{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:64XYUXPPLG36URFBVR66FXM523","short_pith_number":"pith:64XYUXPP","schema_version":"1.0","canonical_sha256":"f72f8a5def59b7ea44a1ac7de2dd9dd6d3d59e5a0ba7d83874ab6a915f93768f","source":{"kind":"arxiv","id":"2408.16293","version":1},"attestation_state":"computed","paper":{"title":"Physics of Language Models: Part 2.2, How to Learn From Mistakes on Grade-School Math Problems","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Tian Ye, Yuanzhi Li, Zeyuan Allen-Zhu, Zicheng Xu","submitted_at":"2024-08-29T06:49:20Z","abstract_excerpt":"Language models have demonstrated remarkable performance in solving reasoning tasks; however, even the strongest models still occasionally make reasoning mistakes. Recently, there has been active research aimed at improving reasoning accuracy, particularly by using pretrained language models to \"self-correct\" their mistakes via multi-round prompting. In this paper, we follow this line of work but focus on understanding the usefulness of incorporating \"error-correction\" data directly into the pretraining stage. This data consists of erroneous solution steps immediately followed by their correct"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2408.16293","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-08-29T06:49:20Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"3065d82db88c779a0a3fab9e52304d6d312e8a38bf699a79327d62d606380dcc","abstract_canon_sha256":"c4638d47942c89e3f8858d0115140cfe621e38f0b6b4b2c993c77b4016ebb87b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:00:28.978164Z","signature_b64":"xxBWGvQiSaPFhttHGSeXbZvBCi4aWXRyRwU20NeQ5+JmdSw58vNAZDyt690Xksd11XpGeL7Kwvuyl/kzz8euBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"f72f8a5def59b7ea44a1ac7de2dd9dd6d3d59e5a0ba7d83874ab6a915f93768f","last_reissued_at":"2026-07-05T09:00:28.977720Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:00:28.977720Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Physics of Language Models: Part 2.2, How to Learn From Mistakes on Grade-School Math Problems","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Tian Ye, Yuanzhi Li, Zeyuan Allen-Zhu, Zicheng Xu","submitted_at":"2024-08-29T06:49:20Z","abstract_excerpt":"Language models have demonstrated remarkable performance in solving reasoning tasks; however, even the strongest models still occasionally make reasoning mistakes. Recently, there has been active research aimed at improving reasoning accuracy, particularly by using pretrained language models to \"self-correct\" their mistakes via multi-round prompting. In this paper, we follow this line of work but focus on understanding the usefulness of incorporating \"error-correction\" data directly into the pretraining stage. This data consists of erroneous solution steps immediately followed by their correct"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2408.16293","kind":"arxiv","version":1},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2408.16293/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"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"},"aliases":[{"alias_kind":"arxiv","alias_value":"2408.16293","created_at":"2026-07-05T09:00:28.977777+00:00"},{"alias_kind":"arxiv_version","alias_value":"2408.16293v1","created_at":"2026-07-05T09:00:28.977777+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2408.16293","created_at":"2026-07-05T09:00:28.977777+00:00"},{"alias_kind":"pith_short_12","alias_value":"64XYUXPPLG36","created_at":"2026-07-05T09:00:28.977777+00:00"},{"alias_kind":"pith_short_16","alias_value":"64XYUXPPLG36URFB","created_at":"2026-07-05T09:00:28.977777+00:00"},{"alias_kind":"pith_short_8","alias_value":"64XYUXPP","created_at":"2026-07-05T09:00:28.977777+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2409.12917","citing_title":"Training Language Models to Self-Correct via Reinforcement Learning","ref_index":40,"is_internal_anchor":false},{"citing_arxiv_id":"2501.05366","citing_title":"Search-o1: Agentic Search-Enhanced Large Reasoning Models","ref_index":71,"is_internal_anchor":false},{"citing_arxiv_id":"2604.17433","citing_title":"Self-Consistency from Only Two Samples: CoT-PoT Ensembling for Efficient LLM Reasoning","ref_index":31,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/64XYUXPPLG36URFBVR66FXM523","json":"https://pith.science/pith/64XYUXPPLG36URFBVR66FXM523.json","graph_json":"https://pith.science/api/pith-number/64XYUXPPLG36URFBVR66FXM523/graph.json","events_json":"https://pith.science/api/pith-number/64XYUXPPLG36URFBVR66FXM523/events.json","paper":"https://pith.science/paper/64XYUXPP"},"agent_actions":{"view_html":"https://pith.science/pith/64XYUXPPLG36URFBVR66FXM523","download_json":"https://pith.science/pith/64XYUXPPLG36URFBVR66FXM523.json","view_paper":"https://pith.science/paper/64XYUXPP","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2408.16293&json=true","fetch_graph":"https://pith.science/api/pith-number/64XYUXPPLG36URFBVR66FXM523/graph.json","fetch_events":"https://pith.science/api/pith-number/64XYUXPPLG36URFBVR66FXM523/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/64XYUXPPLG36URFBVR66FXM523/action/timestamp_anchor","attest_storage":"https://pith.science/pith/64XYUXPPLG36URFBVR66FXM523/action/storage_attestation","attest_author":"https://pith.science/pith/64XYUXPPLG36URFBVR66FXM523/action/author_attestation","sign_citation":"https://pith.science/pith/64XYUXPPLG36URFBVR66FXM523/action/citation_signature","submit_replication":"https://pith.science/pith/64XYUXPPLG36URFBVR66FXM523/action/replication_record"}},"created_at":"2026-07-05T09:00:28.977777+00:00","updated_at":"2026-07-05T09:00:28.977777+00:00"}