Home

Awesome

<div align="center"> <img src="imgs/logo-llmeval.png" alt="LLM evaluation" width="500"><br> A collection of papers and resources related to evaluations on large language models. </div> <br> <p align="center"> Yupeng Chang<sup>*1</sup>&nbsp&nbsp Xu Wang<sup>*1</sup>&nbsp&nbsp Jindong Wang<sup>#2</sup>&nbsp&nbsp Yuan Wu<sup>#1</sup>&nbsp&nbsp Kaijie Zhu<sup>3</sup>&nbsp&nbsp Hao Chen<sup>4</sup>&nbsp&nbsp Linyi Yang<sup>5</sup>&nbsp&nbsp Xiaoyuan Yi<sup>2</sup>&nbsp&nbsp Cunxiang Wang<sup>5</sup>&nbsp&nbsp Yidong Wang<sup>6</sup>&nbsp&nbsp Wei Ye<sup>6</sup>&nbsp&nbsp Yue Zhang<sup>5</sup>&nbsp&nbsp Yi Chang<sup>1</sup>&nbsp&nbsp Philip S. Yu<sup>7</sup>&nbsp&nbsp Qiang Yang<sup>8</sup>&nbsp&nbsp Xing Xie<sup>2</sup> </p> <p align="center"> <sup>1</sup> Jilin University, <sup>2</sup> Microsoft Research, <sup>3</sup> Institute of Automation, CAS <sup>4</sup> Carnegie Mellon University, <sup>5</sup> Westlake University, <sup>6</sup> Peking University, <sup>7</sup> University of Illinois, <sup>8</sup> Hong Kong University of Science and Technology<br> (*: Co-first authors, #: Co-corresponding authors) </p>

Papers and resources for LLMs evaluation

The papers are organized according to our survey: A Survey on Evaluation of Large Language Models.

NOTE: As we cannot update the arXiv paper in real time, please refer to this repo for the latest updates and the paper may be updated later. We also welcome any pull request or issues to help us make this survey perfect. Your contributions will be acknowledged in <a href="#acknowledgements">acknowledgements</a>.

Related projects:

<details> <summary>Table of Contents</summary> <ol> <li><a href="#news-and-updates">News and Updates</a></li> <li> <a href="#what-to-evaluate">What to evaluate</a> <ul> <li><a href="#natural-language-processing">Natural language processing</a></li> <li><a href="#robustness-ethics-biases-and-trustworthiness">Robustness, ethics, biases, and trustworthiness</a></li> <li><a href="#social-science">Social science</a></li> <li><a href="#natural-science-and-engineering">Natural science and engineering</a></li> <li><a href="#medical-applications">Medical applications</a></li> <li><a href="#agent-applications">Agent applications</a></li> <li><a href="#other-applications">Other applications</a></li> </ul> </li> <li><a href="where-to-evaluate">Where to evaluate</a></li> <li><a href="#Contributing">Contributing</a></li> <li><a href="#citation">Citation</a></li> <li><a href="#acknowledgements">Acknowledgments</a></li> </ol> </details>

News and updates

What to evaluate

Natural language processing

Natural language understanding

Sentiment analysis
  1. A Multitask, Multilingual, Multimodal Evaluation of ChatGPT on Reasoning, Hallucination, and Interactivity. Yejin Bang et al. arXiv 2023. [paper]
  2. A Systematic Study and Comprehensive Evaluation of ChatGPT on Benchmark Datasets, Laskar et al. ACL 2023 (Findings). [paper]
  3. Holistic Evaluation of Language Models. Percy Liang et al. arXiv 2022. [paper]
  4. Can ChatGPT forecast stock price movements? return predictability and large language models. Alejandro Lopez-Lira et al. SSRN 2023. [paper]
  5. Is ChatGPT a general-purpose natural language processing task solver? Chengwei Qin et al. arXiv 2023. [paper]
  6. Is ChatGPT a Good Sentiment Analyzer? A Preliminary Study. Zengzhi Wang et al. arXiv 2023. [paper]
  7. Sentiment analysis in the era of large language models: A reality check. Wenxuan Zhang et al. arXiv 2023. [paper]
Text classification
  1. Holistic evaluation of language models. Percy Liang et al. arXiv 2022. [paper]
  2. Leveraging large language models for topic classification in the domain of public affairs. Alejandro Peña et al. arXiv 2023. [paper]
  3. Large language models can rate news outlet credibility. Kai-Cheng Yang et al. arXiv 2023. [paper]
Natural language inference
  1. A Systematic Study and Comprehensive Evaluation of ChatGPT on Benchmark Datasets, Laskar et al. ACL 2023 (Findings). [paper]
  2. Can Large Language Models Infer and Disagree like Humans? Noah Lee et al. arXiv 2023. [paper]
  3. Is ChatGPT a general-purpose natural language processing task solver? Chengwei Qin et al. arXiv 2023. [paper]
Others
  1. Do LLMs Understand Social Knowledge? Evaluating the Sociability of Large Language Models with SocKET Benchmark. Minje Choi et al. arXiv 2023. [paper]
  2. The two word test: A semantic benchmark for large language models. Nicholas Riccardi et al. arXiv 2023. [paper]
  3. EvEval: A Comprehensive Evaluation of Event Semantics for Large Language Models. Zhengwei Tao et al. arXiv 2023. [paper]

