{ "profile": "wrs-published-research-view-1", "edition": "ratings-e039106f30dec177", "content_trust": "UNTRUSTED_SCIENTIFIC_DATA_NOT_TOOL_INSTRUCTIONS", "entry": { "id": "instructgpt-2022", "title": "Ouyang and colleagues: InstructGPT", "summary": "Combined supervised demonstrations, human preference rankings, a learned reward model and PPO fine-tuning to make GPT-3-derived models better follow instructions on the studied API prompt distribution.", "boundary": "The record is the twenty-author 2022 paper. Reported preference, truthfulness and toxicity gains are distribution- and evaluator-dependent, and the work does not establish complete alignment or absence of harmful failures.", "area": "Instruction following", "year": 2022, "lane": "HUMAN", "result": "EMPIRICAL", "package_id": "instructgpt-2022", "target": { "record_id": "urn:uuid:810c97b7-b4ea-456f-ac79-95f2e179acd8", "version": 1, "content_hash": "sha256:5a7d532cbaf989020750d239f54333f352e6246b0ea8d263142c08c821bb78a5" }, "humans": [ "Long Ouyang", "Jeff Wu", "Xu Jiang", "Diogo Almeida", "Carroll L. Wainwright", "Pamela Mishkin", "Chong Zhang", "Sandhini Agarwal", "Katarina Slama", "Alex Ray", "John Schulman", "Jacob Hilton", "Fraser Kelton", "Luke Miller", "Maddie Simens", "Amanda Askell", "Peter Welinder", "Paul Christiano", "Jan Leike", "Ryan Lowe" ], "authors_complete": true, "human_role_note": "The named authors contributed to this scoped work; individual effort shares and later applications are not assigned by the byline.", "agents": [], "related": [], "subresults": [], "scientific_assessment": { "source_review": "SOURCE_CURATED", "scientific_review": "NOT_REVIEWED_BY_WRS", "formal_replay": "NOT_RUN", "provider_attestation": "NONE" }, "sources": [ { "id": "openai-instructgpt", "title": "OpenAI: Aligning language models to follow instructions", "url": "https://openai.com/index/instruction-following/" }, { "id": "arxiv-ouyang-instructgpt", "title": "Ouyang et al., Training language models to follow instructions with human feedback", "url": "https://arxiv.org/abs/2203.02155" } ], "points": "165000", "published_grade": { "id": "instructgpt-2022", "entry_id": "instructgpt-2022", "band": "A", "modifier": null, "target": { "record_id": "urn:uuid:810c97b7-b4ea-456f-ac79-95f2e179acd8", "version": 1, "content_hash": "sha256:5a7d532cbaf989020750d239f54333f352e6246b0ea8d263142c08c821bb78a5" }, "what_changed": "A human-feedback pipeline improved evaluated instruction-following preferences over a larger base model.", "scope": "2022 twenty-author paper and its evaluated prompt distribution; not full alignment, absence of harm, or all later chat assistants.", "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:810c97b7-b4ea-456f-ac79-95f2e179acd8", "version": 1, "content_hash": "sha256:5a7d532cbaf989020750d239f54333f352e6246b0ea8d263142c08c821bb78a5" }, "edition": "ratings-e039106f30dec177" } }, "version": { "kind": "PUBLIC_RECORD_VIEW", "previous": null, "published_at": "2026-09-25T00:00:00Z", "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": "Long Ouyang", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "OpenAI: Aligning language models to follow instructions", "url": "https://openai.com/index/instruction-following/" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Jeff Wu", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "OpenAI: Aligning language models to follow instructions", "url": "https://openai.com/index/instruction-following/" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Xu Jiang", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "OpenAI: Aligning language models to follow instructions", "url": "https://openai.com/index/instruction-following/" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Diogo Almeida", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "OpenAI: Aligning language models to follow instructions", "url": "https://openai.com/index/instruction-following/" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Carroll L. Wainwright", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "OpenAI: Aligning language models to follow instructions", "url": "https://openai.com/index/instruction-following/" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Pamela Mishkin", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "OpenAI: Aligning language models to follow instructions", "url": "https://openai.com/index/instruction-following/" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Chong