, yet their generated text is often surprising even to NLP researchers. In this survey, we discuss over 250 recent studies of English language model behavior before task-specific fine-tuning. Language models possess basic capabilities in syntax, semantics, pragmatics, world knowledge, and reasoning, but these capabilities are sensitive to specific inputs and surface features. Despite dramatic increases in generated text quality as models scale to hundreds of billions of parameters, the models are still prone to unfactual responses, commonsense errors, memorized text, and social biases. Many of these weaknesses can be framed as over-generalizations or under-generalizations of learned patterns in text. We synthesize recent results to highlight what is currently known about what large language models can and cannot do.

As generative large language models (LLMs) grow more performant and prevalent, we must develop comprehensive enough tools to measure and improve their fairness. Different prompt-based datasets can be used to measure social bias across multiple text domains and demographic axes, meaning that testing LLMs on more datasets can potentially help us characterize their biases more fully, and better ensure equal and equitable treatment of marginalized demographic groups. In this work, our focus is two-fold: (1) Benchmarking: a comparison of 6 different prompt-based bias and toxicity metrics across 12 demographic axes and 5 families of generative LLMs. Out of those 6 metrics, AdvPromptSet and HolisticBiasR are novel datasets proposed in the paper. The comparison of those benchmarks gives us insights about the bias and toxicity of the compared models. Therefore, we explore the frequency of demographic terms in common LLM pre-training corpora and how this may relate to model biases. (2) Mitigation: we conduct a comprehensive study of how well 3 bias/toxicity mitigation techniques perform across our suite of measurements. ROBBIE aims to provide insights for practitioners while deploying a model, emphasizing the need to not only measure potential harms, but also understand how they arise by characterizing the data, mitigate harms once found, and balance any trade-offs. We open-source our analysis code in hopes of encouraging broader measurements of bias in future LLMs.

Many empirical studies of labor market questions rely on estimating relatively simple predictive models using small, carefully constructed longitudinal survey datasets based on hand-engineered features. Large Language Models (LLMs), trained on massive datasets, encode vast quantities of world knowledge and can be used for the next job prediction problem. However, while an off-the-shelf LLM produces plausible career trajectories when prompted, the probability with which an LLM predicts a particular job transition conditional on career history will not, in general, align with the true conditional probability in a given population. Recently, Vafa et al. (2024) introduced a transformer-based "foundation model", CAREER, trained using a large, unrepresentative resume dataset, that predicts transitions between jobs; it further demonstrated how transfer learning techniques can be used to leverage the foundation model to build better predictive models of both transitions and wages that reflect conditional transition probabilities found in nationally representative survey datasets. This paper considers an alternative where the fine-tuning of the CAREER foundation model is replaced by fine-tuning LLMs. For the task of next job prediction, we demonstrate that models trained with our approach outperform several alternatives in terms of predictive performance on the survey data, including traditional econometric models, CAREER, and LLMs with in-context learning, even though the LLM can in principle predict job titles that are not allowed in the survey data. Further, we show that our fine-tuned LLM-based models' predictions are more representative of the career trajectories of various workforce subpopulations than off-the-shelf LLM models and CAREER. We conduct experiments and analyses that highlight the sources of the gains in the performance of our models for representative predictions.

Recently, large language models (LLMs), especially those that are pretrained on code, have demonstrated strong capabilities in generating programs from natural language inputs in a few-shot or even zero-shot manner. Despite promising results, there is a notable lack of a comprehensive evaluation of these models language-to-code generation capabilities. Existing studies often focus on specific tasks, model architectures, or learning paradigms, leading to a fragmented understanding of the overall landscape. In this work, we present L2CEval, a systematic evaluation of the language-to-code generation capabilities of LLMs on 7 tasks across the domain spectrum of semantic parsing, math reasoning and Python programming, analyzing the factors that potentially affect their performance, such as model size, pretraining data, instruction tuning, and different prompting methods. In addition to assessing model performance, we measure confidence calibration for the models and conduct human evaluations of the output programs. This enables us to identify and analyze the typical failure modes across various tasks and models. L2CEval offers a comprehensive understanding of the capabilities and limitations of LLMs in language-to-code generation. We also release the evaluation framework and all model outputs, hoping to lay the groundwork for further future research in this domain.

Memorization, or the tendency of large language models (LLMs) to output entire sequences from their training data verbatim, is a key concern for safely deploying language models. In particular, it is vital to minimize a model's memorization of sensitive datapoints such as those containing personal identifiable information (PII). The prevalence of such undesirable memorization can pose issues for model trainers, and may even require discarding an otherwise functional model. We therefore seek to predict which sequences will be memorized before a large model's full train-time by extrapolating the memorization behavior of lower-compute trial runs. We measure memorization of the Pythia model suite and plot scaling laws for forecasting memorization, allowing us to provide equi-compute recommendations to maximize the reliability (recall) of such predictions. We additionally provide further novel discoveries on the distribution of memorization scores across models and data. We release all code and data necessary to reproduce the results in this paper at https://github.com/EleutherAI/pythia

The recent proliferation of research into transformer based natural language processing has led to a number of studies which attempt to detect the presence of human-like cognitive behavior in the models. We contend that, as is true of human psychology, the investigation of cognitive behavior in language models must be conducted in an appropriate population of an appropriate size for the results to be meaningful. We leverage work in uncertainty estimation in a novel approach to efficiently construct experimental populations. The resultant tool, PopulationLM, has been made open source. We provide theoretical grounding in the uncertainty estimation literature and motivation from current cognitive work regarding language models. We discuss the methodological lessons from other scientific communities and attempt to demonstrate their application to two artificial population studies. Through population based experimentation we find that language models exhibit behavior consistent with typicality effects among categories highly represented in training. However, we find that language models don't tend to exhibit structural priming effects. Generally, our results show that single models tend to over estimate the presence of cognitive behaviors in neural models.

Large Language Models (LLMs) changed the way we design and interact with software systems. Their ability to process and extract information from text has drastically improved productivity in a number of routine tasks. Developers that want to include these models in their software stack, however, face a dreadful challenge: debugging LLMs' inconsistent behavior across minor variations of the prompt. We therefore introduce two metrics for classification tasks, namely sensitivity and consistency, which are complementary to task performance. First, sensitivity measures changes of predictions across rephrasings of the prompt, and does not require access to ground truth labels. Instead, consistency measures how predictions vary across rephrasings for elements of the same class. We perform an empirical comparison of these metrics on text classification tasks, using them as guideline for understanding failure modes of the LLM. Our hope is that sensitivity and consistency will be helpful to guide prompt engineering and obtain LLMs that balance robustness with performance.

Despite the utility of Large Language Models (LLMs) across a wide range of tasks and scenarios, developing a method for reliably evaluating LLMs across varied contexts continues to be challenging. Modern evaluation approaches often use LLMs to assess responses generated by LLMs. However, the meta-evaluation conducted to assess the effectiveness of these LLMs as evaluators is typically constrained by the coverage of existing benchmarks or requires extensive human annotation. This underscores the urgency of methods for scalable meta-evaluation that can effectively, reliably, and efficiently evaluate the performance of LLMs as evaluators across diverse tasks and scenarios, particularly in potentially new, user-defined scenarios. To fill this gap, we propose ScaleEval, an agent-debate-assisted meta-evaluation framework that leverages the capabilities of multiple communicative LLM agents. This framework supports multi-round discussions to assist human annotators in discerning the most capable LLMs as evaluators, which significantly eases their workload in cases that used to require large-scale annotations during meta-evaluation. We release the code for our framework, which is publicly available at: https://github.com/GAIR-NLP/scaleeval.