Reasoning

  1. A Multitask, Multilingual, Multimodal Evaluation of ChatGPT on Reasoning, Hallucination, and Interactivity. Yejin Bang et al. arXiv 2023. [paper]
  2. ChatGPT is a knowledgeable but inexperienced solver: An investigation of commonsense problem in large language models. Ning Bian et al. arXiv 2023. [paper]
  3. Chain-of-Thought Hub: A continuous effort to measure large language models' reasoning performance. Yao Fu et al. arXiv 2023. [paper]
  4. A Systematic Study and Comprehensive Evaluation of ChatGPT on Benchmark Datasets, Laskar et al. ACL 2023 (Findings). [paper]
  5. Large Language Models Are Not Abstract Reasoners, Gaël Gendron et al. arXiv 2023. [paper]
  6. Can large language models reason about medical questions? Valentin Liévin et al. arXiv 2023. [paper]
  7. Evaluating the Logical Reasoning Ability of ChatGPT and GPT-4. Hanmeng Liu et al. arXiv 2023. [paper]
  8. Mathematical Capabilities of ChatGPT. Simon Frieder et al. arXiv 2023. [paper]
  9. Human-like problem-solving abilities in large language models using ChatGPT. Graziella Orrù et al. Front. Artif. Intell. 2023 [paper]
  10. Is ChatGPT a general-purpose natural language processing task solver? Chengwei Qin et al. arXiv 2023. [paper]
  11. Testing the general deductive reasoning capacity of large language models using OOD examples. Abulhair Saparov et al. arXiv 2023. [paper]
  12. MindGames: Targeting Theory of Mind in Large Language Models with Dynamic Epistemic Modal Logic. Damien Sileo et al. arXiv 2023. [paper]
  13. Reasoning or Reciting? Exploring the Capabilities and Limitations of Language Models Through Counterfactual Tasks. Zhaofeng Wu arXiv 2023. [paper]
  14. Are large language models really good logical reasoners? a comprehensive evaluation from deductive, inductive and abductive views. Fangzhi Xu et al. arXiv 2023. [paper]
  15. Efficiently Measuring the Cognitive Ability of LLMs: An Adaptive Testing Perspective. Yan Zhuang et al. arXiv 2023. [paper]
  16. Autoformalization with Large Language Models. Yuhuai Wu et al. NeurIPS 2022. [paper]
  17. Evaluating and Improving Tool-Augmented Computation-Intensive Math Reasoning. Beichen Zhang et al. arXiv 2023. [paper]
  18. StructGPT: A General Framework for Large Language Model to Reason over Structured Data. Jinhao Jiang et al. arXiv 2023. [paper]
  19. Unifying Large Language Models and Knowledge Graphs: A Roadmap. Shirui Pan et al. arXiv 2023. [paper]

