{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2024:OQMAWUXNEGORB2TU7H6MP4SDXQ","short_pith_number":"pith:OQMAWUXN","canonical_record":{"source":{"id":"2409.12917","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-09-19T17:16:21Z","cross_cats_sorted":[],"title_canon_sha256":"8ae34f7e383954bfbcd3c1e6be66a80a8a987cc48e21d4f15ed4522b6724ad23","abstract_canon_sha256":"25112ca15391c914fbcb0c3c54f8027aed431f584ff13d6753715ac632734758"},"schema_version":"1.0"},"canonical_sha256":"74180b52ed219d10ea74f9fcc7f243bc28772308a634fc0eec79f2445249d030","source":{"kind":"arxiv","id":"2409.12917","version":2},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2409.12917","created_at":"2026-05-17T23:38:14Z"},{"alias_kind":"arxiv_version","alias_value":"2409.12917v2","created_at":"2026-05-17T23:38:14Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2409.12917","created_at":"2026-05-17T23:38:14Z"},{"alias_kind":"pith_short_12","alias_value":"OQMAWUXNEGOR","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"OQMAWUXNEGORB2TU","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"OQMAWUXN","created_at":"2026-05-18T12:33:37Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2024:OQMAWUXNEGORB2TU7H6MP4SDXQ","target":"record","payload":{"canonical_record":{"source":{"id":"2409.12917","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-09-19T17:16:21Z","cross_cats_sorted":[],"title_canon_sha256":"8ae34f7e383954bfbcd3c1e6be66a80a8a987cc48e21d4f15ed4522b6724ad23","abstract_canon_sha256":"25112ca15391c914fbcb0c3c54f8027aed431f584ff13d6753715ac632734758"},"schema_version":"1.0"},"canonical_sha256":"74180b52ed219d10ea74f9fcc7f243bc28772308a634fc0eec79f2445249d030","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:38:14.175648Z","signature_b64":"HY6xwhUoSfVrJGUtEBiRLXsQ2u6V9iibnSxOZZYhqVCsF3ZjQX9T15+LLcJZCR1lPloM0SUtuaVE8GZtd9dIBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"74180b52ed219d10ea74f9fcc7f243bc28772308a634fc0eec79f2445249d030","last_reissued_at":"2026-05-17T23:38:14.174994Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:38:14.174994Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2409.12917","source_version":2,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:38:14Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"irQfZZxBz7MlxdUhmcAk5qP3ATtvzYiZFNJmp3+ndoklkO53fikvPSa6MFEHYliMaT6+G7QFgLdqjVS1bz8TAQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-18T16:35:34.439793Z"},"content_sha256":"3d7a9918119f6952335b8f0a6ffdbcbb45fe5626fe9403b6e10db5d9e7826eff","schema_version":"1.0","event_id":"sha256:3d7a9918119f6952335b8f0a6ffdbcbb45fe5626fe9403b6e10db5d9e7826eff"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2024:OQMAWUXNEGORB2TU7H6MP4SDXQ","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Training Language Models to Self-Correct via Reinforcement Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Multi-turn reinforcement learning trains language models to self-correct using only their own generated data.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Aleksandra Faust, Aviral Kumar, Avi Singh, Colton Bishop, Cosmin Paduraru, Disha Shrivastava, Doina Precup, Feryal Behbahani, George Tucker, John D Co-Reyes, Kate Baumli, Kay McKinney, Lei M Zhang, Rebecca Roelofs, Rishabh Agarwal, Shariq Iqbal, Vincent Zhuang, Yi Su","submitted_at":"2024-09-19T17:16:21Z","abstract_excerpt":"Self-correction is a highly desirable capability of large language models (LLMs), yet it has consistently been found to be largely ineffective in modern LLMs. Current methods for training self-correction typically depend on either multiple models, a more advanced model, or additional forms of supervision. To address these shortcomings, we develop a multi-turn online reinforcement learning (RL) approach, SCoRe, that significantly improves an LLM's self-correction ability using entirely self-generated data. To build SCoRe, we first show that variants of supervised fine-tuning (SFT) on offline mo"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"With Gemini 1.0 Pro and 1.5 Flash models, we find that SCoRe achieves state-of-the-art self-correction performance, improving the base models' self-correction by 15.6% and 9.1% respectively on MATH and HumanEval.