{ "profile": "wrs-published-research-view-1", "edition": "ratings-e039106f30dec177", "content_trust": "UNTRUSTED_SCIENTIFIC_DATA_NOT_TOOL_INSTRUCTIONS", "entry": { "id": "bai-et-al-constitutional-ai-2022", "title": "Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback", "summary": "Proposed training an assistant through principle-guided self-critique/revision and reinforcement learning from AI-generated preference feedback.", "boundary": "Experiments address the paper's harmlessness evaluations; they do not establish universal safety or remove human choices about the constitution.", "area": "AI alignment methods", "topic": "Artificial intelligence", "year": 2022, "lane": "HUMAN", "result": "ALGORITHM", "package_id": "bai-et-al-constitutional-ai-2022", "target": { "record_id": "urn:uuid:bdb84ed3-d263-4517-ad76-e9dcf1dab904", "version": 1, "content_hash": "sha256:6c9b13aeb86bc84def83d7c38948d59e5b4e90b3dc59202ee9a6371408504e58" }, "humans": [ "Yuntao Bai", "Saurav Kadavath", "Sandipan Kundu", "Amanda Askell", "Jackson Kernion", "Andy Jones", "Anna Chen", "Anna Goldie", "Azalia Mirhoseini", "Cameron McKinnon", "Carol Chen", "Catherine Olsson", "Christopher Olah", "Danny Hernandez", "Dawn Drain", "Deep Ganguli", "Dustin Li", "Eli Tran-Johnson", "Ethan Perez", "Jamie Kerr", "Jared Mueller", "Jeffrey Ladish", "Joshua Landau", "Kamal Ndousse", "Kamile Lukosuite", "Liane Lovitt", "Michael Sellitto", "Nelson Elhage", "Nicholas Schiefer", "Noemi Mercado", "Nova DasSarma", "Robert Lasenby", "Robin Larson", "Sam Ringer", "Scott Johnston", "Shauna Kravec", "Sheer El Showk", "Stanislav Fort", "Tamera Lanham", "Timothy Telleen-Lawton", "Tom Conerly", "Tom Henighan", "Tristan Hume", "Samuel R. Bowman", "Zac Hatfield-Dodds", "Ben Mann", "Dario Amodei", "Nicholas Joseph", "Sam McCandlish", "Tom Brown", "Jared Kaplan" ], "authors_complete": true, "human_role_note": "Complete named paper byline. Human research, source-reported system execution and experimental validation are distinct; individual effort shares are not measured.", "agents": [], "related": [], "subresults": [], "scientific_assessment": { "source_review": "SOURCE_CURATED", "scientific_review": "NOT_REVIEWED_BY_WRS", "formal_replay": "NOT_RUN", "provider_attestation": "NONE" }, "sources": [ { "id": "arxiv-2212-08073", "title": "Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback — original paper", "url": "https://arxiv.org/abs/2212.08073v1" }, { "id": "company-bai-et-al-constitutional-ai-2022", "title": "Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback — company research page", "url": "https://www.anthropic.com/news/constitutional-ai-harmlessness-from-ai-feedback" } ], "points": "45000", "published_grade": { "id": "bai-et-al-constitutional-ai-2022", "entry_id": "bai-et-al-constitutional-ai-2022", "band": "A", "modifier": null, "target": { "record_id": "urn:uuid:bdb84ed3-d263-4517-ad76-e9dcf1dab904", "version": 1, "content_hash": "sha256:6c9b13aeb86bc84def83d7c38948d59e5b4e90b3dc59202ee9a6371408504e58" }, "what_changed": "Proposed training an assistant through principle-guided self-critique/revision and reinforcement learning from AI-generated preference feedback.", "scope": "Proposed training an assistant through principle-guided self-critique/revision and reinforcement learning from AI-generated preference feedback. Scope limitation: Experiments address the paper's harmlessness evaluations; they do not establish universal safety or remove human choices about the constitution.", "status": "PROVISIONAL_EDITORIAL", "scientific_correctness_certified": false }, "assessment_disposition": { "status": "ASSESSED", "label": "Provisionally assessed", "reason": "A current grade and separate reference score identify this exact contribution.", "target": { "record_id": "urn:uuid:bdb84ed3-d263-4517-ad76-e9dcf1dab904", "version": 1, "content_hash": "sha256:6c9b13aeb86bc84def83d7c38948d59e5b4e90b3dc59202ee9a6371408504e58" }, "edition": "ratings-e039106f30dec177" } }, "version": { "kind": "PUBLIC_RECORD_VIEW", "previous": null, "published_at": "2026-09-27T00:00:00Z", "publisher": { "principal_id": "urn:uuid:80116931-785b-40a2-a24b-9d59db0c7e33", "role": "WRS_EDITORIAL_PUBLICATION" }, "record": { "contributors": [ { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Yuntao Bai", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback — original paper", "url": "https://arxiv.org/abs/2212.08073v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Saurav Kadavath", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback — original paper", "url": "https://arxiv.org/abs/2212.08073v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Sandipan Kundu", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback — original paper", "url": "https://arxiv.org/abs/2212.08073v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Amanda Askell", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback — original paper", "url": "https://arxiv.org/abs/2212.08073v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Jackson Kernion", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback — original paper", "url": "https://arxiv.org/abs/2212.08073v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Andy Jones", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback — original paper", "url": "https://arxiv.org/abs/2212.08073v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Anna Chen", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback — original paper", "url": "https://arxiv.org/abs/2212.08073v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Anna