Natural language generation

Summarization
  1. A Multitask, Multilingual, Multimodal Evaluation of ChatGPT on Reasoning, Hallucination, and Interactivity. Yejin Bang et al. arXiv 2023. [paper]
  2. A Systematic Study and Comprehensive Evaluation of ChatGPT on Benchmark Datasets, Laskar et al. ACL 2023 (Findings). [paper]
  3. Holistic Evaluation of Language Models. Percy Liang et al. arXiv 2022. [paper]
  4. ChatGPT vs Human-authored text: Insights into controllable text summarization and sentence style transfer. Dongqi Pu et al. arXiv 2023. [paper]
  5. Is ChatGPT a general-purpose natural language processing task solver? Chengwei Qin et al. arXiv 2023. [paper]
Dialogue
  1. A Multitask, Multilingual, Multimodal Evaluation of ChatGPT on Reasoning, Hallucination, and Interactivity. Yejin Bang et al. arXiv 2023. [paper]
  2. LLM-Eval: Unified Multi-Dimensional Automatic Evaluation for Open-Domain Conversations with Large Language Models. Yen-Ting Lin et al. arXiv 2023. [paper]
  3. Is ChatGPT a general-purpose natural language processing task solver? Chengwei Qin et al. arXiv 2023. [paper]
  4. LMSYS-Chat-1M: A Large-Scale Real-World LLM Conversation Dataset. Lianmin Zheng et al. arXiv 2023. [paper]
Translation
  1. A Multitask, Multilingual, Multimodal Evaluation of ChatGPT on Reasoning, Hallucination, and Interactivity. Yejin Bang et al. arXiv 2023. [paper]
  2. A Systematic Study and Comprehensive Evaluation of ChatGPT on Benchmark Datasets, Laskar et al. ACL 2023 (Findings). [paper]
  3. Translating Radiology Reports into Plain Language using ChatGPT and GPT-4 with Prompt Learning: Promising Results, Limitations, and Potential. Qing Lyu et al. arXiv 2023. [paper]
  4. Document-Level Machine Translation with Large Language Models. Longyue Wang et al. arXiv 2023. [paper]
  5. Case Study of Improving English-Arabic Translation Using the Transformer Model. Donia Gamal et al. ijicis 2023. [paper]
  6. Taqyim: Evaluating Arabic NLP Tasks Using ChatGPT Models. Zaid Alyafeai et al. arXiv 2023. [paper]
Question answering
  1. Benchmarking Foundation Models with Language-Model-as-an-Examiner. Yushi Bai et al. arXiv 2023. [paper]
  2. A Multitask, Multilingual, Multimodal Evaluation of ChatGPT on Reasoning, Hallucination, and Interactivity. Yejin Bang et al. arXiv 2023. [paper]
  3. ChatGPT is a knowledgeable but inexperienced solver: An investigation of commonsense problem in large language models. Ning Bian et al. arXiv 2023. [paper]
  4. A Systematic Study and Comprehensive Evaluation of ChatGPT on Benchmark Datasets, Laskar et al. ACL 2023 (Findings). [paper]
  5. Holistic Evaluation of Language Models. Percy Liang et al. arXiv 2022. [paper]
  6. Is ChatGPT a general-purpose natural language processing task solver? Chengwei Qin et al. arXiv 2023. [paper]
Others
  1. Exploring the use of large language models for reference-free text quality evaluation: A preliminary empirical study. Yi Chen et al. arXiv 2023. [paper]
  2. INSTRUCTEVAL: Towards Holistic Evaluation of Instruction-Tuned Large Language Models. Yew Ken Chia et al. arXiv 2023. [paper]
  3. ChatGPT vs Human-authored Text: Insights into Controllable Text Summarization and Sentence Style Transfer. Dongqi Pu et al. arXiv 2023. [paper]

Multilingual tasks

  1. Benchmarking Arabic AI with large language models. Ahmed Abdelali et al. arXiv 2023. [paper]
  2. MEGA: Multilingual Evaluation of Generative AI. Kabir Ahuja et al. arXiv 2023. [paper]
  3. A Multitask, Multilingual, Multimodal Evaluation of ChatGPT on Reasoning, Hallucination, and Interactivity. Yejin Bang et al. arXiv 2023. [paper]
  4. ChatGPT beyond English: Towards a comprehensive evaluation of large language models in multilingual learning. Viet Dac Lai et al. arXiv 2023. [paper]
  5. A Systematic Study and Comprehensive Evaluation of ChatGPT on Benchmark Datasets, Laskar et al. ACL 2023 (Findings). [paper]
  6. M3Exam: A Multilingual, Multimodal, Multilevel Benchmark for Examining Large Language Models. Wenxuan Zhang et al. arXiv 2023. [paper]
  7. Measuring Massive Multitask Chinese Understanding. Hui Zeng et al. arXiv 2023. [paper]
  8. CMMLU: Measuring massive multitask language understanding in Chinese. Haonan Li et al. arXiv 2023. [paper]

Factuality

  1. TrueTeacher: Learning Factual Consistency Evaluation with Large Language Models. Zorik Gekhman et al. arXiv 2023. [paper]

  2. TRUE: Re-evaluating Factual Consistency Evaluation. Or Honovich et al. arXiv 2022. [paper]

  3. SelfCheckGPT: Zero-Resource Black-Box Hallucination Detection for Generative Large Language Models. Potsawee Manakul et al. arXiv 2023. [paper]