The rapid evolution of Multimodal Large Language Models (MLLMs) has brought substantial advancements in artificial intelligence, significantly enhancing the capability to understand and generate multimodal content. While prior studies have largely concentrated on model architectures and training methodologies, a thorough analysis of the benchmarks used for evaluating these models remains underexplored. This survey addresses this gap by systematically reviewing 211 benchmarks that assess MLLMs across four core domains: understanding, reasoning, generation, and application. We provide a detailed analysis of task designs, evaluation metrics, and dataset constructions, across diverse modalities. We hope that this survey will contribute to the ongoing advancement of MLLM research by offering a comprehensive overview of benchmarking practices and identifying promising directions for future work. An associated GitHub repository collecting the latest papers is available.

Large language models (LLMs) have demonstrated remarkable proficiency in generating detailed and coherent explanations of complex concepts. However, the extent to which these models truly comprehend the concepts they articulate remains unclear. To assess the level of comprehension of a model relative to the content it generates, we implemented a self-evaluation pipeline where models: (i) given a topic generate an excerpt with information about the topic, (ii) given an excerpt generate question-answer pairs, and finally (iii) given a question generate an answer. We refer to this self-evaluation approach as Explain-Query-Test (EQT). Interestingly, the accuracy on generated questions resulting from running the EQT pipeline correlates strongly with the model performance as verified by typical benchmarks such as MMLU-Pro. In other words, EQT's performance is predictive of MMLU-Pro's, and EQT can be used to rank models without the need for any external source of evaluation data other than lists of topics of interest. Moreover, our results reveal a disparity between the models' ability to produce detailed explanations and their performance on questions related to those explanations. This gap highlights fundamental limitations in the internal knowledge representation and reasoning abilities of current LLMs. We release the code at https://github.com/asgsaeid/EQT.

Most research about natural language generation (NLG) relies on evaluation benchmarks with limited references for a sample, which may result in poor correlations with human judgements. The underlying reason is that one semantic meaning can actually be expressed in different forms, and the evaluation with a single or few references may not accurately reflect the quality of the model's hypotheses. To address this issue, this paper presents a simple and effective method, named Div-Ref, to enhance existing evaluation benchmarks by enriching the number of references. We leverage large language models (LLMs) to diversify the expression of a single reference into multiple high-quality ones to cover the semantic space of the reference sentence as much as possible. We conduct comprehensive experiments to empirically demonstrate that diversifying the expression of reference can significantly enhance the correlation between automatic evaluation and human evaluation. This idea is compatible with recent LLM-based evaluation which can similarly derive advantages from incorporating multiple references. We strongly encourage future generation benchmarks to include more references, even if they are generated by LLMs, which is once for all. We release all the code and data at https://github.com/RUCAIBox/Div-Ref to facilitate research.

High-quality preference datasets are essential for training reward models that can effectively guide large language models (LLMs) in generating high-quality responses aligned with human preferences. As LLMs become stronger and better aligned, permissively licensed preference datasets, such as Open Assistant, HH-RLHF, and HelpSteer need to be updated to remain effective for reward modeling. Methods that distil preference data from proprietary LLMs such as GPT-4 have restrictions on commercial usage imposed by model providers. To improve upon both generated responses and attribute labeling quality, we release HelpSteer2, a permissively licensed preference dataset (CC-BY-4.0). Using a powerful internal base model trained on HelpSteer2, we are able to achieve the SOTA score (92.0%) on Reward-Bench's primary dataset, outperforming currently listed open and proprietary models, as of June 12th, 2024. Notably, HelpSteer2 consists of only ten thousand response pairs, an order of magnitude fewer than existing preference datasets (e.g., HH-RLHF), which makes it highly efficient for training reward models. Our extensive experiments demonstrate that reward models trained with HelpSteer2 are effective in aligning LLMs. In particular, we propose SteerLM 2.0, a model alignment approach that can effectively make use of the rich multi-attribute score predicted by our reward models. HelpSteer2 is available at https://huggingface.co/datasets/nvidia/HelpSteer2 and code is available at https://github.com/NVIDIA/NeMo-Aligner

We introduce an approach aimed at enhancing the reasoning capabilities of Large Language Models (LLMs) through an iterative preference learning process inspired by the successful strategy employed by AlphaZero. Our work leverages Monte Carlo Tree Search (MCTS) to iteratively collect preference data, utilizing its look-ahead ability to break down instance-level rewards into more granular step-level signals. To enhance consistency in intermediate steps, we combine outcome validation and stepwise self-evaluation, continually updating the quality assessment of newly generated data. The proposed algorithm employs Direct Preference Optimization (DPO) to update the LLM policy using this newly generated step-level preference data. Theoretical analysis reveals the importance of using on-policy sampled data for successful self-improving. Extensive evaluations on various arithmetic and commonsense reasoning tasks demonstrate remarkable performance improvements over existing models. For instance, our approach outperforms the Mistral-7B Supervised Fine-Tuning (SFT) baseline on GSM8K, MATH, and ARC-C, with substantial increases in accuracy to 81.8% (+5.9%), 34.7% (+5.8%), and 76.4% (+15.8%), respectively. Additionally, our research delves into the training and inference compute tradeoff, providing insights into how our method effectively maximizes performance gains. Our code is publicly available at https://github.com/YuxiXie/MCTS-DPO.

The evaluation of Handwritten Text Recognition (HTR) models during their development is straightforward: because HTR is a supervised problem, the usual data split into training, validation, and test data sets allows the evaluation of models in terms of accuracy or error rates. However, the evaluation process becomes tricky as soon as we switch from development to application. A compilation of a new (and forcibly smaller) ground truth (GT) from a sample of the data that we want to apply the model on and the subsequent evaluation of models thereon only provides hints about the quality of the recognised text, as do confidence scores (if available) the models return. Moreover, if we have several models at hand, we face a model selection problem since we want to obtain the best possible result during the application phase. This calls for GT-free metrics to select the best model, which is why we (re-)introduce and compare different metrics, from simple, lexicon-based to more elaborate ones using standard language models and masked language models (MLM). We show that MLM-based evaluation can compete with lexicon-based methods, with the advantage that large and multilingual transformers are readily available, thus making compiling lexical resources for other metrics superfluous.

To solve complex tasks, large language models (LLMs) often require multiple rounds of interactions with the user, sometimes assisted by external tools. However, current evaluation protocols often emphasize benchmark performance with single-turn exchanges, neglecting the nuanced interactions among the user, LLMs, and external tools, while also underestimating the importance of natural language feedback from users. These oversights contribute to discrepancies between research benchmark evaluations and real-world use cases. We introduce MINT, a benchmark that evaluates LLMs' ability to solve tasks with multi-turn interactions by (1) using tools and (2) leveraging natural language feedback. To ensure reproducibility, we provide an evaluation framework where LLMs can access tools by executing Python code and receive users' natural language feedback simulated by GPT-4. We repurpose a diverse set of established evaluation datasets focusing on reasoning, coding, and decision-making and carefully curate them into a compact subset for efficient evaluation. Our analysis of 20 open- and closed-source LLMs offers intriguing findings. (a) LLMs generally benefit from tools and language feedback, with performance gains (absolute, same below) of 1-8% for each turn of tool use and 2-17% with natural language feedback. (b) Better single-turn performance does not guarantee better multi-turn performance. (c) Surprisingly, on the LLMs evaluated, supervised instruction-finetuning (SIFT) and reinforcement learning from human feedback (RLHF) generally hurt multi-turn capabilities. We expect MINT can help measure progress and incentivize research in improving LLMs' capabilities in multi-turn interactions, especially for open-source communities where multi-turn human evaluation can be less accessible compared to commercial LLMs with a larger user base.

Large language models (LLMs) have a tendency to generate plausible-sounding yet factually incorrect responses, especially when queried on unfamiliar concepts. In this work, we explore the underlying mechanisms that govern how finetuned LLMs hallucinate. Our investigation reveals an interesting pattern: as inputs become more unfamiliar, LLM outputs tend to default towards a ``hedged'' prediction, whose form is determined by how the unfamiliar examples in the finetuning data are supervised. Thus, by strategically modifying these examples' supervision, we can control LLM predictions for unfamiliar inputs (e.g., teach them to say ``I don't know''). Based on these principles, we develop an RL approach that more reliably mitigates hallucinations for long-form generation tasks, by tackling the challenges presented by reward model hallucinations. We validate our findings with a series of controlled experiments in multiple-choice QA on MMLU, as well as long-form biography and book/movie plot generation tasks.

