{"id":"arxiv:2303.17651","title":"Self-Refine: Iterative Refinement with Self-Feedback","date":"2023-03-30","section":"Post-training","quality":0.100328,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/","source_url":"https://arxiv.org/abs/2303.17651","text_available":true,"authors":["Aman Madaan","Niket Tandon","Prakhar Gupta","Skyler Hallinan"],"outgoing":[{"source":"arxiv:2303.17651","target":"url:rlhfbook.com","evidence":"title","id":"url:rlhfbook.com","title":"Reinforcement Learning from Human Feedback","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/20e354559fad09ad4d6338d104c9c3f822a48e7fbf86cd8afa1ddf951b223658/"}],"incoming":[{"source":"arxiv:2309.16797","target":"arxiv:2303.17651","evidence":"title","id":"arxiv:2309.16797","title":"Promptbreeder: Self-Referential Self-Improvement Via Prompt Evolution","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/daac3bcfcd7f39084d995905e9ef96ae01c5aab7bfe8e565bd7622b4e44088f7/"},{"source":"arxiv:2402.13669","target":"arxiv:2303.17651","evidence":"title","id":"arxiv:2402.13669","title":"Self-Distillation Bridges Distribution Gap in Language Model Fine-Tuning","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1a8d11d48a58f35055e7f88f699cf42688a3698c63bd7026d5789f72a80581bf/"},{"source":"arxiv:2407.21783","target":"arxiv:2303.17651","evidence":"title","id":"arxiv:2407.21783","title":"The Llama 3 Herd of Models","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/"},{"source":"arxiv:2408.08435","target":"arxiv:2303.17651","evidence":"title","id":"arxiv:2408.08435","title":"Automated Design of Agentic Systems","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/c9890d5668d1c1b33e29569b74626fd586f456af6a97c657b8c99895c90f59a5/"},{"source":"arxiv:2410.04444","target":"arxiv:2303.17651","evidence":"title","id":"arxiv:2410.04444","title":"Gödel Agent: A Self-Referential Agent Framework for Recursive Self-Improvement","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/d04ae5d37f10547be79c728efed48a629e631ab5384155564e52558e12ff4362/"},{"source":"arxiv:2412.14135","target":"arxiv:2303.17651","evidence":"title","id":"arxiv:2412.14135","title":"Scaling of Search and Learning: A Roadmap to Reproduce o1 from Reinforcement Learning Perspective","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5a0b4b0b524698ee5f30eab659b7cd5104f5fe6497c8dec2d0c4b9d332350047/"},{"source":"arxiv:2504.16084","target":"arxiv:2303.17651","evidence":"title","id":"arxiv:2504.16084","title":"TTRL: Test-Time Reinforcement Learning","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/ad54dd58d24afdfefc8aa5087d07c4aab2143e90ef06f6f7db8d3f3333452c48/"},{"source":"arxiv:2505.15134","target":"arxiv:2303.17651","evidence":"title","id":"arxiv:2505.15134","title":"The Unreasonable Effectiveness of Entropy Minimization in LLM Reasoning","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/12b01b403e280a32d41fd991a27fa30f33315e09f0f42a4f43e9bece5ad9c53d/"},{"source":"arxiv:2505.21444","target":"arxiv:2303.17651","evidence":"title","id":"arxiv:2505.21444","title":"Can Large Reasoning Models Self-Train?","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/206224ca50098e0bdc230140722feec107cfc90c1cef2ed8d3ed52355ef56bca/"},{"source":"arxiv:2505.22954","target":"arxiv:2303.17651","evidence":"title","id":"arxiv:2505.22954","title":"Darwin Godel Machine: Open-Ended Evolution of Self-Improving Agents","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/c29f716871f80314868b5647692de387f26e12166b93bd362915fd272b6e081c/"},{"source":"arxiv:2507.19457","target":"arxiv:2303.17651","evidence":"title","id":"arxiv:2507.19457","title":"GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/c1ef72d96fa3292590a6a3d112ab509840081b59aaa771cf42b8b0e8ef62ba08/"},{"source":"arxiv:2509.20317","target":"arxiv:2303.17651","evidence":"title","id":"arxiv:2509.20317","title":"SIM-CoT: Supervised Implicit Chain-of-Thought","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/72b64dd6a30c71d99920cafabbb2c909d03febab8991eeaf3c091b16cb7ed348/"},{"source":"arxiv:2601.20802","target":"arxiv:2303.17651","evidence":"title","id":"arxiv:2601.20802","title":"Reinforcement Learning via Self-Distillation","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/6073ec3d17b9a932a68e79ae078928f4635e40fc79cfcbc31709715d1f1e1ffa/"},{"source":"arxiv:2602.09000","target":"arxiv:2303.17651","evidence":"title","id":"arxiv:2602.09000","title":"iGRPO: Self-Feedback-Driven LLM Reasoning","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1cd6a66823fe24e8e2d83e817177d3a1e8cda1d7d5b59a4397f8669bee6306ae/"},{"source":"arxiv:2603.28052","target":"arxiv:2303.17651","evidence":"title","id":"arxiv:2603.28052","title":"Meta-Harness: End-to-End Optimization of Model Harnesses","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/8bb0b87384dfdd42c419cf26cc7e52f1f0d89f63afa5d5968e246b73a44c99cb/"},{"source":"arxiv:2604.08224","target":"arxiv:2303.17651","evidence":"id","id":"arxiv:2604.08224","title":"Externalization in LLM Agents: A Unified Review of Memory, Skills, Protocols and Harness Engineering","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/392317a8b3bedec8258ff8288d69ac72bc9e60299c5f7bec8962838bb448c2ce/"},{"source":"arxiv:2604.19341","target":"arxiv:2303.17651","evidence":"title","id":"arxiv:2604.19341","title":"Structured Scaling of AI Discovery Across Diverse Scientific Domains","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/33ad3375247f86197210840a19064548b997e8617475e606d7a7ac2cda64740e/"},{"source":"arxiv:2604.25850","target":"arxiv:2303.17651","evidence":"title","id":"arxiv:2604.25850","title":"Agentic