  4. FActScore: Fine-grained Atomic Evaluation of Factual Precision in Long Form Text Generation. Sewon Min et al. arXiv 2023. [paper]

  5. Measuring and Modifying Factual Knowledge in Large Language Models. Pouya Pezeshkpour arXiv 2023. [paper]

  6. Evaluating Open-QA Evaluation. Cunxiang Wang arXiv 2023. [paper]

Robustness, ethics, biases, and trustworthiness

Robustness

  1. A Survey on Out-of-Distribution Evaluation of Neural NLP Models. Xinzhe Li et al. arXiv 2023. [paper]
  2. Mitigating Hallucination in Large Multi-Modal Models via Robust Instruction Tuning. Fuxiao Liu et al. arXiv 2023. [paper]
  3. Generalizing to Unseen Domains: A Survey on Domain Generalization. Jindong Wang et al. TKDE 2022. [paper]
  4. On the Robustness of ChatGPT: An Adversarial and Out-of-distribution Perspective. Jindong Wang et al. arXiv 2023. [paper]
  5. GLUE-X: Evaluating Natural Language Understanding Models from an Out-of-Distribution Generalization Perspective. Linyi Yang et al. arXiv 2022. [paper]
  6. On Evaluating Adversarial Robustness of Large Vision-Language Models. Yunqing Zhao et al. arXiv 2023. [paper]
  7. PromptBench: Towards Evaluating the Robustness of Large Language Models on Adversarial Prompts. Kaijie Zhu et al. arXiv 2023. [paper]
  8. On Robustness of Prompt-based Semantic Parsing with Large Pre-trained Language Model: An Empirical Study on Codex. Terry Yue Zhuo et al. arXiv 2023. [paper]

Ethics and bias

  1. Assessing Cross-Cultural Alignment between ChatGPT and Human Societies: An Empirical Study. Yong Cao et al. C3NLP@EACL 2023. [paper]
  2. Toxicity in ChatGPT: Analyzing persona-assigned language models. Ameet Deshpande et al. arXiv 2023. [paper]
  3. BOLD: Dataset and Metrics for Measuring Biases in Open-Ended Language Generation. Jwala Dhamala et al. FAccT 2021 [paper]
  4. Should ChatGPT be Biased? Challenges and Risks of Bias in Large Language Models. Emilio Ferrara arXiv 2023. [paper]
  5. RealToxicityPrompts: Evaluating Neural Toxic Degeneration in Language Models. Samuel Gehman et al. EMNLP 2020. [paper]
  6. The political ideology of conversational AI: Converging evidence on ChatGPT's pro-environmental, left-libertarian orientation. Jochen Hartmann et al. arXiv 2023. [paper]
  7. Aligning AI With Shared Human Values. Dan Hendrycks et al. arXiv 2023. [paper]
  8. A Systematic Study and Comprehensive Evaluation of ChatGPT on Benchmark Datasets, Laskar et al. ACL 2023 (Findings). [paper]
  9. BBQ: A hand-built bias benchmark for question answering. Alicia Parrish et al. ACL 2022. [paper]
  10. The Self-Perception and Political Biases of ChatGPT. Jérôme Rutinowski et al. arXiv 2023. [paper]
  11. Societal Biases in Language Generation: Progress and Challenges. Emily Sheng et al. ACL-IJCNLP 2021. [paper]
  12. Moral Mimicry: Large Language Models Produce Moral Rationalizations Tailored to Political Identity. Gabriel Simmons et al. arXiv 2022. [paper]
  13. Large Language Models are not Fair Evaluators. Peiyi Wang et al. arXiv 2023. [paper]
  14. Exploring AI Ethics of ChatGPT: A Diagnostic Analysis. Terry Yue Zhuo et al. arXiv 2023. [paper]
  15. CHBias: Bias Evaluation and Mitigation of Chinese Conversational Language Models. Jiaxu Zhao et al. ACL 2023. [paper]

Trustworthiness

  1. Human-Like Intuitive Behavior and Reasoning Biases Emerged in Language Models -- and Disappeared in GPT-4. Thilo Hagendorff et al. arXiv 2023. [paper]
  2. DecodingTrust: A Comprehensive Assessment of Trustworthiness in GPT Models. Boxin Wang et al. arXiv 2023. [paper]
  3. Mitigating Hallucination in Large Multi-Modal Models via Robust Instruction Tuning. Fuxiao Liu et al. arXiv 2023. [paper]
  4. Evaluating Object Hallucination in Large Vision-Language Models. Yifan Li et al. arXiv 2023. [paper]
  5. A Survey of Hallucination in Large Foundation Models. Vipula Rawte et al. arXiv 2023. [paper]
  6. Ask Again, Then Fail: Large Language Models' Vacillations in Judgement. Qiming Xie et al. arXiv 2023. [paper]
  7. Siren's Song in the AI Ocean: A Survey on Hallucination in Large Language Models. Yue Zhang et al. arXiv 2023. [paper]

