{ "profile": "wrs-published-research-view-1", "edition": "ratings-e039106f30dec177", "content_trust": "UNTRUSTED_SCIENTIFIC_DATA_NOT_TOOL_INSTRUCTIONS", "entry": { "id": "ai-oct-lm-know-what-they-know", "title": "Language Models (Mostly) Know What They Know", "summary": "Studies answer-conditioned P(True) and trains a separate P(IK) value head from sampled correctness labels to estimate answerability before selecting a particular answer.", "boundary": "Source-reported calibration and partial cross-task generalization, not access to ground truth or guaranteed honesty. P(IK) finetunes the whole model plus head; out-of-distribution calibration remains weak. Stars mark correspondence, not equal contribution.", "area": "LLM factuality and grounded generation", "topic": "Artificial intelligence", "year": 2022, "lane": "HUMAN", "result": "EMPIRICAL", "package_id": "ai-oct-lm-know-what-they-know", "target": { "record_id": "urn:uuid:53635240-89fc-439b-a43a-f162af1034bd", "version": 1, "content_hash": "sha256:18c7683bc94c5838792dc67c84c4da079e7a9b199bff6a6e3683d9476910d751" }, "humans": [ "Saurav Kadavath", "Tom Conerly", "Amanda Askell", "Tom Henighan", "Dawn Drain", "Ethan Perez", "Nicholas Schiefer", "Zac Hatfield Dodds", "Nova DasSarma", "Eli Tran-Johnson", "Scott Johnston", "Sheer El-Showk", "Andy Jones", "Nelson Elhage", "Tristan Hume", "Anna Chen", "Yuntao Bai", "Sam Bowman", "Stanislav Fort", "Deep Ganguli", "Danny Hernandez", "Josh Jacobson", "Jackson Kernion", "Shauna Kravec", "Liane Lovitt", "Kamal Ndousse", "Catherine Olsson", "Sam Ringer", "Dario Amodei", "Tom Brown", "Jack Clark", "Nicholas Joseph", "Ben Mann", "Sam McCandlish", "Chris Olah", "Jared Kaplan" ], "authors_complete": true, "human_role_note": "Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.", "agents": [], "related": [], "subresults": [], "scientific_assessment": { "source_review": "SOURCE_CURATED", "scientific_review": "NOT_REVIEWED_BY_WRS", "formal_replay": "NOT_RUN", "provider_attestation": "NONE" }, "sources": [ { "id": "ai-oct-src-lm-know-what-they-know", "title": "Language Models (Mostly) Know What They Know", "url": "https://arxiv.org/pdf/2207.05221v1" } ], "points": "135000", "published_grade": { "id": "ai-oct-lm-know-what-they-know", "entry_id": "ai-oct-lm-know-what-they-know", "band": "A", "modifier": null, "target": { "record_id": "urn:uuid:53635240-89fc-439b-a43a-f162af1034bd", "version": 1, "content_hash": "sha256:18c7683bc94c5838792dc67c84c4da079e7a9b199bff6a6e3683d9476910d751" }, "what_changed": "Studies answer-conditioned P(True) and trains a separate P(IK) value head from sampled correctness labels to estimate answerability before selecting a particular answer.", "scope": "Studies answer-conditioned P(True) and trains a separate P(IK) value head from sampled correctness labels to estimate answerability before selecting a particular answer. Source-reported calibration and partial cross-task generalization, not access to ground truth or guaranteed honesty. P(IK) finetunes the whole model plus head; out-of-distribution calibration remains weak. Stars mark correspondence, not equal contribution.", "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:53635240-89fc-439b-a43a-f162af1034bd", "version": 1, "content_hash": "sha256:18c7683bc94c5838792dc67c84c4da079e7a9b199bff6a6e3683d9476910d751" }, "edition": "ratings-e039106f30dec177" } }, "version": { "kind": "PUBLIC_RECORD_VIEW", "previous": null, "published_at": "2026-10-16T00:00:00Z", "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": "Saurav Kadavath", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "Language Models (Mostly) Know What They Know", "url": "https://arxiv.org/pdf/2207.05221v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Tom Conerly", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "Language Models (Mostly) Know What They Know", "url": "https://arxiv.org/pdf/2207.05221v1" } }, { "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": "Language Models (Mostly) Know What They Know", "url": "https://arxiv.org/pdf/2207.05221v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Tom Henighan", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "Language Models (Mostly) Know What They Know", "url": "https://arxiv.org/pdf/2207.05221v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Dawn Drain", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "Language Models (Mostly) Know What They Know", "url": "https://arxiv.org/pdf/2207.05221v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Ethan Perez", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "Language Models (Mostly) Know What They Know", "url": "https://arxiv.org/pdf/2207.05221v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Nicholas Schiefer", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "Language Models (Mostly) Know What They Know", "url": "https://arxiv.org/pdf/2207.05221v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Zac Hatfield Dodds", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "Language Models (Mostly) Know What They