Large language models (LLM) have achieved remarkable performance on various NLP tasks and are augmented by tools for broader applications. Yet, how to evaluate and analyze the tool-utilization capability of LLMs is still under-explored. In contrast to previous works that evaluate models holistically, we comprehensively decompose the tool utilization into multiple sub-processes, including instruction following, planning, reasoning, retrieval, understanding, and review. Based on that, we further introduce T-Eval to evaluate the tool utilization capability step by step. T-Eval disentangles the tool utilization evaluation into several sub-domains along model capabilities, facilitating the inner understanding of both holistic and isolated competency of LLMs. We conduct extensive experiments on T-Eval and in-depth analysis of various LLMs. T-Eval not only exhibits consistency with the outcome-oriented evaluation but also provides a more fine-grained analysis of the capabilities of LLMs, providing a new perspective in LLM evaluation on tool-utilization ability. The benchmark will be available at https://github.com/open-compass/T-Eval.

The proliferation of generative models, combined with pretraining on web-scale data, raises a timely question: what happens when these models are trained on their own generated outputs? Recent investigations into model-data feedback loops proposed that such loops would lead to a phenomenon termed model collapse, under which performance progressively degrades with each model-data feedback iteration until fitted models become useless. However, those studies largely assumed that new data replace old data over time, where an arguably more realistic assumption is that data accumulate over time. In this paper, we ask: what effect does accumulating data have on model collapse? We empirically study this question by pretraining sequences of language models on text corpora. We confirm that replacing the original real data by each generation's synthetic data does indeed tend towards model collapse, then demonstrate that accumulating the successive generations of synthetic data alongside the original real data avoids model collapse; these results hold across a range of model sizes, architectures, and hyperparameters. We obtain similar results for deep generative models on other types of real data: diffusion models for molecule conformation generation and variational autoencoders for image generation. To understand why accumulating data can avoid model collapse, we use an analytically tractable framework introduced by prior work in which a sequence of linear models are fit to the previous models' outputs. Previous work used this framework to show that if data are replaced, the test error increases with the number of model-fitting iterations; we extend this argument to prove that if data instead accumulate, the test error has a finite upper bound independent of the number of iterations, meaning model collapse no longer occurs.

Recently, significant public efforts have been directed towards developing low-cost models with capabilities akin to ChatGPT, thereby fostering the growth of open-source conversational models. However, there remains a scarcity of comprehensive and in-depth evaluations of these models' performance. In this study, we examine the influence of training data factors, including quantity, quality, and linguistic distribution, on model performance. Our analysis is grounded in several publicly accessible, high-quality instruction datasets, as well as our own Chinese multi-turn conversations. We assess various models using a evaluation set of 1,000 samples, encompassing nine real-world scenarios. Our goal is to supplement manual evaluations with quantitative analyses, offering valuable insights for the continued advancement of open-source chat models. Furthermore, to enhance the performance and training and inference efficiency of models in the Chinese domain, we extend the vocabulary of LLaMA - the model with the closest open-source performance to proprietary language models like GPT-3 - and conduct secondary pre-training on 3.4B Chinese words. We make our model, data, as well as code publicly available.

In this work, we explore whether modeling recurrence into the Transformer architecture can both be beneficial and efficient, by building an extremely simple recurrent module into the Transformer. We compare our model to baselines following the training and evaluation recipe of BERT. Our results confirm that recurrence can indeed improve Transformer models by a consistent margin, without requiring low-level performance optimizations, and while keeping the number of parameters constant. For example, our base model achieves an absolute improvement of 2.1 points averaged across 10 tasks and also demonstrates increased stability in fine-tuning over a range of learning rates.

This paper explores the use of Large Language Models (LLMs) for sequential recommendation, which predicts users' future interactions based on their past behavior. We introduce a new concept, "Integrating Recommendation Systems as a New Language in Large Models" (RSLLM), which combines the strengths of traditional recommenders and LLMs. RSLLM uses a unique prompting method that combines ID-based item embeddings from conventional recommendation models with textual item features. It treats users' sequential behaviors as a distinct language and aligns the ID embeddings with the LLM's input space using a projector. We also propose a two-stage LLM fine-tuning framework that refines a pretrained LLM using a combination of two contrastive losses and a language modeling loss. The LLM is first fine-tuned using text-only prompts, followed by target domain fine-tuning with unified prompts. This trains the model to incorporate behavioral knowledge from the traditional sequential recommender into the LLM. Our empirical results validate the effectiveness of our proposed framework.

Perceiving and generating diverse modalities are crucial for AI models to effectively learn from and engage with real-world signals, necessitating reliable evaluations for their development. We identify two major issues in current evaluations: (1) inconsistent standards, shaped by different communities with varying protocols and maturity levels; and (2) significant query, grading, and generalization biases. To address these, we introduce MixEval-X, the first any-to-any real-world benchmark designed to optimize and standardize evaluations across input and output modalities. We propose multi-modal benchmark mixture and adaptation-rectification pipelines to reconstruct real-world task distributions, ensuring evaluations generalize effectively to real-world use cases. Extensive meta-evaluations show our approach effectively aligns benchmark samples with real-world task distributions and the model rankings correlate strongly with that of crowd-sourced real-world evaluations (up to 0.98). We provide comprehensive leaderboards to rerank existing models and organizations and offer insights to enhance understanding of multi-modal evaluations and inform future research.

Large language models (LLMs) have shown tremendous success in following user instructions and generating helpful responses. Nevertheless, their robustness is still far from optimal, as they may generate significantly inconsistent responses due to minor changes in the verbalized instructions. Recent literature has explored this inconsistency issue, highlighting the importance of continued improvement in the robustness of response generation. However, systematic analysis and solutions are still lacking. In this paper, we quantitatively define the inconsistency problem and propose a two-stage training framework consisting of instruction-augmented supervised fine-tuning and consistency alignment training. The first stage helps a model generalize on following instructions via similar instruction augmentations. In the second stage, we improve the diversity and help the model understand which responses are more aligned with human expectations by differentiating subtle differences in similar responses. The training process is accomplished by self-rewards inferred from the trained model at the first stage without referring to external human preference resources. We conduct extensive experiments on recent publicly available LLMs on instruction-following tasks and demonstrate the effectiveness of our training framework.

Reinforcement learning from human feedback (RLHF) has emerged as a powerful technique to make large language models (LLMs) more capable in complex settings. RLHF proceeds as collecting human preference data, training a reward model on said data, and optimizing a base ML model with respect to said reward for extrinsic evaluation metrics (e.g. MMLU, GSM8k). RLHF relies on many assumptions about how the various pieces fit together, such as a reward model capturing human preferences and an RL optimizer extracting the right signal from a reward model. As the RLHF process involves many distinct design decisions, it is easy to assume that multiple processes are correlated and therefore numerically linked. This apparent correlation is often not true, where reward models are easily overoptimized or RL optimizers can reduce performance on tasks not modeled in the data. Notable manifestations of models trained with imperfect RLHF systems are those that are prone to refusing basic requests for safety reasons or appearing lazy in generations. As chat model evaluation becomes increasingly nuanced, the reliance on a perceived link between reward model training, RL scores, and downstream performance drives these issues, which we describe as an objective mismatch. In this paper, we illustrate the causes of this issue, reviewing relevant literature from model-based reinforcement learning, and argue for solutions. By solving objective mismatch in RLHF, the ML models of the future will be more precisely aligned to user instructions for both safety and helpfulness.