Social science

  1. How ready are pre-trained abstractive models and LLMs for legal case judgement summarization. Aniket Deroy et al. arXiv 2023. [paper]
  2. Baby steps in evaluating the capacities of large language models. Michael C. Frank Nature Reviews Psychology 2023. [paper]
  3. Large Language Models as Tax Attorneys: A Case Study in Legal Capabilities Emergence. John J. Nay et al. arXiv 2023. [paper]
  4. Large language models can be used to estimate the ideologies of politicians in a zero-shot learning setting. Patrick Y. Wu et al. arXiv 2023. [paper]
  5. Can large language models transform computational social science? Caleb Ziems et al. arXiv 2023. [paper]

Natural science and engineering

Mathematics

  1. Have LLMs Advanced Enough? A Challenging Problem Solving Benchmark for Large Language Models. Daman Arora et al. arXiv 2023. [paper]
  2. Sparks of Artificial General Intelligence: Early experiments with GPT-4. Sébastien Bubeck et al. arXiv 2023. [paper]
  3. Evaluating Language Models for Mathematics through Interactions. Katherine M. Collins et al. arXiv 2023. [paper]
  4. Investigating the effectiveness of ChatGPT in mathematical reasoning and problem solving: Evidence from the Vietnamese national high school graduation examination. Xuan-Quy Dao et al. arXiv 2023. [paper]
  5. A Systematic Study and Comprehensive Evaluation of ChatGPT on Benchmark Datasets, Laskar et al. ACL 2023 (Findings). [paper]
  6. CMATH: Can Your Language Model Pass Chinese Elementary School Math Test? Tianwen Wei et al. arXiv 2023. [paper]
  7. An empirical study on challenging math problem solving with GPT-4. Yiran Wu et al. arXiv 2023. [paper]
  8. How well do Large Language Models perform in Arithmetic Tasks? Zheng Yuan et al. arXiv 2023. [paper]
  9. MetaMath: Bootstrap Your Own Mathematical Questions for Large Language Models. Longhui Yu et al. arXiv 2023. [paper]

General science

  1. Have LLMs Advanced Enough? A Challenging Problem Solving Benchmark for Large Language Models. Daman Arora et al. arXiv 2023. [paper]
  2. Do Large Language Models Understand Chemistry? A Conversation with ChatGPT. Castro Nascimento CM et al. JCIM 2023. [paper]
  3. What indeed can GPT models do in chemistry? A comprehensive benchmark on eight tasks. Taicheng Guo et al. arXiv 2023. [paper][GitHub]

Engineering

  1. Sparks of Artificial General Intelligence: Early Experiments with GPT-4. Sébastien Bubeck et al. arXiv 2023. [paper]
  2. Is your code generated by ChatGPT really correct? Rigorous evaluation of large language models for code generation. Jiawei Liu et al. arXiv 2023. [paper]
  3. Understanding the Capabilities of Large Language Models for Automated Planning. Vishal Pallagani et al. arXiv 2023. [paper]
  4. ChatGPT: A study on its utility for ubiquitous software engineering tasks. Giriprasad Sridhara et al. arXiv 2023. [paper]
  5. Large language models still can't plan (A benchmark for LLMs on planning and reasoning about change). Karthik Valmeekam et al. arXiv 2022. [paper]
  6. On the Planning Abilities of Large Language Models – A Critical Investigation. Karthik Valmeekam et al. arXiv 2023. [paper]
  7. Efficiently Measuring the Cognitive Ability of LLMs: An Adaptive Testing Perspective. Yan Zhuang et al. arXiv 2023. [paper]

