{"paper":{"title":"Chain-of-Thought Prompting Elicits Reasoning in Large Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Chain of thought prompting lets large language models reach state-of-the-art accuracy on math word problems using only eight examples.","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Brian Ichter, Dale Schuurmans, Denny Zhou, Ed Chi, Fei Xia, Jason Wei, Maarten Bosma, Quoc Le, Xuezhi Wang","submitted_at":"2022-01-28T02:33:07Z","abstract_excerpt":"We explore how generating a chain of thought -- a series of intermediate reasoning steps -- significantly improves the ability of large language models to perform complex reasoning. In particular, we show how such reasoning abilities emerge naturally in sufficiently large language models via a simple method called chain of thought prompting, where a few chain of thought demonstrations are provided as exemplars in prompting. Experiments on three large language models show that chain of thought prompting improves performance on a range of arithmetic, commonsense, and symbolic reasoning tasks. Th"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"prompting a 540B-parameter language model with just eight chain of thought exemplars achieves state of the art accuracy on the GSM8K benchmark of math word problems, surpassing even finetuned GPT-3 with a verifier.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the performance gains are caused by the explicit reasoning steps rather than simply by providing longer or more detailed prompts; the paper compares against standard few-shot prompting but does not exhaustively rule out all alternative explanations for the improvement.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Chain-of-thought prompting, by including intermediate reasoning steps in few-shot examples, elicits strong reasoning abilities in large language models on arithmetic, commonsense, and symbolic tasks.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Chain of thought prompting lets large language models reach state-of-the-art accuracy on math word problems using only eight examples.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"62017840d700a56709ec2533a3fae53fca980a1f57a6890ebe29ec5407864bc2"},"source":{"id":"2201.11903","kind":"arxiv","version":6},"verdict":{"id":"6818f78d-2a6e-4a5f-bc52-fd2318df83cb","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-10T12:49:15.009832Z","strongest_claim":"prompting a 540B-parameter language model with just eight chain of thought exemplars achieves state of the art accuracy on the GSM8K benchmark of math word problems, surpassing even finetuned GPT-3 with a verifier.","one_line_summary":"Chain-of-thought prompting, by including intermediate reasoning steps in few-shot examples, elicits strong reasoning abilities in large language models on arithmetic, commonsense, and symbolic tasks.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the performance gains are caused by the explicit reasoning steps rather than simply by providing longer or more detailed prompts; the paper compares against standard few-shot prompting but does not exhaustively rule out all alternative explanations for the improvement.","pith_extraction_headline":"Chain of thought prompting lets large language models reach state-of-the-art accuracy on math word problems using only eight examples."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2201.11903/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":86,"sample":[{"doi":"","year":2022,"title":"Do As I Can, Not As I Say: Grounding Language in Robotic Affordances","work_id":"037320f1-b0a9-4cbe-a639-bfb25409ce71","ref_index":1,"cited_arxiv_id":"2204.01691","is_internal_anchor":true},{"doi":"","year":2019,"title":"Aida Amini, Saadia Gabriel, Shanchuan Lin, Rik Koncel-Kedziorski, Yejin Choi, and Hannaneh Hajishirzi. 2019. https://aclanthology.org/N19-1245 M ath QA : Towards interpretable math word problem solvin","work_id":"52727e9e-5988-4c18-9609-e3b38dfb0d99","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.18653/v1/d19-1609","year":2019,"title":"Giving BERT a Calculator: Finding Operations and Arguments with Reading Comprehension","work_id":"c346b2eb-e194-4648-aeee-aa600865fe15","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2018,"title":"Jacob Andreas, Dan Klein, and Sergey Levine. 2018. https://aclanthology.org/N18-1197 Learning with latent language . NAACL","work_id":"f64450f7-143b-4001-8a7f-a8f6f9c4a58e","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"Program Synthesis with Large Language Models","work_id":"fd241a05-03b9-4de2-9588-9d77ce176125","ref_index":5,"cited_arxiv_id":"2108.07732","is_internal_anchor":true}],"resolved_work":86,"snapshot_sha256":"72f62b9f4a6f48785008da5e6374785cbb2e85f43be22329bec5b8e65889c9a6","internal_anchors":14},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}