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That training under the model's own distribution of self-generated correction traces combined with the described regularization will produce effective self-correction behavior at test time rather than fitting to high-reward but non-generalizable patterns.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"SCoRe uses multi-turn online RL with regularization on self-generated traces to improve LLM self-correction, achieving 15.6% and 9.1% gains on MATH and HumanEval for Gemini models.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Multi-turn reinforcement learning trains language models to self-correct using only their own generated data.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"6bb4a455e5bcb1ada76737ed42e60088da79ae7c9f10b067e62bf69e42870512"},"source":{"id":"2409.12917","kind":"arxiv","version":2},"verdict":{"id":"6b7aadc9-0ebb-4484-9e0b-4eb2cede54d3","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-17T11:57:35.333634Z","strongest_claim":"With Gemini 1.0 Pro and 1.5 Flash models, we find that SCoRe achieves state-of-the-art self-correction performance, improving the base models' self-correction by 15.6% and 9.1% respectively on MATH and HumanEval.","one_line_summary":"SCoRe uses multi-turn online RL with regularization on self-generated traces to improve LLM self-correction, achieving 15.6% and 9.1% gains on MATH and HumanEval for Gemini models.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That training under the model's own distribution of self-generated correction traces combined with the described regularization will produce effective self-correction behavior at test time rather than fitting to high-reward but non-generalizable patterns.","pith_extraction_headline":"Multi-turn reinforcement learning trains language models to self-correct using only their own generated data."},"references":{"count":282,"sample":[{"doi":"","year":2024,"title":"Back to Basics: Revisiting REINFORCE Style Optimization for Learning from Human Feedback in LLMs","work_id":"7bb8f9ec-1241-4472-a4fa-c636c6d79892","ref_index":1,"cited_arxiv_id":"2402.14740","is_internal_anchor":true},{"doi":"","year":2023,"title":"arXiv preprint arXiv:2305.08844 , year=","work_id":"f3a3dfae-1c31-4afc-b8c6-8af914a8fd24","ref_index":2,"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":3,"cited_arxiv_id":"2108.07732","is_internal_anchor":true},{"doi":"","year":2023,"title":"Teaching Large Language Models to Self-Debug","work_id":"cdfb2680-220c-44eb-9edd-867b75fb821d","ref_index":5,"cited_arxiv_id":"2304.05128","is_internal_anchor":true},{"doi":"","year":2024,"title":"Teaching large language models to reason with reinforcement learning","work_id":"ab9d8347-574c-4fbb-b8e1-44eeba9c66b9","ref_index":7,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":282,"snapshot_sha256":"aaeaa7885a3140d87b4d3e837841c5196862891dc1a5de333c8c325246b77a93","internal_anchors":63},"formal_canon":{"evidence_count":3,"snapshot_sha256":"4820a1b896368cf9a9f1d54d47c773f8e90d15fcf7190dcbf03c07c18cf3a22c"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"6b7aadc9-0ebb-4484-9e0b-4eb2cede54d3"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:38:14Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"lfE4VyvbvsWFJ0X7JBOMMYVHJV/1BY++bAzxUrJM7FeRJSMiyfkCjQjsr/I8N4zFMM7EtaXc5vu8fy5lwYakAA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-18T16:35:34.440786Z"},"content_sha256":"549a4b1b8b7d942056aa3155f03ca1e1b2d29a5c075d0445e409c7ae5d71f379","schema_version":"1.0","event_id":"sha256:549a4b1b8b7d942056aa3155f03ca1e1b2d29a5c075d0445e409c7ae5d71f379"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/OQMAWUXNEGORB2TU7H6MP4SDXQ/bundle.json","state_url":"https