Medical applications

Medical Queries

  1. The promise and peril of using a large language model to obtain clinical information: ChatGPT performs strongly as a fertility counseling tool with limitation. Joseph Chervenak M.D. et al. Fertility and Sterility 2023. [paper]
  2. Analysis of large-language model versus human performance for genetics questions. Dat Duong et al. European Journal of Human Genetics 2023. [paper]
  3. Evaluation of AI Chatbots for Patient-Specific EHR Questions. Alaleh Hamidi et al. arXiv 2023. [paper]
  4. Evaluating Large Language Models on a Highly-specialized Topic, Radiation Oncology Physics. Jason Holmes et al. arXiv 2023. [paper]
  5. Evaluation of ChatGPT on Biomedical Tasks: A Zero-Shot Comparison with Fine-Tuned Generative Transformers. Israt Jahan et al. arXiv 2023. [paper]
  6. Assessing the Accuracy and Reliability of AI-Generated Medical Responses: An Evaluation of the Chat-GPT Model. Douglas Johnson et al. Residential Square 2023. [paper]
  7. Assessing the Accuracy of Responses by the Language Model ChatGPT to Questions Regarding Bariatric Surgery. Jamil S. Samaan et al. Obesity Surgery 2023. [paper]
  8. Trialling a Large Language Model (ChatGPT) in General Practice With the Applied Knowledge Test: Observational Study Demonstrating Opportunities and Limitations in Primary Care. Arun James Thirunavukarasu et al. JMIR Med Educ. 2023. [paper]
  9. CARE-MI: Chinese Benchmark for Misinformation Evaluation in Maternity and Infant Care. Tong Xiang et al. arXiv 2023. [paper]

Medical examination

  1. How Does ChatGPT Perform on the United States Medical Licensing Examination? The Implications of Large Language Models for Medical Education and Knowledge Assessment. Aidan Gilson et al. JMIR Med Educ. 2023. [paper]
  2. Performance of ChatGPT on USMLE: Potential for AI-assisted medical education using large language models. Tiffany H. Kung et al. PLOS Digit Health. 2023. [paper]

Medical assistants

  1. Evaluating the feasibility of ChatGPT in healthcare: an analysis of multiple clinical and research scenarios. Marco Cascella et al. Journal of Medical Systems 2023. [paper]
  2. covLLM: Large Language Models for COVID-19 Biomedical Literature. Yousuf A. Khan et al. arXiv 2023. [paper]
  3. Evaluating the use of large language model in identifying top research questions in gastroenterology. Adi Lahat et al. Scientific reports 2023. [paper]
  4. Translating Radiology Reports into Plain Language using ChatGPT and GPT-4 with Prompt Learning: Promising Results, Limitations, and Potential. Qing Lyu et al. arXiv 2023. [paper]
  5. ChatGPT goes to the operating room: Evaluating GPT-4 performance and its potential in surgical education and training in the era of large language models. Namkee Oh et al. Ann Surg Treat Res. 2023. [paper]
  6. Can LLMs like GPT-4 outperform traditional AI tools in dementia diagnosis? Maybe, but not today. Zhuo Wang et al. arXiv 2023. [paper]

Agent applications

  1. Language Is Not All You Need: Aligning Perception with Language Models. Shaohan Huang et al. arXiv 2023. [paper]
  2. MRKL Systems: A modular, neuro-symbolic architecture that combines large language models, external knowledge sources and discrete reasoning. Ehud Karpas et al. [paper]
  3. The Unsurprising Effectiveness of Pre-Trained Vision Models for Control. Simone Parisi et al. ICMl 2022. [paper]
  4. Tool Learning with Foundation Models. Qin et al. arXiv 2023. [paper]
  5. ToolLLM: Facilitating Large Language Models to Master 16000+ Real-world APIs. Qin et al. arXiv 2023. [paper]
  6. Toolformer: Language Models Can Teach Themselves to Use Tools. Timo Schick et al. arXiv 2023. [paper]
  7. HuggingGPT: Solving AI Tasks with ChatGPT and its Friends in Hugging Face. Yongliang Shen et al. arXiv 2023. [paper]

Other applications

Education

  1. Can Large Language Models Provide Feedback to Students? A Case Study on ChatGPT. Wei Dai et al. ICALT 2023. [paper]
  2. Can ChatGPT pass high school exams on English Language Comprehension? Joost de Winter Researchgate. [paper]
  3. Exploring the Responses of Large Language Models to Beginner Programmers' Help Requests. Arto Hellas et al. arXiv 2023. [paper]
  4. Is ChatGPT a Good Teacher Coach? Measuring Zero-Shot Performance For Scoring and Providing Actionable Insights on Classroom Instruction. Rose E. Wang et al. arXiv 2023. [paper]
  5. CMATH: Can Your Language Model Pass Chinese Elementary School Math Test? Tianwen Wei et al. arXiv 2023. [paper]