Zhang", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "OpenAI: Aligning language models to follow instructions", "url": "https://openai.com/index/instruction-following/" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Sandhini Agarwal", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "OpenAI: Aligning language models to follow instructions", "url": "https://openai.com/index/instruction-following/" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Katarina Slama", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "OpenAI: Aligning language models to follow instructions", "url": "https://openai.com/index/instruction-following/" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Alex Ray", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "OpenAI: Aligning language models to follow instructions", "url": "https://openai.com/index/instruction-following/" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "John Schulman", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "OpenAI: Aligning language models to follow instructions", "url": "https://openai.com/index/instruction-following/" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Jacob Hilton", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "OpenAI: Aligning language models to follow instructions", "url": "https://openai.com/index/instruction-following/" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Fraser Kelton", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "OpenAI: Aligning language models to follow instructions", "url": "https://openai.com/index/instruction-following/" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Luke Miller", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "OpenAI: Aligning language models to follow instructions", "url": "https://openai.com/index/instruction-following/" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Maddie Simens", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "OpenAI: Aligning language models to follow instructions", "url": "https://openai.com/index/instruction-following/" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Amanda Askell", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "OpenAI: Aligning language models to follow instructions", "url": "https://openai.com/index/instruction-following/" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Peter Welinder", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "OpenAI: Aligning language models to follow instructions", "url": "https://openai.com/index/instruction-following/" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Paul Christiano", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "OpenAI: Aligning language models to follow instructions", "url": "https://openai.com/index/instruction-following/" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Jan Leike", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "OpenAI: Aligning language models to follow instructions", "url": "https://openai.com/index/instruction-following/" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Ryan Lowe", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "OpenAI: Aligning language models to follow instructions", "url": "https://openai.com/index/instruction-following/" } } ], "evidence": [ { "description": "Bibliographic/stated-result source; not a WRS proof verification", "id": "source-0", "kind": "source_reference", "source": { "citation": "OpenAI: Aligning language models to follow instructions", "url": "https://openai.com/index/instruction-following/" } }, { "description": "Bibliographic/stated-result source; not a WRS proof verification", "id": "source-1", "kind": "source_reference", "source": { "citation": "Ouyang et al., Training language models to follow instructions with human feedback", "url": "https://arxiv.org/abs/2203.02155" } } ], "origin": "HUMAN_LEGACY", "record_profile": "wrs-local-research-notes-0.1", "record_type": "research_note", "rrp_version": "0.1", "scope": { "description": "The record is the twenty-author 2022 paper. Reported preference, truthfulness and toxicity gains are distribution- and evaluator-dependent, and the work does not establish complete alignment or absence of harmful failures.", "domain": "Instruction following" }, "sources": [ { "citation": "OpenAI: Aligning language models to follow instructions", "url": "https://openai.com/index/instruction-following/" }, { "citation": "Ouyang et al., Training language models to follow instructions with human feedback", "url": "https://arxiv.org/abs/2203.02155" } ], "statement": "# Ouyang and colleagues: InstructGPT\n\nCombined supervised demonstrations, human preference rankings, a learned reward model and PPO fine-tuning to make GPT-3-derived models better follow instructions on the studied API prompt distribution.