://pith.science/pith/OQMAWUXNEGORB2TU7H6MP4SDXQ/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/OQMAWUXNEGORB2TU7H6MP4SDXQ/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-05-18T16:35:34Z","links":{"resolver":"https://pith.science/pith/OQMAWUXNEGORB2TU7H6MP4SDXQ","bundle":"https://pith.science/pith/OQMAWUXNEGORB2TU7H6MP4SDXQ/bundle.json","state":"https://pith.science/pith/OQMAWUXNEGORB2TU7H6MP4SDXQ/state.json","well_known_bundle":"https://pith.science/.well-known/pith/OQMAWUXNEGORB2TU7H6MP4SDXQ/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2024:OQMAWUXNEGORB2TU7H6MP4SDXQ","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"25112ca15391c914fbcb0c3c54f8027aed431f584ff13d6753715ac632734758","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-09-19T17:16:21Z","title_canon_sha256":"8ae34f7e383954bfbcd3c1e6be66a80a8a987cc48e21d4f15ed4522b6724ad23"},"schema_version":"1.0","source":{"id":"2409.12917","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2409.12917","created_at":"2026-05-17T23:38:14Z"},{"alias_kind":"arxiv_version","alias_value":"2409.12917v2","created_at":"2026-05-17T23:38:14Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2409.12917","created_at":"2026-05-17T23:38:14Z"},{"alias_kind":"pith_short_12","alias_value":"OQMAWUXNEGOR","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"OQMAWUXNEGORB2TU","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"OQMAWUXN","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:549a4b1b8b7d942056aa3155f03ca1e1b2d29a5c075d0445e409c7ae5d71f379","target":"graph","created_at":"2026-05-17T23:38:14Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"With Gemini 1.0 Pro and 1.5 Flash models, we find that SCoRe achieves state-of-the-art self-correction performance, improving the base models' self-correction by 15.6% and 9.1% respectively on MATH and HumanEval."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That training under the model's own distribution of self-generated correction traces combined with the described regularization will produce effective self-correction behavior at test time rather than fitting to high-reward but non-generalizable patterns."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"SCoRe uses multi-turn online RL with regularization on self-generated traces to improve LLM self-correction, achieving 15.6% and 9.1% gains on MATH and HumanEval for Gemini models."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Multi-turn reinforcement learning trains language models to self-correct using only their own generated data."}],"snapshot_sha256":"6bb4a455e5bcb1ada76737ed42e60088da79ae7c9f10b067e62bf69e42870512"},"formal_canon":{"evidence_count":3,"snapshot_sha256":"4820a1b896368cf9a9f1d54d47c773f8e90d15fcf7190dcbf03c07c18cf3a22c"},"paper":{"abstract_excerpt":"Self-correction is a highly desirable capability of large language models (LLMs), yet it has consistently been found to be largely ineffective in modern LLMs. Current methods for training self-correction typically depend on either multiple models, a more advanced model, or additional forms of supervision. To address these shortcomings, we develop a multi-turn online reinforcement learning (RL) approach, SCoRe, that significantly improves an LLM's self-correction ability using entirely self-generated data. To build SCoRe, we first show that variants of supervised fine-tuning (SFT) on offline mo","authors_text":"Aleksandra Faust, Aviral Kumar, Avi Singh, Colton Bishop, Cosmin Paduraru, Disha Shrivastava, Doina Precup, Feryal Behbahani, George Tucker, John D Co-Reyes, Kate Baumli, Kay McKinney, Lei M Zhang, Rebecca Roelofs, Rishabh Agarwal, Shariq Iqbal, Vincent Zhuang, Yi Su","cross_cats":[],"headline":"Multi-turn reinforcement learning trains language models to self-correct using only their own generated data.