The rapid development of Large Language Models (LLMs) has substantially expanded the range of tasks they can address. In the field of Natural Language Processing (NLP), researchers have shifted their focus from conventional NLP tasks (e.g., sequence tagging and parsing) towards tasks that revolve around aligning with human needs (e.g., brainstorming and email writing). This shift in task distribution imposes new requirements on evaluating these aligned models regarding generality (i.e., assessing performance across diverse scenarios), flexibility (i.e., examining under different protocols), and interpretability (i.e., scrutinizing models with explanations). In this paper, we propose a generative judge with 13B parameters, Auto-J, designed to address these challenges. Our model is trained on user queries and LLM-generated responses under massive real-world scenarios and accommodates diverse evaluation protocols (e.g., pairwise response comparison and single-response evaluation) with well-structured natural language critiques. To demonstrate the efficacy of our approach, we construct a new testbed covering 58 different scenarios. Experimentally, Auto-J outperforms a series of strong competitors, including both open-source and closed-source models, by a large margin. We also provide detailed analysis and case studies to further reveal the potential of our method and make a variety of resources public at https://github.com/GAIR-NLP/auto-j.

Many cognitive approaches to well-being, such as recognizing and reframing unhelpful thoughts, have received considerable empirical support over the past decades, yet still lack truly widespread adoption in self-help format. A barrier to that adoption is a lack of adequately specific and diverse dedicated practice material. This work examines whether current language models can be leveraged to both produce a virtually unlimited quantity of practice material illustrating standard unhelpful thought patterns matching specific given contexts, and generate suitable positive reframing proposals. We propose PATTERNREFRAME, a novel dataset of about 10k examples of thoughts containing unhelpful thought patterns conditioned on a given persona, accompanied by about 27k positive reframes. By using this dataset to train and/or evaluate current models, we show that existing models can already be powerful tools to help generate an abundance of tailored practice material and hypotheses, with no or minimal additional model training required.

The impressive performance of large language models (LLMs) has attracted considerable attention from the academic and industrial communities. Besides how to construct and train LLMs, how to effectively evaluate and compare the capacity of LLMs has also been well recognized as an important yet difficult problem. Existing paradigms rely on either human annotators or model-based evaluators to evaluate the performance of LLMs on different tasks. However, these paradigms often suffer from high cost, low generalizability, and inherited biases in practice, which make them incapable of supporting the sustainable development of LLMs in long term. In order to address these issues, inspired by the peer review systems widely used in academic publication process, we propose a novel framework that can automatically evaluate LLMs through a peer-review process. Specifically, for the evaluation of a specific task, we first construct a small qualification exam to select "reviewers" from a couple of powerful LLMs. Then, to actually evaluate the "submissions" written by different candidate LLMs, i.e., the evaluatees, we use the reviewer LLMs to rate or compare the submissions. The final ranking of evaluatee LLMs is generated based on the results provided by all reviewers. We conducted extensive experiments on text summarization tasks with eleven LLMs including GPT-4. The results demonstrate the existence of biasness when evaluating using a single LLM. Also, our PRE model outperforms all the baselines, illustrating the effectiveness of the peer review mechanism.

The emergence of pretrained models has significantly impacted from Natural Language Processing (NLP) and Computer Vision to relational datasets. Traditionally, these models are assessed through fine-tuned downstream tasks. However, this raises the question of how to evaluate these models more efficiently and more effectively. In this study, we explore a novel approach where we leverage the meta features associated with each entity as a source of worldly knowledge and employ entity representations from the models. We propose using the consistency between these representations and the meta features as a metric for evaluating pretrained models. Our method's effectiveness is demonstrated across various domains, including models with relational datasets, large language models and images models.

As Language Model (LM) capabilities advance, evaluating and supervising them at scale is getting harder for humans. There is hope that other language models can automate both these tasks, which we refer to as "AI Oversight". We study how model similarity affects both aspects of AI oversight by proposing a probabilistic metric for LM similarity based on overlap in model mistakes. Using this metric, we first show that LLM-as-a-judge scores favor models similar to the judge, generalizing recent self-preference results. Then, we study training on LM annotations, and find complementary knowledge between the weak supervisor and strong student model plays a crucial role in gains from "weak-to-strong generalization". As model capabilities increase, it becomes harder to find their mistakes, and we might defer more to AI oversight. However, we observe a concerning trend -- model mistakes are becoming more similar with increasing capabilities, pointing to risks from correlated failures. Our work underscores the importance of reporting and correcting for model similarity, especially in the emerging paradigm of AI oversight.

Large Language Models (LLMs) exhibit impressive capabilities but require careful alignment with human preferences. Traditional training-time methods finetune LLMs using human preference datasets but incur significant training costs and require repeated training to handle diverse user preferences. Test-time alignment methods address this by using reward models (RMs) to guide frozen LLMs without retraining. However, existing test-time approaches rely on trajectory-level RMs which are designed to evaluate complete responses, making them unsuitable for autoregressive text generation that requires computing next-token rewards from partial responses. To address this, we introduce GenARM, a test-time alignment approach that leverages the Autoregressive Reward Model--a novel reward parametrization designed to predict next-token rewards for efficient and effective autoregressive generation. Theoretically, we demonstrate that this parametrization can provably guide frozen LLMs toward any distribution achievable by traditional RMs within the KL-regularized reinforcement learning framework. Experimental results show that GenARM significantly outperforms prior test-time alignment baselines and matches the performance of training-time methods. Additionally, GenARM enables efficient weak-to-strong guidance, aligning larger LLMs with smaller RMs without the high costs of training larger models. Furthermore, GenARM supports multi-objective alignment, allowing real-time trade-offs between preference dimensions and catering to diverse user preferences without retraining.

As language models (LMs) advance, interest is growing in applying them to high-stakes societal decisions, such as determining financing or housing eligibility. However, their potential for discrimination in such contexts raises ethical concerns, motivating the need for better methods to evaluate these risks. We present a method for proactively evaluating the potential discriminatory impact of LMs in a wide range of use cases, including hypothetical use cases where they have not yet been deployed. Specifically, we use an LM to generate a wide array of potential prompts that decision-makers may input into an LM, spanning 70 diverse decision scenarios across society, and systematically vary the demographic information in each prompt. Applying this methodology reveals patterns of both positive and negative discrimination in the Claude 2.0 model in select settings when no interventions are applied. While we do not endorse or permit the use of language models to make automated decisions for the high-risk use cases we study, we demonstrate techniques to significantly decrease both positive and negative discrimination through careful prompt engineering, providing pathways toward safer deployment in use cases where they may be appropriate. Our work enables developers and policymakers to anticipate, measure, and address discrimination as language model capabilities and applications continue to expand. We release our dataset and prompts at https://huggingface.co/datasets/Anthropic/discrim-eval

Large language models (LLMs) currently dominate the field of natural language processing (NLP), representing the state-of-the-art across a diverse array of tasks. Developing a model of this nature, from training to inference, requires making numerous decisions which define a combinatorial search problem. For example, selecting the optimal pre-trained LLM, prompt, or hyperparameters to attain the best performance for a task often requires evaluating multiple candidates on an entire test set. This exhaustive evaluation can be time-consuming and costly, as both inference and metric computation with LLMs are resource-intensive. In this paper, we address the challenge of identifying the best method within a limited budget for evaluating methods on test examples. By leveraging the well-studied multi-armed bandit framework, which sequentially selects the next method-example pair to evaluate, our approach, combining multi-armed bandit algorithms with low-rank factorization, significantly reduces the required resources. Experiments show that our algorithms can identify the top-performing method using only 5-15\% of the typically needed resources, resulting in an 85-95\% reduction in cost.