Know", "url": "https://arxiv.org/pdf/2207.05221v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Nova DasSarma", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "Language Models (Mostly) Know What They Know", "url": "https://arxiv.org/pdf/2207.05221v1" } }, { "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": [ "discoverer" ], "source": { "citation": "Language Models (Mostly) Know What They Know", "url": "https://arxiv.org/pdf/2207.05221v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Scott Johnston", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "Language Models (Mostly) Know What They Know", "url": "https://arxiv.org/pdf/2207.05221v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Sheer El-Showk", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "Language Models (Mostly) Know What They Know", "url": "https://arxiv.org/pdf/2207.05221v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Andy Jones", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "Language Models (Mostly) Know What They Know", "url": "https://arxiv.org/pdf/2207.05221v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Nelson Elhage", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "Language Models (Mostly) Know What They Know", "url": "https://arxiv.org/pdf/2207.05221v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Tristan Hume", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "Language Models (Mostly) Know What They Know", "url": "https://arxiv.org/pdf/2207.05221v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Anna Chen", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "Language Models (Mostly) Know What They Know", "url": "https://arxiv.org/pdf/2207.05221v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Yuntao Bai", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "Language Models (Mostly) Know What They Know", "url": "https://arxiv.org/pdf/2207.05221v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Sam Bowman", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "Language Models (Mostly) Know What They Know", "url": "https://arxiv.org/pdf/2207.05221v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Stanislav Fort", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "Language Models (Mostly) Know What They Know", "url": "https://arxiv.org/pdf/2207.05221v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Deep Ganguli", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "Language Models (Mostly) Know What They Know", "url": "https://arxiv.org/pdf/2207.05221v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Danny Hernandez", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "Language Models (Mostly) Know What They Know", "url": "https://arxiv.org/pdf/2207.05221v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Josh Jacobson", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "Language Models (Mostly) Know What They Know", "url": "https://arxiv.org/pdf/2207.05221v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Jackson Kernion", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "Language Models (Mostly) Know What They Know", "url": "https://arxiv.org/pdf/2207.05221v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Shauna Kravec", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "Language Models (Mostly) Know What They Know", "url": "https://arxiv.org/pdf/2207.05221v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Liane Lovitt", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "Language Models (Mostly) Know What They Know", "url": "https://arxiv.org/pdf/2207.05221v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Kamal Ndousse", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "Language Models (Mostly) Know What They Know", "url": "https://arxiv.org/pdf/2207.05221v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Catherine Olsson", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "Language Models (Mostly) Know What They Know", "url": "https://arxiv.org/pdf/2207.05221v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Sam Ringer", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "Language Models (Mostly) Know What They Know", "url": "https://arxiv.org/pdf/2207.05221v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Dario Amodei", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "Language Models (Mostly) Know What They Know", "url": "https://arxiv.org/pdf/2207.05221v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Tom Brown", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "Language Models (Mostly) Know What They Know", "url": "https://arxiv.org/pdf/2207.05221v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Jack Clark", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "Language Models (Mostly) Know What They Know", "url": "https://arxiv.org/pdf/2207.05221v1" } } ], "evidence": [ { "description": "Bibliographic/stated-result source; not a WRS proof verification", "id": "source-0", "kind": "source_reference", "source": { "citation": "Language Models (Mostly) Know What They