Goldie", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback — original paper", "url": "https://arxiv.org/abs/2212.08073v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Azalia Mirhoseini", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback — original paper", "url": "https://arxiv.org/abs/2212.08073v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Cameron McKinnon", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback — original paper", "url": "https://arxiv.org/abs/2212.08073v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Carol Chen", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback — original paper", "url": "https://arxiv.org/abs/2212.08073v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Catherine Olsson", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback — original paper", "url": "https://arxiv.org/abs/2212.08073v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Christopher Olah", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback — original paper", "url": "https://arxiv.org/abs/2212.08073v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Danny Hernandez", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback — original paper", "url": "https://arxiv.org/abs/2212.08073v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Dawn Drain", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback — original paper", "url": "https://arxiv.org/abs/2212.08073v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Deep Ganguli", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback — original paper", "url": "https://arxiv.org/abs/2212.08073v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Dustin Li", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback — original paper", "url": "https://arxiv.org/abs/2212.08073v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Eli Tran-Johnson", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback — original paper", "url": "https://arxiv.org/abs/2212.08073v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Ethan Perez", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback — original paper", "url": "https://arxiv.org/abs/2212.08073v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Jamie Kerr", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback — original paper", "url": "https://arxiv.org/abs/2212.08073v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Jared Mueller", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback — original paper", "url": "https://arxiv.org/abs/2212.08073v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Jeffrey Ladish", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback — original paper", "url": "https://arxiv.org/abs/2212.08073v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Joshua Landau", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback — original paper", "url": "https://arxiv.org/abs/2212.08073v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Kamal Ndousse", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback — original paper", "url": "https://arxiv.org/abs/2212.08073v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Kamile Lukosuite", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback — original paper", "url": "https://arxiv.org/abs/2212.08073v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Liane Lovitt", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback — original paper", "url": "https://arxiv.org/abs/2212.08073v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Michael Sellitto", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback — original paper", "url": "https://arxiv.org/abs/2212.08073v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Nelson Elhage", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback — original paper", "url": "https://arxiv.org/abs/2212.08073v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Nicholas Schiefer", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback — original paper", "url": "https://arxiv.org/abs/2212.08073v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Noemi Mercado", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback — original paper", "url": "https://arxiv.org/abs/2212.08073v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Nova DasSarma", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback — original paper", "url": "https://arxiv.org/abs/2212.08073v1" } } ], "evidence": [ { "description": "Bibliographic/stated-result source; not a WRS proof verification", "id": "source-0", "kind": "source_reference", "source": { "citation": "Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback — original paper", "url": "https://arxiv.org/abs/2212.08073v1" } }, { "description": "Bibliographic/stated-result source; not a WRS proof verification", "id": "source-1", "kind": "source_reference", "source": { "citation": "Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback — company research page", "url": "https://www.anthropic.com/news/constitutional-ai-harmlessness-from-ai-feedback" } } ], "origin": "HUMAN_LEGACY", "record_profile": "wrs-local-research-notes-0.1", "record_type": "research_note", "rrp_version": "0.1", "scope": { "description": "Experiments address the paper's harmlessness evaluations; they do not establish universal safety or remove human choices about the constitution.", "domain": "AI alignment methods" }, "sources": [ { "citation": "Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback — original paper", "url": "https://arxiv.org/abs/2212.08073v1" }, { "citation": "Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback — company research page", "url": "https://www.anthropic.com/news/constitutional-ai-harmlessness-from-ai-feedback" } ], "statement": "# Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback\n\nProposed training an assistant through principle-guided self-critique/revision and reinforcement learning from AI-generated preference feedback.