Clinical trials are fundamental in developing new drugs, medical devices, and treatments. However, they are often time-consuming and have low success rates. Although there have been initial attempts to create large language models (LLMs) for clinical trial design and patient-trial matching, these models remain task-specific and not adaptable to diverse clinical trial tasks. To address this challenge, we propose a clinical trial foundation model named Panacea, designed to handle multiple tasks, including trial search, trial summarization, trial design, and patient-trial matching. We also assemble a large-scale dataset, named TrialAlign, of 793,279 trial documents and 1,113,207 trial-related scientific papers, to infuse clinical knowledge into the model by pre-training. We further curate TrialInstruct, which has 200,866 of instruction data for fine-tuning. These resources enable Panacea to be widely applicable for a range of clinical trial tasks based on user requirements. We evaluated Panacea on a new benchmark, named TrialPanorama, which covers eight clinical trial tasks. Our method performed the best on seven of the eight tasks compared to six cutting-edge generic or medicine-specific LLMs. Specifically, Panacea showed great potential to collaborate with human experts in crafting the design of eligibility criteria, study arms, and outcome measures, in multi-round conversations. In addition, Panacea achieved 14.42% improvement in patient-trial matching, 41.78% to 52.02% improvement in trial search, and consistently ranked at the top for five aspects of trial summarization. Our approach demonstrates the effectiveness of Panacea in clinical trials and establishes a comprehensive resource, including training data, model, and benchmark, for developing clinical trial foundation models, paving the path for AI-based clinical trial development.

Due to the cumbersome nature of human evaluation and limitations of code-based evaluation, Large Language Models (LLMs) are increasingly being used to assist humans in evaluating LLM outputs. Yet LLM-generated evaluators simply inherit all the problems of the LLMs they evaluate, requiring further human validation. We present a mixed-initiative approach to ``validate the validators'' -- aligning LLM-generated evaluation functions (be it prompts or code) with human requirements. Our interface, EvalGen, provides automated assistance to users in generating evaluation criteria and implementing assertions. While generating candidate implementations (Python functions, LLM grader prompts), EvalGen asks humans to grade a subset of LLM outputs; this feedback is used to select implementations that better align with user grades. A qualitative study finds overall support for EvalGen but underscores the subjectivity and iterative process of alignment. In particular, we identify a phenomenon we dub criteria drift: users need criteria to grade outputs, but grading outputs helps users define criteria. What is more, some criteria appears dependent on the specific LLM outputs observed (rather than independent criteria that can be defined a priori), raising serious questions for approaches that assume the independence of evaluation from observation of model outputs. We present our interface and implementation details, a comparison of our algorithm with a baseline approach, and implications for the design of future LLM evaluation assistants.

Long-context models (LCMs) have made remarkable strides in recent years, offering users great convenience for handling tasks that involve long context, such as document summarization. As the community increasingly prioritizes the faithfulness of generated results, merely ensuring the accuracy of LCM outputs is insufficient, as it is quite challenging for humans to verify the results from the extremely lengthy context. Yet, although some efforts have been made to assess whether LCMs respond truly based on the context, these works either are limited to specific tasks or heavily rely on external evaluation resources like GPT-4.In this work, we introduce L-CiteEval, a comprehensive multi-task benchmark for long-context understanding with citations, aiming to evaluate both the understanding capability and faithfulness of LCMs. L-CiteEval covers 11 tasks from diverse domains, spanning context lengths from 8K to 48K, and provides a fully automated evaluation suite. Through testing with 11 cutting-edge closed-source and open-source LCMs, we find that although these models show minor differences in their generated results, open-source models substantially trail behind their closed-source counterparts in terms of citation accuracy and recall. This suggests that current open-source LCMs are prone to responding based on their inherent knowledge rather than the given context, posing a significant risk to the user experience in practical applications. We also evaluate the RAG approach and observe that RAG can significantly improve the faithfulness of LCMs, albeit with a slight decrease in the generation quality. Furthermore, we discover a correlation between the attention mechanisms of LCMs and the citation generation process.

The development of generative language models that can create long and coherent textual outputs via autoregression has lead to a proliferation of uses and a corresponding sweep of analyses as researches work to determine the limitations of this new paradigm. Unlike humans, these 'Large Language Models' (LLMs) are highly sensitive to small changes in their inputs, leading to unwanted inconsistency in their behavior. One problematic inconsistency when LLMs are used to answer multiple-choice questions or analyze multiple inputs is order dependency: the output of an LLM can (and often does) change significantly when sub-sequences are swapped, despite both orderings being semantically identical. In this paper we present , a technique that guarantees the output of an LLM will not have order dependence on a specified set of sub-sequences. We show that this method provably eliminates order dependency, and that it can be applied to any transformer-based LLM to enable text generation that is unaffected by re-orderings. Delving into the implications of our method, we show that, despite our inputs being out of distribution, the impact on expected accuracy is small, where the expectation is over the order of uniformly chosen shuffling of the candidate responses, and usually significantly less in practice. Thus, can be used as a 'dropped-in' method on fully trained models. Finally, we discuss how our method's success suggests that other strong guarantees can be obtained on LLM performance via modifying the input representations.

Large Language Model (LLM) leaderboards based on benchmark rankings are regularly used to guide practitioners in model selection. Often, the published leaderboard rankings are taken at face value - we show this is a (potentially costly) mistake. Under existing leaderboards, the relative performance of LLMs is highly sensitive to (often minute) details. We show that for popular multiple choice question benchmarks (e.g. MMLU) minor perturbations to the benchmark, such as changing the order of choices or the method of answer selection, result in changes in rankings up to 8 positions. We explain this phenomenon by conducting systematic experiments over three broad categories of benchmark perturbations and identifying the sources of this behavior. Our analysis results in several best-practice recommendations, including the advantage of a hybrid scoring method for answer selection. Our study highlights the dangers of relying on simple benchmark evaluations and charts the path for more robust evaluation schemes on the existing benchmarks.

Multimodal Large Language Models (MLLMs) are gaining increasing popularity in both academia and industry due to their remarkable performance in various applications such as visual question answering, visual perception, understanding, and reasoning. Over the past few years, significant efforts have been made to examine MLLMs from multiple perspectives. This paper presents a comprehensive review of 180 benchmarks and evaluation for MLLMs, focusing on (1)perception and understanding, (2)cognition and reasoning, (3)specific domains, (4)key capabilities, and (5)other modalities. Finally, we discuss the limitations of the current evaluation methods for MLLMs and explore promising future directions. Our key argument is that evaluation should be regarded as a crucial discipline to better support the development of MLLMs. For more details, please visit our GitHub repository: https://github.com/swordlidev/Evaluation-Multimodal-LLMs-Survey.

As large language models (LLMs) are adopted as a fundamental component of language technologies, it is crucial to accurately characterize their performance. Because choices in prompt design can strongly influence model behavior, this design process is critical in effectively using any modern pre-trained generative language model. In this work, we focus on LLM sensitivity to a quintessential class of meaning-preserving design choices: prompt formatting. We find that several widely used open-source LLMs are extremely sensitive to subtle changes in prompt formatting in few-shot settings, with performance differences of up to 76 accuracy points when evaluated using LLaMA-2-13B. Sensitivity remains even when increasing model size, the number of few-shot examples, or performing instruction tuning. Our analysis suggests that work evaluating LLMs with prompting-based methods would benefit from reporting a range of performance across plausible prompt formats, instead of the currently-standard practice of reporting performance on a single format. We also show that format performance only weakly correlates between models, which puts into question the methodological validity of comparing models with an arbitrarily chosen, fixed prompt format. To facilitate systematic analysis we propose FormatSpread, an algorithm that rapidly evaluates a sampled set of plausible prompt formats for a given task, and reports the interval of expected performance without accessing model weights. Furthermore, we present a suite of analyses that characterize the nature of this sensitivity, including exploring the influence of particular atomic perturbations and the internal representation of particular formats.