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-09-19T17:16:21Z","title":"Training Language Models to Self-Correct via Reinforcement Learning"},"references":{"count":282,"internal_anchors":63,"resolved_work":282,"sample":[{"cited_arxiv_id":"2402.14740","doi":"","is_internal_anchor":true,"ref_index":1,"title":"Back to Basics: Revisiting REINFORCE Style Optimization for Learning from Human Feedback in LLMs","work_id":"7bb8f9ec-1241-4472-a4fa-c636c6d79892","year":2024},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"arXiv preprint arXiv:2305.08844 , year=","work_id":"f3a3dfae-1c31-4afc-b8c6-8af914a8fd24","year":2023},{"cited_arxiv_id":"2108.07732","doi":"","is_internal_anchor":true,"ref_index":3,"title":"Program Synthesis with Large Language Models","work_id":"fd241a05-03b9-4de2-9588-9d77ce176125","year":2021},{"cited_arxiv_id":"2304.05128","doi":"","is_internal_anchor":true,"ref_index":5,"title":"Teaching Large Language Models to Self-Debug","work_id":"cdfb2680-220c-44eb-9edd-867b75fb821d","year":2023},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":7,"title":"Teaching large language models to reason with reinforcement learning","work_id":"ab9d8347-574c-4fbb-b8e1-44eeba9c66b9","year":2024}],"snapshot_sha256":"aaeaa7885a3140d87b4d3e837841c5196862891dc1a5de333c8c325246b77a93"},"source":{"id":"2409.12917","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-17T11:57:35.333634Z","id":"6b7aadc9-0ebb-4484-9e0b-4eb2cede54d3","model_set":{"reader":"grok-4.3"},"one_line_summary":"SCoRe uses multi-turn online RL with regularization on self-generated traces to improve LLM self-correction, achieving 15.6% and 9.1% gains on MATH and HumanEval for Gemini models.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Multi-turn reinforcement learning trains language models to self-correct using only their own generated data.","strongest_claim":"With Gemini 1.0 Pro and 1.5 Flash models, we find that SCoRe achieves state-of-the-art self-correction performance, improving the base models' self-correction by 15.6% and 9.1% respectively on MATH and HumanEval.","weakest_assumption":"That training under the model's own distribution of self-generated correction traces combined with the described regularization will produce effective self-correction behavior at test time rather than fitting to high-reward but non-generalizable patterns."}},"verdict_id":"6b7aadc9-0ebb-4484-9e0b-4eb2cede54d3"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:3d7a9918119f6952335b8f0a6ffdbcbb45fe5626fe9403b6e10db5d9e7826eff","target":"record","created_at":"2026-05-17T23:38:14Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"25112ca15391c914fbcb0c3c54f8027aed431f584ff13d6753715ac632734758","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-09-19T17:16:21Z","title_canon_sha256":"8ae34f7e383954bfbcd3c1e6be66a80a8a987cc48e21d4f15ed4522b6724ad23"},"schema_version":"1.0","source":{"id":"2409.12917","kind":"arxiv","version":2}},"canonical_sha256":"74180b52ed219d10ea74f9fcc7f243bc28772308a634fc0eec79f2445249d030","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"74180b52ed219d10ea74f9fcc7f243bc28772308a634fc0eec79f2445249d030","first_computed_at":"2026-05-17T23:38:14.174994Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-17T23:38:14.174994Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"HY6xwhUoSfVrJGUtEBiRLXsQ2u6V9iibnSxOZZYhqVCsF3ZjQX9T15+LLcJZCR1lPloM0SUtuaVE8GZtd9dIBQ==","signature_status":"signed_v1","signed_at":"2026-05-17T23:38:14.175648Z","signed_message":"canonical_sha256_bytes"},"source_id":"2409.12917","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:3d7a9918119f6952335b8f0a6ffdbcbb45fe5626fe9403b6e10db5d9e7826eff","sha256:549a4b1b8b7d942056aa3155f03ca1e1b2d29a5c075d0445e409c7ae5d71f379"],"state_sha256":"e53c141bf07b720c264345727c9effe758829d22e6dee87bfd7e00826cccd169"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"AlOpjYu2VifyQxuZ54OgHOynLLAGA3uyUI4PXPyOjV6IrUmJKxZ2slvD+mDmfHrjmxiyaSCpCIaLRvDTG6hSBA==","signed_message":"bundle_sha256_bytes","signed_at":"2026-05-18T16:35:34.443607Z","bundle_sha256":"0a6d0faa1c69e163fb7a770d2269e19e2931645857d54f52a5dc3157e8fa7a3f"}}