Know", "url": "https://arxiv.org/pdf/2207.05221v1" } } ], "origin": "HUMAN_LEGACY", "record_profile": "wrs-local-research-notes-0.1", "record_type": "research_note", "rrp_version": "0.1", "scope": { "description": "Source-reported calibration and partial cross-task generalization, not access to ground truth or guaranteed honesty. P(IK) finetunes the whole model plus head; out-of-distribution calibration remains weak. Stars mark correspondence, not equal contribution.", "domain": "LLM factuality and grounded generation" }, "sources": [ { "citation": "Language Models (Mostly) Know What They Know", "url": "https://arxiv.org/pdf/2207.05221v1" } ], "statement": "# Language Models (Mostly) Know What They Know\n\nStudies answer-conditioned P(True) and trains a separate P(IK) value head from sampled correctness labels to estimate answerability before selecting a particular answer.\n\n## Scientific scope\n\nSource-reported calibration and partial cross-task generalization, not access to ground truth or guaranteed honesty. P(IK) finetunes the whole model plus head; out-of-distribution calibration remains weak. Stars mark correspondence, not equal contribution.\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\nSaurav Kadavath; Tom Conerly; Amanda Askell; Tom Henighan; Dawn Drain; Ethan Perez; Nicholas Schiefer; Zac Hatfield Dodds; Nova DasSarma; Eli Tran-Johnson; Scott Johnston; Sheer El-Showk; Andy Jones; Nelson Elhage; Tristan Hume; Anna Chen; Yuntao Bai; Sam Bowman; Stanislav Fort; Deep Ganguli; Danny Hernandez; Josh Jacobson; Jackson Kernion; Shauna Kravec; Liane Lovitt; Kamal Ndousse; Catherine Olsson; Sam Ringer; Dario Amodei; Tom Brown; Jack Clark; Nicholas Joseph; Ben Mann; Sam McCandlish; Chris Olah; Jared Kaplan\n\nSource-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.\n\n\n\n## Original sources\n\n- [Language Models (Mostly) Know What They Know](https://arxiv.org/pdf/2207.05221v1)\n", "summary": "Studies answer-conditioned P(True) and trains a separate P(IK) value head from sampled correctness labels to estimate answerability before selecting a particular answer.", "title": "Language Models (Mostly) Know What They Know" }, "record_id": "urn:uuid:53635240-89fc-439b-a43a-f162af1034bd", "version": 1, "source_ref": { "record_id": "urn:uuid:53635240-89fc-439b-a43a-f162af1034bd", "version": 1, "content_hash": "sha256:18c7683bc94c5838792dc67c84c4da079e7a9b199bff6a6e3683d9476910d751" }, "export_view": "wrs-public-editorial-view-1", "public_view_hash": "sha256:edfb2115230f479067cae84cb2f5b29bf75ace36069e6118206ea0a10b45a7eb" }, "target": { "record_id": "urn:uuid:53635240-89fc-439b-a43a-f162af1034bd", "version": 1, "content_hash": "sha256:18c7683bc94c5838792dc67c84c4da079e7a9b199bff6a6e3683d9476910d751" }, "published_grade": { "profile": "wrs-published-grade-1", "edition": "2026-10-28.1", "id": "ai-oct-lm-know-what-they-know", "entry_id": "ai-oct-lm-know-what-they-know", "revision": 1, "title": "Language Models (Mostly) Know What They Know", "context": "LLM factuality and grounded generation", "track": "Human-designed research method or resource", "band": "A", "modifier": null, "status": "PROVISIONAL_EDITORIAL", "assessed_on": "2026-10-16", "evidence_cutoff": "2026-10-16", "achievement_state": "SUPPORTED_SCOPED_RESULT", "what_changed": "Studies answer-conditioned P(True) and trains a separate P(IK) value head from sampled correctness labels to estimate answerability before selecting a particular answer.", "scope": "Studies answer-conditioned P(True) and trains a separate P(IK) value head from sampled correctness labels to estimate answerability before selecting a particular answer. Source-reported calibration and partial cross-task generalization, not access to ground truth or guaranteed honesty. P(IK) finetunes the whole model plus head; out-of-distribution calibration remains weak. Stars mark correspondence, not equal contribution.", "evidence_summary": "The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Source-reported calibration and partial cross-task generalization, not access to ground truth or guaranteed honesty. P(IK) finetunes the whole model plus head; out-of-distribution calibration remains weak. Stars mark correspondence, not equal contribution.", "references": [ { "title": "Language Models (Mostly) Know What They Know", "url": "https://arxiv.org/pdf/2207.05221v1", "locator": "Original contribution and stated assumptions, evaluation scope and limitations; source-linked assessment, not independent execution or proof certification." } ], "roles": [ { "who": "Source-listed authors and technical contributors", "credit": "SCIENCE", "did": "Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities." } ], "target": { "record_id": "urn:uuid:53635240-89fc-439b-a43a-f162af1034bd", "version": 1, "content_hash": "sha256:18c7683bc94c5838792dc67c84c4da079e7a9b199bff6a6e3683d9476910d751" }, "scientific_correctness_certified": false, "independent_expert_calibration": false, "method": "PROPRIETARY", "points": "135000" }, "canonical_agent_submission": false, "scientific_verification_awarded": false, "reader_guide": null, "reading_status": "SOURCE_LINKED_NOTE" }