This paper presents synthetic Preference Optimization (PO) datasets generated using multi-agent workflows and evaluates the effectiveness and potential of these workflows in the dataset generation process. PO dataset generation requires two modules: (1) response evaluation, and (2) response generation. In the response evaluation module, the responses from Large Language Models (LLMs) are evaluated and ranked - a task typically carried out by human annotators that we automate using LLMs. We assess the response evaluation module in a 2 step process. In step 1, we assess LLMs as evaluators using three distinct prompting strategies. In step 2, we apply the winning prompting strategy to compare the performance of LLM-as-a-Judge, LLMs-as-a-Jury, and LLM Debate. In each step, we use inter-rater agreement using Cohen's Kappa between human annotators and LLMs. For the response generation module, we compare different configurations for the LLM Feedback Loop using the identified LLM evaluator configuration. We use the win rate (the fraction of times a generation framework is selected as the best by an LLM evaluator) to determine the best multi-agent configuration for generation. After identifying the best configurations for both modules, we use models from the GPT, Gemma, and Llama families to generate our PO datasets using the above pipeline. We generate two types of PO datasets, one to improve the generation capabilities of individual LLM and the other to improve the multi-agent workflow. Our evaluation shows that GPT-4o-as-a-Judge is more consistent across datasets when the candidate responses do not include responses from the GPT family. Additionally, we find that the LLM Feedback Loop, with Llama as the generator and Gemma as the reviewer, achieves a notable 71.8% and 73.8% win rate over single-agent Llama and Gemma, respectively.

Large language models achieve high performance on many but not all downstream tasks. The interaction between pretraining data and task data is commonly assumed to determine this variance: a task with data that is more similar to a model's pretraining data is assumed to be easier for that model. We test whether distributional and example-specific similarity measures (embedding-, token- and model-based) correlate with language model performance through a large-scale comparison of the Pile and C4 pretraining datasets with downstream benchmarks. Similarity correlates with performance for multilingual datasets, but in other benchmarks, we surprisingly find that similarity metrics are not correlated with accuracy or even each other. This suggests that the relationship between pretraining data and downstream tasks is more complex than often assumed.

This paper introduces Patched Round-Trip Correctness (Patched RTC), a novel evaluation technique for Large Language Models (LLMs) applied to diverse software development tasks, particularly focusing on "outer loop" activities such as bug fixing, code review, and documentation updates. Patched RTC extends the original Round-Trip Correctness method to work with any LLM and downstream task, offering a self-evaluating framework that measures consistency and robustness of model responses without human intervention. The study demonstrates a correlation between Patched RTC scores and task-specific accuracy metrics, presenting it as an alternative to the LLM-as-Judge paradigm for open-domain task evaluation. We implement Patched RTC in an open-source framework called patchwork, allowing for transparent evaluation during inference across various patchflows. Experiments comparing GPT-3.5 and GPT-4 models across different software development tasks reveal that Patched RTC effectively distinguishes model performance and task difficulty. The paper also explores the impact of consistency prompts on improving model accuracy, suggesting that Patched RTC can guide prompt refinement and model selection for complex software development workflows.

We introduce LLaVA-Critic, the first open-source large multimodal model (LMM) designed as a generalist evaluator to assess performance across a wide range of multimodal tasks. LLaVA-Critic is trained using a high-quality critic instruction-following dataset that incorporates diverse evaluation criteria and scenarios. Our experiments demonstrate the model's effectiveness in two key areas: (1) LMM-as-a-Judge, where LLaVA-Critic provides reliable evaluation scores, performing on par with or surpassing GPT models on multiple evaluation benchmarks; and (2) Preference Learning, where it generates reward signals for preference learning, enhancing model alignment capabilities. This work underscores the potential of open-source LMMs in self-critique and evaluation, setting the stage for future research into scalable, superhuman alignment feedback mechanisms for LMMs.

While finetuning language models from pairwise preferences has proven remarkably effective, the underspecified nature of natural language presents critical challenges. Direct preference feedback is uninterpretable, difficult to provide where multidimensional criteria may apply, and often inconsistent, either because it is based on incomplete instructions or provided by diverse principals. To address these challenges, we consider the two-step preference modeling procedure that first resolves the under-specification by selecting a context, and then evaluates preference with respect to the chosen context. We decompose reward modeling error according to these two steps, which suggests that supervising context in addition to context-specific preference may be a viable approach to aligning models with diverse human preferences. For this to work, the ability of models to evaluate context-specific preference is critical. To this end, we contribute context-conditioned preference datasets and accompanying experiments that investigate the ability of language models to evaluate context-specific preference. We use our datasets to (1) show that existing preference models benefit from, but fail to fully consider, added context, (2) finetune a context-aware reward model with context-specific performance exceeding that of GPT-4 and Llama 3 70B on tested datasets, and (3) investigate the value of context-aware preference modeling.

Large language models (LMs) have been shown to memorize parts of their training data, and when prompted appropriately, they will emit the memorized training data verbatim. This is undesirable because memorization violates privacy (exposing user data), degrades utility (repeated easy-to-memorize text is often low quality), and hurts fairness (some texts are memorized over others). We describe three log-linear relationships that quantify the degree to which LMs emit memorized training data. Memorization significantly grows as we increase (1) the capacity of a model, (2) the number of times an example has been duplicated, and (3) the number of tokens of context used to prompt the model. Surprisingly, we find the situation becomes more complicated when generalizing these results across model families. On the whole, we find that memorization in LMs is more prevalent than previously believed and will likely get worse as models continues to scale, at least without active mitigations.

Language model pretraining has led to significant performance gains but careful comparison between different approaches is challenging. Training is computationally expensive, often done on private datasets of different sizes, and, as we will show, hyperparameter choices have significant impact on the final results. We present a replication study of BERT pretraining (Devlin et al., 2019) that carefully measures the impact of many key hyperparameters and training data size. We find that BERT was significantly undertrained, and can match or exceed the performance of every model published after it. Our best model achieves state-of-the-art results on GLUE, RACE and SQuAD. These results highlight the importance of previously overlooked design choices, and raise questions about the source of recently reported improvements. We release our models and code.

Recently, there has been a growing trend of utilizing Large Language Model (LLM) to evaluate the quality of other LLMs. Many studies have employed proprietary close-source models, especially GPT4, as the evaluator. Alternatively, other works have fine-tuned judge models based on open-source LLMs as the evaluator. In this study, we conduct an empirical study of different judge models on their evaluation capability. Our findings indicate that although the fine-tuned judge models achieve high accuracy on in-domain test sets, even surpassing GPT4, they are inherently task-specific classifiers, and their generalizability and fairness severely underperform GPT4.

Researchers are usually interested in examining the impact of covariates when separating heterogeneous samples into latent classes that are more homogeneous. The majority of theoretical and empirical studies with such aims have focused on identifying covariates as predictors of class membership in the structural equation modeling framework. In other words, the covariates only indirectly affect the sample heterogeneity. However, the covariates' influence on between-individual differences can also be direct. This article presents a mixture model that investigates covariates to explain within-cluster and between-cluster heterogeneity simultaneously, known as a mixture-of-experts (MoE) model. This study aims to extend the MoE framework to investigate heterogeneity in nonlinear trajectories: to identify latent classes, covariates as predictors to clusters, and covariates that explain within-cluster differences in change patterns over time. Our simulation studies demonstrate that the proposed model generally estimates the parameters unbiasedly, precisely and exhibits appropriate empirical coverage for a nominal 95% confidence interval. This study also proposes implementing structural equation model forests to shrink the covariate space of the proposed mixture model. We illustrate how to select covariates and construct the proposed model with longitudinal mathematics achievement data. Additionally, we demonstrate that the proposed mixture model can be further extended in the structural equation modeling framework by allowing the covariates that have direct effects to be time-varying.

Given the recent introduction of multiple language models and the ongoing demand for improved Natural Language Processing tasks, particularly summarization, this work provides a comprehensive benchmarking of 20 recent language models, focusing on smaller ones for the news summarization task. In this work, we systematically test the capabilities and effectiveness of these models in summarizing news article texts which are written in different styles and presented in three distinct datasets. Specifically, we focus in this study on zero-shot and few-shot learning settings and we apply a robust evaluation methodology that combines different evaluation concepts including automatic metrics, human evaluation, and LLM-as-a-judge. Interestingly, including demonstration examples in the few-shot learning setting did not enhance models' performance and, in some cases, even led to worse quality of the generated summaries. This issue arises mainly due to the poor quality of the gold summaries that have been used as reference summaries, which negatively impacts the models' performance. Furthermore, our study's results highlight the exceptional performance of GPT-3.5-Turbo and GPT-4, which generally dominate due to their advanced capabilities. However, among the public models evaluated, certain models such as Qwen1.5-7B, SOLAR-10.7B-Instruct-v1.0, Meta-Llama-3-8B and Zephyr-7B-Beta demonstrated promising results. These models showed significant potential, positioning them as competitive alternatives to large models for the task of news summarization.