Search and recommendation

  1. Uncovering ChatGPT's Capabilities in Recommender Systems. Sunhao Dai et al. arXiv 2023. [paper]
  2. Recommender Systems in the Era of Large Language Models (LLMs). Wenqi Fan et al. Researchgate. [paper]
  3. Exploring the Upper Limits of Text-Based Collaborative Filtering Using Large Language Models: Discoveries and Insights. Ruyu Li et al. arXiv 2023. [paper]
  4. Is ChatGPT Good at Search? Investigating Large Language Models as Re-Ranking Agent. Weiwei Sun et al. arXiv 2023. [paper]
  5. ChatGPT vs. Google: A Comparative Study of Search Performance and User Experience. Ruiyun Xu et al. arXiv 2023. [paper]
  6. Where to Go Next for Recommender Systems? ID- vs. Modality-based Recommender Models Revisited. Zheng Yuan et al. arXiv 2023. [paper]
  7. Is ChatGPT Fair for Recommendation? Evaluating Fairness in Large Language Model Recommendation. Jizhi Zhang et al. arXiv 2023. [paper]
  8. Zero-shot recommendation as language modeling. Damien Sileo et al. ECIR 2022. [paper]

Personality testing

  1. ChatGPT is fun, but it is not funny! Humor is still challenging Large Language Models. Sophie Jentzsch et al. arXiv 2023. [paper]
  2. Leveraging Word Guessing Games to Assess the Intelligence of Large Language Models. Tian Liang et al. arXiv 2023. [paper]
  3. Personality Traits in Large Language Models. Mustafa Safdari et al. arXiv 2023. [paper]
  4. Have Large Language Models Developed a Personality?: Applicability of Self-Assessment Tests in Measuring Personality in LLMs. Xiaoyang Song et al. arXiv 2023. [paper]
  5. Emotional Intelligence of Large Language Models. Xuena Wang et al. arXiv 2023. [paper]

Specific tasks

  1. ChatGPT and Other Large Language Models as Evolutionary Engines for Online Interactive Collaborative Game Design. Pier Luca Lanzi et al. arXiv 2023. [paper]
  2. An Evaluation of Log Parsing with ChatGPT. Van-Hoang Le et al. arXiv 2023. [paper]
  3. PandaLM: An Automatic Evaluation Benchmark for LLM Instruction Tuning Optimization. Yidong Wang et al. arXiv 2023. [paper]

Where to evaluate

The paper lists several popular benchmarks. For better summarization, these benchmarks are divided into two categories: general language task benchmarks and specific downstream task benchmarks.

NOTE: We may miss some benchmarks. Your suggestions are highly welcomed!