Harness Engineering: Observability-Driven Automatic Evolution of Coding-Agent Harnesses","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/c2358d2261b7478833ca59529278f0f4872896ca5ae4434e8db5d420532e10bd/"},{"source":"arxiv:2607.13104","target":"arxiv:2303.17651","evidence":"title","id":"arxiv:2607.13104","title":"Self-Improvements in Modern Agentic Systems: A Survey","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/"},{"source":"arxiv:2608.08311","target":"arxiv:2303.17651","evidence":"id","id":"arxiv:2608.08311","title":"Ouroboros: A Self-Developing Frontier Coding Agent with Reviewed Core Evolution","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4b8f53ea59f810bf1717d78f1907b7596b91df2bec6751aba12accc8598c78e9/"},{"source":"url:lilianweng.github.io/posts/2026-07-04-harness","target":"arxiv:2303.17651","evidence":"id","id":"url:lilianweng.github.io/posts/2026-07-04-harness","title":"Harnesses for models (Lilian Weng, 2026)","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/c1ea249334ba128a9fff468e25fa4f8d13d4cccd13661ee130bec020ebb6463b/"}],"text_parts":[{"start":0,"end":15862,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-1.html"},{"start":15862,"end":30656,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-2.html"},{"start":30656,"end":46394,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-3.html"},{"start":46394,"end":61530,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-4.html"},{"start":61530,"end":66278,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-5.html"}],"contents":[{"title":"Self-Refine : Iterative Refinement with Self-Feedback","offset":0,"level":1,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-1.html#text"},{"title":"Abstract","offset":57,"level":6,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-1.html#text"},{"title":"1 Introduction","offset":1310,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-1.html#text"},{"title":"2 Iterative Refinement with Self-Refine","offset":5232,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-1.html#text"},{"title":"Initial generation","offset":5759,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-1.html#text"},{"title":"feedback","offset":6618,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-1.html#text"},{"title":"refine","offset":7912,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-1.html#text"},{"title":"3 Evaluation","offset":10154,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-1.html#text"},{"title":"3.1 Instantiating Self-Refine","offset":10811,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-1.html#text"},{"title":"Base LLMs","offset":11272,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-1.html#text"},{"title":"3.2 Metrics","offset":12109,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-1.html#text"},{"title":"3.3 Results","offset":13288,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-1.html#text"},{"title":"4 Analysis","offset":15847,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-1.html#text"},{"title":"The impact of the feedback quality","offset":16056,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-2.html#text"},{"title":"How important are the multiple iterations of feedback - refine ?","offset":17559,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-2.html#text"},{"title":"Can we just generate multiple outputs instead of refining?","offset":18849,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-2.html#text"},{"title":"Does Self-Refine work with weaker models?","offset":19681,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-2.html#text"},{"title":"Qualitative Analysis","offset":20712,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-2.html#text"},{"title":"Going Beyond Benchmarks","offset":22538,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-2.html#text"},{"title":"5 Related work","offset":23185,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-2.html#text"},{"title":"Source of feedback.","offset":23821,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-2.html#text"},{"title":"Representation of feedback.","offset":24646,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-2.html#text"},{"title":"Types of refiners.","offset":25110,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-2.html#text"},{"title":"Non-refinement reinforcement learning (RL) approaches.","offset":25776,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-2.html#text"},{"title":"6 Limitations and Discussion","offset":26268,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-2.html#text"},{"title":"7 Conclusion","offset":27437,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-2.html#text"},{"title":"Appendix A Evaluation Tasks","offset":28233,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-2.html#text"},{"title":"Appendix B Broader Related Work","offset":28338,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-2.html#text"},{"title":"Comparison with Welleck et al. (2022)","offset":29056,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-2.html#text"},{"title":"Comparison with non-refinement reinforcement learning (RL) approaches.","offset":30002,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-2.html#text"},{"title":"Appendix C Human Evaluation","offset":32492,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-3.html#text"},{"title":"Appendix D GPT-4 