Evaluation is the baton for the development of large language models. Current evaluations typically employ a single-item assessment paradigm for each atomic test objective, which struggles to discern whether a model genuinely possesses the required capabilities or merely memorizes/guesses the answers to specific questions. To this end, we propose a novel evaluation framework referred to as StructEval. Starting from an atomic test objective, StructEval deepens and broadens the evaluation by conducting a structured assessment across multiple cognitive levels and critical concepts, and therefore offers a comprehensive, robust and consistent evaluation for LLMs. Experiments on three widely-used benchmarks demonstrate that StructEval serves as a reliable tool for resisting the risk of data contamination and reducing the interference of potential biases, thereby providing more reliable and consistent conclusions regarding model capabilities. Our framework also sheds light on the design of future principled and trustworthy LLM evaluation protocols.

We investigate the predictability of large language model (LLM) capabilities: given records of past experiments using different model families, numbers of parameters, tasks, and numbers of in-context examples, can we accurately predict LLM performance on new experiment configurations? Answering this question has practical implications for LLM users (e.g., deciding which models to try), developers (e.g., prioritizing evaluation on representative tasks), and the research community (e.g., identifying hard-to-predict capabilities that warrant further investigation). We study the performance prediction problem on experiment records from BIG-bench. On a random train-test split, an MLP-based predictor achieves an R^2 score greater than 95%, indicating the presence of learnable patterns within the experiment records. We then formulate the problem of searching for "small-bench," an informative subset of BIG-bench tasks from which the performance on the full set can be maximally recovered. We find a subset as informative as BIG-bench Hard for evaluating new model families, while being 3times smaller. Additionally, we find competitive subsets by clustering task representations learned by our MLP-based predictor and selecting tasks close to cluster centroids, highlighting the importance of task diversity in constructing "small-bench."

Large datasets have become commonplace in NLP research. However, the increased emphasis on data quantity has made it challenging to assess the quality of data. We introduce Data Maps---a model-based tool to characterize and diagnose datasets. We leverage a largely ignored source of information: the behavior of the model on individual instances during training (training dynamics) for building data maps. This yields two intuitive measures for each example---the model's confidence in the true class, and the variability of this confidence across epochs---obtained in a single run of training. Experiments across four datasets show that these model-dependent measures reveal three distinct regions in the data map, each with pronounced characteristics. First, our data maps show the presence of "ambiguous" regions with respect to the model, which contribute the most towards out-of-distribution generalization. Second, the most populous regions in the data are "easy to learn" for the model, and play an important role in model optimization. Finally, data maps uncover a region with instances that the model finds "hard to learn"; these often correspond to labeling errors. Our results indicate that a shift in focus from quantity to quality of data could lead to robust models and improved out-of-distribution generalization.

We test the robustness of debate as a method of scalable oversight by training models to debate with data generated via self-play. In a long-context reading comprehension task, we find that language model based evaluators answer questions more accurately when judging models optimized to win debates. By contrast, we find no such relationship for consultancy models trained to persuade a judge without an opposing debater present. In quantitative and qualitative comparisons between our debate models and novel consultancy baselines, we find evidence that debate training encourages stronger and more informative arguments, showing promise that it can help provide high-quality supervision for tasks that are difficult to directly evaluate.

To answer a question, language models often need to integrate prior knowledge learned during pretraining and new information presented in context. We hypothesize that models perform this integration in a predictable way across different questions and contexts: models will rely more on prior knowledge for questions about entities (e.g., persons, places, etc.) that they are more familiar with due to higher exposure in the training corpus, and be more easily persuaded by some contexts than others. To formalize this problem, we propose two mutual information-based metrics to measure a model's dependency on a context and on its prior about an entity: first, the persuasion score of a given context represents how much a model depends on the context in its decision, and second, the susceptibility score of a given entity represents how much the model can be swayed away from its original answer distribution about an entity. Following well-established measurement modeling methods, we empirically test for the validity and reliability of these metrics. Finally, we explore and find a relationship between the scores and the model's expected familiarity with an entity, and provide two use cases to illustrate their benefits.

We present the call for papers for the BabyLM Challenge: Sample-efficient pretraining on a developmentally plausible corpus. This shared task is intended for participants with an interest in small scale language modeling, human language acquisition, low-resource NLP, and cognitive modeling. In partnership with CoNLL and CMCL, we provide a platform for approaches to pretraining with a limited-size corpus sourced from data inspired by the input to children. The task has three tracks, two of which restrict the training data to pre-released datasets of 10M and 100M words and are dedicated to explorations of approaches such as architectural variations, self-supervised objectives, or curriculum learning. The final track only restricts the amount of text used, allowing innovation in the choice of the data, its domain, and even its modality (i.e., data from sources other than text is welcome). We will release a shared evaluation pipeline which scores models on a variety of benchmarks and tasks, including targeted syntactic evaluations and natural language understanding.

Large Language Models (LLMs) are trained on vast amounts of data, most of which is automatically scraped from the internet. This data includes encyclopedic documents that harbor a vast amount of general knowledge (e.g., Wikipedia) but also potentially overlap with benchmark datasets used for evaluating LLMs. Consequently, evaluating models on test splits that might have leaked into the training set is prone to misleading conclusions. To foster sound evaluation of language models, we introduce a new test dataset named RepLiQA, suited for question-answering and topic retrieval tasks. RepLiQA is a collection of five splits of test sets, four of which have not been released to the internet or exposed to LLM APIs prior to this publication. Each sample in RepLiQA comprises (1) a reference document crafted by a human annotator and depicting an imaginary scenario (e.g., a news article) absent from the internet; (2) a question about the document's topic; (3) a ground-truth answer derived directly from the information in the document; and (4) the paragraph extracted from the reference document containing the answer. As such, accurate answers can only be generated if a model can find relevant content within the provided document. We run a large-scale benchmark comprising several state-of-the-art LLMs to uncover differences in performance across models of various types and sizes in a context-conditional language modeling setting. Released splits of RepLiQA can be found here: https://huggingface.co/datasets/ServiceNow/repliqa.

Leveraging Large Language Models (LLMs) as judges for evaluating the performance of LLMs has recently garnered attention. Nonetheless, this type of approach concurrently introduces potential biases from LLMs, raising concerns about the reliability of the evaluation results. To mitigate this issue, we propose and study two versions of many-shot in-context prompts, Reinforced and Unsupervised ICL, for helping GPT-4o-as-a-Judge in single answer grading. The former uses in-context examples with model-generated rationales, and the latter without. Based on the designed prompts, we investigate the impact of scaling the number of in-context examples on the agreement and quality of the evaluation. Furthermore, we first reveal the symbol bias in GPT-4o-as-a-Judge for pairwise comparison and then propose a simple yet effective approach to mitigate it. Experimental results show that advanced long-context LLMs, such as GPT-4o, perform better in the many-shot regime than in the zero-shot regime. Meanwhile, the experimental results further verify the effectiveness of the symbol bias mitigation approach.