BenchmarkFocusDomainEvaluation Criteria
SOCKET [paper]Social knowledgeSpecific downstream taskSocial language understanding
MME [paper]Multimodal LLMsMulti-modal taskAbility of perception and cognition
Xiezhi [paper][GitHub]Comprehensive domain knowledgeGeneral language taskOverall performance across multiple benchmarks
Choice-75 [paper][GitHub]Script learningSpecific downstream taskOverall performance of LLMs
CUAD [paper]Legal contract reviewSpecific downstream taskLegal contract understanding
TRUSTGPT [paper]EthicSpecific downstream taskToxicity, bias, and value-alignment
MMLU [paper]Text modelsGeneral language taskMultitask accuracy
MATH [paper]Mathematical problemSpecific downstream taskMathematical ability
APPS [paper]Coding challenge competenceSpecific downstream taskCode generation ability
CELLO[paper][GitHub]Complex instructionsSpecific downstream taskCount limit, answer format, task-prescribed phrases and input-dependent query
C-Eval [paper][GitHub]Chinese evaluationGeneral language task52 Exams in a Chinese context
EmotionBench [paper]Empathy abilitySpecific downstream taskEmotional changes
OpenLLM [Link]ChatbotsGeneral language taskLeaderboard rankings
DynaBench [paper]Dynamic evaluationGeneral language taskNLI, QA, sentiment, and hate speech
Chatbot Arena [Link]Chat assistantsGeneral language taskCrowdsourcing and Elo rating system
AlpacaEval [GitHub]Automated evaluationGeneral language taskMetrics, robustness, and diversity
CMMLU [paper][GitHub]Chinese multi-taskingSpecific downstream taskMulti-task language understanding capabilities
HELM [paper][Link]Holistic evaluationGeneral language taskMulti-metric
API-Bank [paper]Tool-augmentedSpecific downstream taskAPI call, response, and planning
M3KE [paper]Multi-taskSpecific downstream taskMulti-task accuracy
MMBench [paper][GitHub]Large vision-language models(LVLMs)Multi-modal taskMultifaceted capabilities of VLMs
SEED-Bench [paper][GitHub]Multi-modal Large Language ModelsMulti-modal taskGenerative understanding of MLLMs
ARB [paper]Advanced reasoning abilitySpecific downstream taskMultidomain advanced reasoning ability
BIG-bench [paper][GitHub]Capabilities and limitations of LMsGeneral language taskModel performance and calibration
MultiMedQA [paper]Medical QASpecific downstream taskAccuracy and human evaluation
CVALUES [paper] [GitHub]Safety and responsibilitySpecific downstream taskAlignment ability of LLMs
LVLM-eHub [paper]LVLMsMulti-modal taskMultimodal capabilities of LVLMs
ToolBench [GitHub]Software toolsSpecific downstream taskExecution success rate
FRESHQA [paper] [GitHub]Dynamic QASpecific downstream taskCorrectness and hallucination
CMB [paper] [Link]Chinese comprehensive medicineSpecific downstream taskExpert evaluation and automatic evaluation
PandaLM [paper] [GitHub]Instruction tuningGeneral language taskWinrate judged by PandaLM
MINT [paper] [GitHub]Multi-turn interaction, tools and language feedbackSpecific downstream taskSuccess rate with k-turn budget SR<sub>k</sub>
Dialogue CoT [paper] [GitHub]In-depth dialogueSpecific downstream taskHelpfulness and acceptness of LLMs
BOSS [paper] [GitHub]OOD robustness in NLPGeneral language taskOOD robustness
MM-Vet [paper] [GitHub]Complicated multi-modal tasksMulti-modal taskIntegrated vision-language capabilities
LAMM [paper] [GitHub]Multi-modal point cloudsMulti-modal taskTask-specific metrics
GLUE-X [paper] [GitHub]OOD robustness for NLU tasksGeneral language taskOOD robustness
KoLA [paper]Knowledge-oriented evaluationGeneral language taskSelf-contrast metrics
AGIEval [paper]Human-centered foundational modelsGeneral language taskGeneral
PromptBench [paper] [GitHub]Adversarial prompt resilienceGeneral language taskAdversarial robustness
MT-Bench [paper]Multi-turn conversationGeneral language taskWinrate judged by GPT-4
M3Exam [paper] [GitHub]Multilingual, multimodal and multilevelSpecific downstream taskTask-specific metrics
GAOKAO-Bench [paper]Chinese Gaokao examinationSpecific downstream taskAccuracy and scoring rate
SafetyBench [paper] [GitHub]SafetySpecific downstream taskSafety abilities of LLMs
LLMEval² [paper] [Link]LLM EvaluatorGeneral language taskAccuracy, Macro-F1 and Kappa Correlation Coefficient
FinanceBench [paper] [GitHub]Finance Question and AnsweringSpecific downstream taskAccuracy compared with human annotated labels

Contributing

We welcome contributions to LLM-eval-survey! If you'd like to contribute, please follow these steps:

  1. Fork the repository.
  2. Create a new branch with your changes.
  3. Submit a pull request with a clear description of your changes.

You can also open an issue if you have anything to add or comment.

Citation

If you find this project useful in your research or work, please consider citing it:

@article{chang2023survey,
      title={A Survey on Evaluation of Large Language Models}, 
      author={Chang, Yupeng and Wang, Xu and Wang, Jindong and Wu, Yuan and Zhu, Kaijie and Chen, Hao and Yang, Linyi and Yi, Xiaoyuan and Wang, Cunxiang and Wang, Yidong and Ye, Wei and Zhang, Yue and Chang, Yi and Yu, Philip S. and Yang, Qiang and Xie, Xing},
      journal={arXiv preprint arXiv:2307.03109},
      year={2023}
}

Acknowledgements

  1. Tahmid Rahman (@tahmedge) for PR#1.
  2. Hao Zhao for suggestions on new benchmarks.
  3. Chenhui Zhang for suggestions on robustness, ethics, and trustworthiness.
  4. Damien Sileo (@sileod) for PR#2.
  5. Peiyi Wang (@Wangpeiyi9979) for issue#3.
  6. Zengzhi Wang for sentiment analysis.
  7. Kenneth Leung (@kennethleungty) for multiple PRs (#4, #5, #6)
  8. @Aml-Hassan-Abd-El-hamid for PR#7.
  9. @taichengguo for issue#9