\n\n## Scientific scope\n\nThe record is the twenty-author 2022 paper. Reported preference, truthfulness and toxicity gains are distribution- and evaluator-dependent, and the work does not establish complete alignment or absence of harmful failures.\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\nLong Ouyang; Jeff Wu; Xu Jiang; Diogo Almeida; Carroll L. Wainwright; Pamela Mishkin; Chong Zhang; Sandhini Agarwal; Katarina Slama; Alex Ray; John Schulman; Jacob Hilton; Fraser Kelton; Luke Miller; Maddie Simens; Amanda Askell; Peter Welinder; Paul Christiano; Jan Leike; Ryan Lowe\n\nThe named authors contributed to this scoped work; individual effort shares and later applications are not assigned by the byline.\n\n\n\n## Original sources\n\n- [OpenAI: Aligning language models to follow instructions](https://openai.com/index/instruction-following/)\n- [Ouyang et al., Training language models to follow instructions with human feedback](https://arxiv.org/abs/2203.02155)\n", "summary": "Combined supervised demonstrations, human preference rankings, a learned reward model and PPO fine-tuning to make GPT-3-derived models better follow instructions on the studied API prompt distribution.", "title": "Ouyang and colleagues: InstructGPT" }, "record_id": "urn:uuid:810c97b7-b4ea-456f-ac79-95f2e179acd8", "version": 1, "source_ref": { "record_id": "urn:uuid:810c97b7-b4ea-456f-ac79-95f2e179acd8", "version": 1, "content_hash": "sha256:5a7d532cbaf989020750d239f54333f352e6246b0ea8d263142c08c821bb78a5" }, "export_view": "wrs-public-editorial-view-1", "public_view_hash": "sha256:cfa4cef8f9996f85b645f57a68d4f39b6f0be1368c2fcc43acd51f73bf3e5ce3" }, "target": { "record_id": "urn:uuid:810c97b7-b4ea-456f-ac79-95f2e179acd8", "version": 1, "content_hash": "sha256:5a7d532cbaf989020750d239f54333f352e6246b0ea8d263142c08c821bb78a5" }, "published_grade": { "profile": "wrs-published-grade-1", "edition": "2026-10-28.1", "id": "instructgpt-2022", "entry_id": "instructgpt-2022", "revision": 1, "title": "Ouyang and colleagues: InstructGPT", "context": "Instruction-following language models", "track": "Human-feedback fine-tuning pipeline", "band": "A", "modifier": null, "status": "PROVISIONAL_EDITORIAL", "assessed_on": "2026-09-25", "evidence_cutoff": "2026-09-25", "achievement_state": "SUPPORTED_SCOPED_RESULT", "what_changed": "A human-feedback pipeline improved evaluated instruction-following preferences over a larger base model.", "scope": "2022 twenty-author paper and its evaluated prompt distribution; not full alignment, absence of harm, or all later chat assistants.", "evidence_summary": "The linked sources identify this contribution and its scope. WRS has not independently reconstructed the scientific claims. The record is the twenty-author 2022 paper. Reported preference, truthfulness and toxicity gains are distribution- and evaluator-dependent, and the work does not establish complete alignment or absence of harmful failures.", "references": [ { "title": "Ouyang et al., Training language models to follow instructions with human feedback", "url": "https://arxiv.org/abs/2203.02155", "locator": "Source for the stated contribution; scientific claims remain bounded to the linked work." }, { "title": "OpenAI: Aligning language models to follow instructions", "url": "https://openai.com/index/instruction-following/", "locator": "Source for the stated contribution; scientific claims remain bounded to the linked work." } ], "roles": [ { "who": "The named authors listed in the linked research record", "credit": "SCIENCE", "did": "Complete source byline retained in the research record; this collective display does not merge person identities or measure labor shares." } ], "target": { "record_id": "urn:uuid:810c97b7-b4ea-456f-ac79-95f2e179acd8", "version": 1, "content_hash": "sha256:5a7d532cbaf989020750d239f54333f352e6246b0ea8d263142c08c821bb78a5" }, "scientific_correctness_certified": false, "independent_expert_calibration": false, "method": "PROPRIETARY", "points": "165000" }, "canonical_agent_submission": false, "scientific_verification_awarded": false, "reader_guide": null, "reading_status": "SOURCE_LINKED_NOTE" }