Large language models (LLMs) hold immense promise to serve complex health information needs but also have the potential to introduce harm and exacerbate health disparities. Reliably evaluating equity-related model failures is a critical step toward developing systems that promote health equity. In this work, we present resources and methodologies for surfacing biases with potential to precipitate equity-related harms in long-form, LLM-generated answers to medical questions and then conduct an empirical case study with Med-PaLM 2, resulting in the largest human evaluation study in this area to date. Our contributions include a multifactorial framework for human assessment of LLM-generated answers for biases, and EquityMedQA, a collection of seven newly-released datasets comprising both manually-curated and LLM-generated questions enriched for adversarial queries. Both our human assessment framework and dataset design process are grounded in an iterative participatory approach and review of possible biases in Med-PaLM 2 answers to adversarial queries. Through our empirical study, we find that the use of a collection of datasets curated through a variety of methodologies, coupled with a thorough evaluation protocol that leverages multiple assessment rubric designs and diverse rater groups, surfaces biases that may be missed via narrower evaluation approaches. Our experience underscores the importance of using diverse assessment methodologies and involving raters of varying backgrounds and expertise. We emphasize that while our framework can identify specific forms of bias, it is not sufficient to holistically assess whether the deployment of an AI system promotes equitable health outcomes. We hope the broader community leverages and builds on these tools and methods towards realizing a shared goal of LLMs that promote accessible and equitable healthcare for all.

Large Language Models (LLMs) have achieved remarkable results in the machine translation evaluation task, yet there remains a gap in knowledge regarding how they utilize the provided data to conduct evaluations. This study aims to explore how LLMs leverage source and reference information in evaluating translations, with the ultimate goal of better understanding the working mechanism of LLMs. To this end, we design the controlled experiments across various input modes and model types, and employ both coarse-grained and fine-grained prompts to discern the utility of source versus reference information. Surprisingly, we find that reference information significantly enhances the evaluation accuracy, while source information sometimes is counterproductive, indicating a lack of cross-lingual capability when using LLMs to evaluate translations. We further conduct a meta-evaluation for translation error detection of LLMs, observing a similar phenomenon. These findings also suggest a potential research direction for LLMs that fully exploits the cross-lingual capability of LLMs to achieve better performance in machine translation evaluation tasks.

LLM-as-a-Judge and reward models are widely used alternatives of multiple-choice questions or human annotators for large language model (LLM) evaluation. Their efficacy shines in evaluating long-form responses, serving a critical role as evaluators of leaderboards and as proxies to align LLMs via reinforcement learning. However, despite their popularity, their effectiveness outside of English remains largely unexplored. In this paper, we conduct a comprehensive analysis on automated evaluators, reporting key findings on their behavior in a non-English environment. First, we discover that English evaluation capabilities significantly influence language-specific capabilities, often more than the language proficiency itself, enabling evaluators trained in English to easily transfer their skills to other languages. Second, we identify critical shortcomings, where LLMs fail to detect and penalize errors, such as factual inaccuracies, cultural misrepresentations, and the presence of unwanted language. Finally, we release Kudge, the first non-English meta-evaluation dataset containing 5,012 human annotations in Korean.

Long-context large language models (LC LLMs) promise to increase reliability of LLMs in real-world tasks requiring processing and understanding of long input documents. However, this ability of LC LLMs to reliably utilize their growing context windows remains under investigation. In this work, we evaluate the performance of state-of-the-art GPT-4 suite of LC LLMs in solving a series of progressively challenging tasks, as a function of factors such as context length, task difficulty, and position of key information by creating a real world financial news dataset. Our findings indicate that LC LLMs exhibit brittleness at longer context lengths even for simple tasks, with performance deteriorating sharply as task complexity increases. At longer context lengths, these state-of-the-art models experience catastrophic failures in instruction following resulting in degenerate outputs. Our prompt ablations also reveal unfortunate continued sensitivity to both the placement of the task instruction in the context window as well as minor markdown formatting. Finally, we advocate for more rigorous evaluation of LC LLMs by employing holistic metrics such as F1 (rather than recall) and reporting confidence intervals, thereby ensuring robust and conclusive findings.

Evaluation of large language models (LLMs) has raised great concerns in the community due to the issue of data contamination. Existing work designed evaluation protocols using well-defined algorithms for specific tasks, which cannot be easily extended to diverse scenarios. Moreover, current evaluation benchmarks can only provide the overall benchmark results and cannot support a fine-grained and multifaceted analysis of LLMs' abilities. In this paper, we propose meta probing agents (MPA), a general dynamic evaluation protocol inspired by psychometrics to evaluate LLMs. MPA is the key component of DyVal 2, which naturally extends the previous DyVal~zhu2023dyval. MPA designs the probing and judging agents to automatically transform an original evaluation problem into a new one following psychometric theory on three basic cognitive abilities: language understanding, problem solving, and domain knowledge. These basic abilities are also dynamically configurable, allowing multifaceted analysis. We conducted extensive evaluations using MPA and found that most LLMs achieve poorer performance, indicating room for improvement. Our multifaceted analysis demonstrated the strong correlation between the basic abilities and an implicit Matthew effect on model size, i.e., larger models possess stronger correlations of the abilities. MPA can also be used as a data augmentation approach to enhance LLMs. Code is available at: https://github.com/microsoft/promptbench.

We introduce LLM-as-an-Interviewer, a novel paradigm for evaluating large language models (LLMs). This approach leverages multi-turn interactions where the LLM interviewer actively provides feedback on responses and poses follow-up questions to the evaluated LLM. At the start of the interview, the LLM interviewer dynamically modifies datasets to generate initial questions, mitigating data contamination. We apply the LLM-as-an-Interviewer framework to evaluate six models on the MATH and DepthQA tasks. Our results show that the framework effectively provides insights into LLM performance, including the quality of initial responses, adaptability to feedback, and ability to address follow-up queries like clarification or additional knowledge requests. The framework also addresses key limitations of conventional methods like LLM-as-a-Judge, including verbosity bias and inconsistency across runs. Finally, we propose the Interview Report, which aggregates insights from the interview process, providing examples and a comprehensive analysis of the LLM's strengths and weaknesses. This report offers a detailed snapshot of the model's real-world applicability. The code for our framework is publicly available at https://github.com/interview-eval/.

Multimodal Large Language Models (MLLMs) have advanced in integrating diverse modalities but frequently suffer from hallucination. A promising solution to mitigate this issue is to generate text with citations, providing a transparent chain for verification. However, existing work primarily focuses on generating citations for text-only content, overlooking the challenges and opportunities of multimodal contexts. To address this gap, we introduce MCiteBench, the first benchmark designed to evaluate and analyze the multimodal citation text generation ability of MLLMs. Our benchmark comprises data derived from academic papers and review-rebuttal interactions, featuring diverse information sources and multimodal content. We comprehensively evaluate models from multiple dimensions, including citation quality, source reliability, and answer accuracy. Through extensive experiments, we observe that MLLMs struggle with multimodal citation text generation. We also conduct deep analyses of models' performance, revealing that the bottleneck lies in attributing the correct sources rather than understanding the multimodal content.

The NLP community has long advocated for the construction of multi-annotator datasets to better capture the nuances of language interpretation, subjectivity, and ambiguity. This paper conducts a retrospective study to show how performance scores can vary when a dataset expands from a single annotation per instance to multiple annotations. We propose a novel multi-annotator simulation process to generate datasets with varying annotation budgets. We show that similar datasets with the same annotation budget can lead to varying performance gains. Our findings challenge the popular belief that models trained on multi-annotation examples always lead to better performance than models trained on single or few-annotation examples.

Great successes have been reported using Reinforcement Learning from Human Feedback (RLHF) to align large language models. Open-source preference datasets and reward models have enabled wider experimentation beyond generic chat settings, particularly to make systems more "helpful" for tasks like web question answering, summarization, and multi-turn dialogue. When optimizing for helpfulness, RLHF has been consistently observed to drive models to produce longer outputs. This paper demonstrates that optimizing for response length is a significant factor behind RLHF's reported improvements in these settings. First, we study the relationship between reward and length for reward models trained on three open-source preference datasets for helpfulness. Here, length correlates strongly with reward, and improvements in reward score are driven in large part by shifting the distribution over output lengths. We then explore interventions during both RL and reward model learning to see if we can achieve the same downstream improvements as RLHF without increasing length. While our interventions mitigate length increases, they aren't uniformly effective across settings. Furthermore, we find that even running RLHF with a reward based solely on length can reproduce most of the downstream improvements over the initial policy model, showing that reward models in these settings have a long way to go.