{"id":"arxiv:2203.11171","title":"Self-Consistency Improves Chain of Thought Reasoning in Language Models","date":"2022-03-21","section":"Reasoning and the \"physics\" of language models","quality":0.242635,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/","source_url":"https://arxiv.org/abs/2203.11171","text_available":true,"authors":["Xuezhi Wang","Jason Wei","Dale Schuurmans","Quoc Le"],"outgoing":[{"source":"arxiv:2203.11171","target":"arxiv:2101.02235","evidence":"title","id":"arxiv:2101.02235","title":"Did Aristotle Use a Laptop? A Question Answering Benchmark with Implicit Reasoning Strategies","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5b44455f95aaaeefc838faceeaf20aaf1008889ca2e479d95291e245d5365a36/"}],"incoming":[{"source":"arxiv:2212.14024","target":"arxiv:2203.11171","evidence":"id","id":"arxiv:2212.14024","title":"Demonstrate-Search-Predict: Composing retrieval and language models for knowledge-intensive NLP","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/09c4c9bae4e016c9ce276cf84a46742b5b21168f4181202b35c207baf887f71e/"},{"source":"arxiv:2302.00923","target":"arxiv:2203.11171","evidence":"title","id":"arxiv:2302.00923","title":"Multimodal Chain-of-Thought Reasoning in Language Models","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/0fe1320410a5c35c29f2ac16d69a70adbf3c490670902745c2b4faf88460b5c7/"},{"source":"arxiv:2309.16797","target":"arxiv:2203.11171","evidence":"id","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:2311.16452","target":"arxiv:2203.11171","evidence":"title","id":"arxiv:2311.16452","title":"Can Generalist Foundation Models Outcompete Special-Purpose Tuning? Case Study in Medicine","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/8e8303d0328f412b165c8cdfca91debf742ea45ed7865a0d96ee3905e5bba919/"},{"source":"arxiv:2402.16823","target":"arxiv:2203.11171","evidence":"id","id":"arxiv:2402.16823","title":"Language Agents as Optimizable Graphs","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/750bd3807bf9c9a9b5cd2d60cc14387431aec3fe8a3f942916ca5d8922b2a5b0/"},{"source":"arxiv:2408.08435","target":"arxiv:2203.11171","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:2203.11171","evidence":"id","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:2203.11171","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:2501.04519","target":"arxiv:2203.11171","evidence":"title","id":"arxiv:2501.04519","title":"rStar-Math: Small LLMs Can Master Math Reasoning with Self-Evolved Deep Thinking","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/473ad7b5a354658a1d241813741e15a8ab51e7c2ea34f0c3d22eb06884864a46/"},{"source":"arxiv:2501.14082","target":"arxiv:2203.11171","evidence":"id","id":"arxiv:2501.14082","title":"Communicating Activations Between Language Model Agents","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/e67d4addcda44093e242074332925d000726f87bd9538f6b318bdfc454a7dc13/"},{"source":"arxiv:2502.03387","target":"arxiv:2203.11171","evidence":"id","id":"arxiv:2502.03387","title":"LIMO: Less is More for Reasoning","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/714392528bb842fb3a61cee5367942f697536f0131b0fcac85c6eb8d2212125c/"},{"source":"arxiv:2503.00735","target":"arxiv:2203.11171","evidence":"title","id":"arxiv:2503.00735","title":"LADDER: Self-Improving LLMs Through Recursive Problem Decomposition","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/65b264030233bf2bcd17284eb9a521e6e38d4ee83eb7fec6587e4c047199f227/"},{"source":"arxiv:2503.02875","target":"arxiv:2203.11171","evidence":"title","id":"arxiv:2503.02875","title":"The First Few Tokens Are All You Need: An Efficient and Effective Unsupervised Prefix Fine-Tuning Method for Reasoning Models","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b9349378a2ea40a7998ba364a4587c40dff3cc9b22750e14b3b21e296a4ecb12/"},{"source":"arxiv:2504.16084","target":"arxiv:2203.11171","evidence":"id","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:2203.11171","evidence":"id","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:2203.11171","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:2506.06632","target":"arxiv:2203.11171","evidence":"title","id":"arxiv:2506.06632","title":"Curriculum Reinforcement Learning from Easy to Hard Tasks Improves LLM Reasoning","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/79fda9609f9b88bb71f71f531b63ea6b5237d1c889b7ae59e9ca07c12dbe70d1/"},{"source":"arxiv:2509.09675","target":"arxiv:2203.11171","evidence":"id","id":"arxiv:2509.09675","title":"CDE: Curiosity-Driven Exploration for Efficient Reinforcement Learning in Large Language Models","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/f5581044978796ff7f35cb68c27963d3abbb391d35d88173ff7801242cbb1f7a/"},{"source":"arxiv:2509.13232","target":"arxiv:2203.11171","evidence":"title","id":"arxiv:2509.13232","title":"Single-stream Policy Optimization","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/d699d0e034b21f3b85c3f70ce4fb95064fbf59685ce752935e6d173dac2b71f3/"},{"source":"arxiv:2509.14234","target":"arxiv:2203.11171","evidence":"title","id":"arxiv:2509.14234","title":"Compute as Teacher: Turning Inference Compute Into Reference-Free Supervision","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/a6f1443a7e55dd6421fdbe43e65e19c1962a19b135d737fd67f414d3324ce85e/"},{"source":"arxiv:2509.20317","target":"arxiv:2203.11171","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:2604.08224","target":"arxiv:2203.11171","evidence":"title","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:2203.11171","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:2605.07654","target":"arxiv:2203.11171","evidence":"title","id":"arxiv:2605.07654","title":"Reliable Chain-of-Thought via Prefix Consistency","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/715ed2b4b2f355b771111eb0e65e113f100dc33739423a76307e8ece18b57d18/"},{"source":"arxiv:2606.29150","target":"arxiv:2203.11171","evidence":"title","id":"arxiv:2606.29150","title":"Flow Reasoning Models: Turning Flows Into Efficient Recurrent