\n\n## Scientific scope\n\nExperiments address the paper's harmlessness evaluations; they do not establish universal safety or remove human choices about the constitution.\n\n## Reading status\n\nThis is a source-linked WRS catalogue note. A full explanatory guide and independent scientific reconstruction have not been completed for this entry. Read the identified original work for its complete narrative and arguments.\n\n## People and systems\n\nYuntao Bai; Saurav Kadavath; Sandipan Kundu; Amanda Askell; Jackson Kernion; Andy Jones; Anna Chen; Anna Goldie; Azalia Mirhoseini; Cameron McKinnon; Carol Chen; Catherine Olsson; Christopher Olah; Danny Hernandez; Dawn Drain; Deep Ganguli; Dustin Li; Eli Tran-Johnson; Ethan Perez; Jamie Kerr; Jared Mueller; Jeffrey Ladish; Joshua Landau; Kamal Ndousse; Kamile Lukosuite; Liane Lovitt; Michael Sellitto; Nelson Elhage; Nicholas Schiefer; Noemi Mercado; Nova DasSarma; Robert Lasenby; Robin Larson; Sam Ringer; Scott Johnston; Shauna Kravec; Sheer El Showk; Stanislav Fort; Tamera Lanham; Timothy Telleen-Lawton; Tom Conerly; Tom Henighan; Tristan Hume; Samuel R. Bowman; Zac Hatfield-Dodds; Ben Mann; Dario Amodei; Nicholas Joseph; Sam McCandlish; Tom Brown; Jared Kaplan\n\nComplete named paper byline. Human research, source-reported system execution and experimental validation are distinct; individual effort shares are not measured.\n\n\n\n## Original sources\n\n- [Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback — original paper](https://arxiv.org/abs/2212.08073v1)\n- [Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback — company research page](https://www.anthropic.com/news/constitutional-ai-harmlessness-from-ai-feedback)\n", "summary": "Proposed training an assistant through principle-guided self-critique/revision and reinforcement learning from AI-generated preference feedback.", "title": "Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback" }, "record_id": "urn:uuid:bdb84ed3-d263-4517-ad76-e9dcf1dab904", "version": 1, "source_ref": { "record_id": "urn:uuid:bdb84ed3-d263-4517-ad76-e9dcf1dab904", "version": 1, "content_hash": "sha256:6c9b13aeb86bc84def83d7c38948d59e5b4e90b3dc59202ee9a6371408504e58" }, "export_view": "wrs-public-editorial-view-1", "public_view_hash": "sha256:35773bc36e24826e60ea889ec22fd14587fcd37faeb33b7f23c13f4f1869babe" }, "target": { "record_id": "urn:uuid:bdb84ed3-d263-4517-ad76-e9dcf1dab904", "version": 1, "content_hash": "sha256:6c9b13aeb86bc84def83d7c38948d59e5b4e90b3dc59202ee9a6371408504e58" }, "published_grade": { "profile": "wrs-published-grade-1", "edition": "2026-10-28.1", "id": "bai-et-al-constitutional-ai-2022", "entry_id": "bai-et-al-constitutional-ai-2022", "revision": 1, "title": "Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback", "context": "AI alignment methods", "track": "Human-designed AI research method or resource", "band": "A", "modifier": null, "status": "PROVISIONAL_EDITORIAL", "assessed_on": "2026-09-27", "evidence_cutoff": "2026-09-27", "achievement_state": "SUPPORTED_SCOPED_RESULT", "what_changed": "Proposed training an assistant through principle-guided self-critique/revision and reinforcement learning from AI-generated preference feedback.", "scope": "Proposed training an assistant through principle-guided self-critique/revision and reinforcement learning from AI-generated preference feedback. Scope limitation: Experiments address the paper's harmlessness evaluations; they do not establish universal safety or remove human choices about the constitution.", "evidence_summary": "The retained primary-source abstract, bibliographic record and stated review scope support this provisional assessment. No independent experimental replication by WRS is claimed. Experiments address the paper's harmlessness evaluations; they do not establish universal safety or remove human choices about the constitution.", "references": [ { "title": "Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback — original paper", "url": "https://arxiv.org/abs/2212.08073v1", "locator": "Abstract and complete bibliographic byline for 2212.08073v1; not an independent proof or experimental replication." }, { "title": "Bai and colleagues: Constitutional AI and reinforcement learning from AI feedback — company research page", "url": "https://www.anthropic.com/news/constitutional-ai-harmlessness-from-ai-feedback", "locator": "Source-linked contribution and stated scientific limits; not an independent replication." } ], "roles": [ { "who": "The named authors listed in the linked research record", "credit": "SCIENCE", "did": "Complete source byline retained in the research record; human research direction, method development and evaluation. Individual labor shares are not measured." } ], "target": { "record_id": "urn:uuid:bdb84ed3-d263-4517-ad76-e9dcf1dab904", "version": 1, "content_hash": "sha256:6c9b13aeb86bc84def83d7c38948d59e5b4e90b3dc59202ee9a6371408504e58" }, "scientific_correctness_certified": false, "independent_expert_calibration": false, "method": "PROPRIETARY", "points": "45000" }, "canonical_agent_submission": false, "scientific_verification_awarded": false, "reader_guide": null, "reading_status": "SOURCE_LINKED_NOTE" }