Evaluation","offset":33227,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-3.html#text"},{"title":"Appendix E Model Key","offset":33645,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-3.html#text"},{"title":"Appendix F Comparison of Self-Refine with State-of-the-art of Few-Shot Learning Models and Fine-Tuned Baselines","offset":33744,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-3.html#text"},{"title":"Appendix G Evaluation of Vicuna-13b","offset":34317,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-3.html#text"},{"title":"Mixed-refine: Improving Vicuna-13b with ChatGPT","offset":35062,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-3.html#text"},{"title":"Appendix H Additional Analysis","offset":35653,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-3.html#text"},{"title":"H.1 Using Oracle Feedback","offset":35688,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-3.html#text"},{"title":"Non-monotonic increase in output quality for acronym generation","offset":36158,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-3.html#text"},{"title":"Feedback and Refinement Error Analysis for Response Generation","offset":36946,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-3.html#text"},{"title":"Appendix I Beyond Benchmarks","offset":38164,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-3.html#text"},{"title":"Ice Cream Generation","offset":38790,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-3.html#text"},{"title":"Photosynthesis","offset":39516,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-3.html#text"},{"title":"Appendix J Statistical Confidence Intervals","offset":39884,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-3.html#text"},{"title":"Appendix K New Tasks","offset":40380,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-3.html#text"},{"title":"Constrained Generation","offset":40405,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-3.html#text"},{"title":"Acronym Generation","offset":41003,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-3.html#text"},{"title":"Appendix L Code Readability","offset":41494,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-3.html#text"},{"title":"L.1 Method","offset":42054,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-3.html#text"},{"title":"L.2 Experiments","offset":43196,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-3.html#text"},{"title":"Dataset","offset":43217,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-3.html#text"},{"title":"Implementation","offset":43675,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-3.html#text"},{"title":"Evaluation Methods","offset":44185,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-3.html#text"},{"title":"Result","offset":45007,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-3.html#text"},{"title":"Example","offset":46278,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-3.html#text"},{"title":"Appendix M Dialogue Response Generation","offset":46350,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-3.html#text"},{"title":"M.1 Modules","offset":46829,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-4.html#text"},{"title":"Init","offset":46960,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-4.html#text"},{"title":"Feedback","offset":47127,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-4.html#text"},{"title":"Iterate","offset":48614,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-4.html#text"},{"title":"M.2 Setup and Experiments","offset":48874,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-4.html#text"},{"title":"Model and Baseline","offset":48905,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-4.html#text"},{"title":"Evaluation","offset":50006,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-4.html#text"},{"title":"Results","offset":51369,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-4.html#text"},{"title":"Appendix N Code Optimization","offset":51957,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-4.html#text"},{"title":"Appendix O Math Reasoning","offset":52466,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-4.html#text"},{"title":"Appendix P Sentiment Reversal","offset":53650,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-4.html#text"},{"title":"Instantiating Self-Refine for sentiment reversal","offset":54212,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-4.html#text"},{"title":"P.1 Details","offset":54857,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-4.html#text"},{"title":"Evaluation","offset":54874,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-4.html#text"},{"title":"Pin-pointed feedback","offset":56625,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-4.html#text"},{"title":"Evaluation","offset":57374,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-4.html#text"},{"title":"Appendix Q Acronym Generation","offset":57517,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-4.html#text"},{"title":"feedback","offset":58430,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-4.html#text"},{"title":"Appendix R Constrained Generation","offset":60694,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-4.html#text"},{"title":"Appendix S Prompts","offset":62402,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-5.html#text"},{"title":"Instructions for reporting errors","offset":65314,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/part-5.html#text"}],"text_code_points":66278,"source_markdown":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/source.md"}