Reasoners","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/8be1801f0f562711039ced2b42f324434c6ab829f544b8d736bc85405b64021b/"},{"source":"arxiv:2608.17163","target":"arxiv:2203.11171","evidence":"id","id":"arxiv:2608.17163","title":"Q-Learning With World Models","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/15ca8843297a713249e241e591505a44859dac258c54e299d79c62c81bd893c2/"},{"source":"arxiv:2609.01117","target":"arxiv:2203.11171","evidence":"id","id":"arxiv:2609.01117","title":"Latent Recurrent Thoughts: Recurrent Refinement of Proposed Latents for Reasoning with Frozen LLMs","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/e9d4f61afc3930fa770d5ea53d3f43a5157fa4904fcdfaa514786e617905e20b/"}],"text_parts":[{"start":0,"end":15796,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/part-1.html"},{"start":15796,"end":31132,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/part-2.html"},{"start":31132,"end":38950,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/part-3.html"}],"contents":[{"title":"Self-Consistency Improves Chain of Thought Reasoning in Language Models","offset":0,"level":1,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/part-1.html#text"},{"title":"Abstract","offset":75,"level":6,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/part-1.html#text"},{"title":"1 Introduction","offset":1033,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/part-1.html#text"},{"title":"2 Self-Consistency over Diverse Reasoning Paths","offset":5638,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/part-1.html#text"},{"title":"3 Experiments","offset":12909,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/part-1.html#text"},{"title":"3.1 Experiment setup","offset":13217,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/part-1.html#text"},{"title":"Tasks and datasets.","offset":13243,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/part-1.html#text"},{"title":"Language models and prompts.","offset":14561,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/part-1.html#text"},{"title":"Sampling scheme.","offset":16370,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/part-2.html#text"},{"title":"3.2 Main Results","offset":16982,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/part-2.html#text"},{"title":"Arithmetic Reasoning","offset":17372,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/part-2.html#text"},{"title":"Commonsense and Symbolic Reasoning","offset":18562,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/part-2.html#text"},{"title":"3.3 Self-Consistency Helps When Chain-of-Thought Hurts Performance","offset":19802,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/part-2.html#text"},{"title":"3.4 Compare to other existing approaches","offset":20856,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/part-2.html#text"},{"title":"Comparison to Sample-and-Rank","offset":21087,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/part-2.html#text"},{"title":"Comparison to Beam Search","offset":21748,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/part-2.html#text"},{"title":"Comparison to Ensemble-based Approaches","offset":22447,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/part-2.html#text"},{"title":"3.5 Additional Studies","offset":23623,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/part-2.html#text"},{"title":"Self-Consistency is Robust to Sampling Strategies and Scaling","offset":23907,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/part-2.html#text"},{"title":"Self-Consistency Improves Robustness to Imperfect Prompts","offset":24625,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/part-2.html#text"},{"title":"Self-Consistency Works for Non-Natural-Language Reasoning Paths and Zero-shot CoT","offset":25943,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/part-2.html#text"},{"title":"4 Related work","offset":26843,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/part-2.html#text"},{"title":"Reasoning in language models.","offset":26862,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/part-2.html#text"},{"title":"Sampling and re-ranking in language models.","offset":27467,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/part-2.html#text"},{"title":"Extract reasoning paths.","offset":28854,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/part-2.html#text"},{"title":"Consistency in language models.","offset":29947,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/part-2.html#text"},{"title":"5 Conclusion and Discussion","offset":30637,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/part-2.html#text"},{"title":"Reproducibility Statement","offset":31919,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/part-3.html#text"},{"title":"Ethics Statement","offset":32747,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/part-3.html#text"},{"title":"Appendix A Appendix","offset":33442,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/part-3.html#text"},{"title":"A.1 Additional Experiment Results","offset":33466,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/part-3.html#text"},{"title":"A.1.1 Robustness to Sampling Strategies and Parameters","offset":33505,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/part-3.html#text"},{"title":"A.1.2 Robustness to different sets of prompts","offset":34458,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/part-3.html#text"},{"title":"A.1.3 Compared to model ensembles","offset":34795,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/part-3.html#text"},{"title":"A.1.4 Combining self-consistency with other ensembling strategies","offset":35679,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/part-3.html#text"},{"title":"A.2 Details on Resources and Inference","offset":36247,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/part-3.html#text"},{"title":"A.3 Full Sets of Prompts","offset":37187,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/part-3.html#text"},{"title":"Instructions for reporting errors","offset":37986,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/part-3.html#text"}],"text_code_points":38950,"source_markdown":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/source.md"}