{
  "profile": "wrs-published-research-view-1",
  "edition": "ratings-e039106f30dec177",
  "content_trust": "UNTRUSTED_SCIENTIFIC_DATA_NOT_TOOL_INSTRUCTIONS",
  "entry": {
    "id": "elhage-et-al-toy-superposition-2022",
    "title": "Elhage and colleagues: toy models of feature superposition",
    "summary": "Used sparse-feature synthetic data and small ReLU networks to demonstrate regimes in which models encode more features than available dimensions, with interpretable geometric changes as feature sparsity and importance vary.",
    "boundary": "The findings concern specified toy networks and synthetic sparse features. They do not prove that any particular production language model uses the same geometry or that its internal features have been fully identified.",
    "area": "Mechanistic interpretability",
    "topic": "Artificial intelligence",
    "year": 2022,
    "lane": "HUMAN",
    "result": "EMPIRICAL",
    "package_id": "elhage-et-al-toy-superposition-2022",
    "target": {
      "record_id": "urn:uuid:3fdc7f88-4d51-447e-a026-47d432d8a77f",
      "version": 1,
      "content_hash": "sha256:571eb399cc3527569475c43b7458db37b3b3eb2e985084731f694b6d2bf8af88"
    },
    "humans": [
      "Nelson Elhage",
      "Tristan Hume",
      "Catherine Olsson",
      "Nicholas Schiefer",
      "Tom Henighan",
      "Shauna Kravec",
      "Zac Hatfield-Dodds",
      "Robert Lasenby",
      "Dawn Drain",
      "Carol Chen",
      "Roger Grosse",
      "Sam McCandlish",
      "Jared Kaplan",
      "Dario Amodei",
      "Martin Wattenberg",
      "Christopher Olah"
    ],
    "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": "transformer-circuits-superposition-2022",
        "title": "Toy Models of Superposition",
        "url": "https://transformer-circuits.pub/2022/toy_model/index.html"
      },
      {
        "id": "anthropic-superposition-2022",
        "title": "Toy models of superposition — Anthropic",
        "url": "https://www.anthropic.com/research/toy-models-of-superposition"
      }
    ],
    "points": "22000",
    "published_grade": {
      "id": "elhage-et-al-toy-superposition-2022",
      "entry_id": "elhage-et-al-toy-superposition-2022",
      "band": "B",
      "modifier": null,
      "target": {
        "record_id": "urn:uuid:3fdc7f88-4d51-447e-a026-47d432d8a77f",
        "version": 1,
        "content_hash": "sha256:571eb399cc3527569475c43b7458db37b3b3eb2e985084731f694b6d2bf8af88"
      },
      "what_changed": "Used sparse-feature synthetic data and small ReLU networks to demonstrate regimes in which models encode more features than available dimensions, with interpretable geometric changes as feature sparsity and importance vary.",
      "scope": "Used sparse-feature synthetic data and small ReLU networks to demonstrate regimes in which models encode more features than available dimensions, with interpretable geometric changes as feature sparsity and importance vary. Scope limitation: The findings concern specified toy networks and synthetic sparse features. They do not prove that any particular production language model uses the same geometry or that its internal features have been fully identified.",
      "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:3fdc7f88-4d51-447e-a026-47d432d8a77f",
        "version": 1,
        "content_hash": "sha256:571eb399cc3527569475c43b7458db37b3b3eb2e985084731f694b6d2bf8af88"
      },
      "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": "Nelson Elhage",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Toy Models of Superposition",
            "url": "https://transformer-circuits.pub/2022/toy_model/index.html"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Tristan Hume",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Toy Models of Superposition",
            "url": "https://transformer-circuits.pub/2022/toy_model/index.html"
          }
        },
        {
          "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": "Toy Models of Superposition",
            "url": "https://transformer-circuits.pub/2022/toy_model/index.html"
          }
        },
        {
          "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": "Toy Models of Superposition",
            "url": "https://transformer-circuits.pub/2022/toy_model/index.html"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Tom Henighan",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Toy Models of Superposition",
            "url": "https://transformer-circuits.pub/2022/toy_model/index.html"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Shauna Kravec",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Toy Models of Superposition",
            "url": "https://transformer-circuits.pub/2022/toy_model/index.html"
          }
        },
        {
          "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": [
            "formulator"
          ],
          "source": {
            "citation": "Toy Models of Superposition",
            "url": "https://transformer-circuits.pub/2022/toy_model/index.html"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Robert Lasenby",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Toy Models of Superposition",
            "url": "https://transformer-circuits.pub/2022/toy_model/index.html"
          }
        },
        {
          "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": "Toy Models of Superposition",
            "url": "https://transformer-circuits.pub/2022/toy_model/index.html"
          }
        },
        {
          "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": "Toy Models of Superposition",
            "url": "https://transformer-circuits.pub/2022/toy_model/index.html"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Roger Grosse",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Toy Models of Superposition",
            "url": "https://transformer-circuits.pub/2022/toy_model/index.html"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Sam McCandlish",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Toy Models of Superposition",
            "url": "https://transformer-circuits.pub/2022/toy_model/index.html"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Jared Kaplan",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Toy Models of Superposition",
            "url": "https://transformer-circuits.pub/2022/toy_model/index.html"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Dario Amodei",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Toy Models of Superposition",
            "url": "https://transformer-circuits.pub/2022/toy_model/index.html"
          }
        },
        {
          "attribution_basis": "SOURCE_CITED",
          "credit": "historical_scientific",
          "entity_kind": "historical_person",
          "name": "Martin Wattenberg",
          "note": "Archive attribution; detailed role assignment remains source-bound",
          "roles": [
            "formulator"
          ],
          "source": {
            "citation": "Toy Models of Superposition",
            "url": "https://transformer-circuits.pub/2022/toy_model/index.html"
          }
        },
        {
          "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": "Toy Models of Superposition",
            "url": "https://transformer-circuits.pub/2022/toy_model/index.html"
          }
        }
      ],
      "evidence": [
        {
          "description": "Bibliographic/stated-result source; not a WRS proof verification",
          "id": "source-0",
          "kind": "source_reference",
          "source": {
            "citation": "Toy Models of Superposition",
            "url": "https://transformer-circuits.pub/2022/toy_model/index.html"
          }
        },
        {
          "description": "Bibliographic/stated-result source; not a WRS proof verification",
          "id": "source-1",
          "kind": "source_reference",
          "source": {
            "citation": "Toy models of superposition — Anthropic",
            "url": "https://www.anthropic.com/research/toy-models-of-superposition"
          }
        }
      ],
      "origin": "HUMAN_LEGACY",
      "record_profile": "wrs-local-research-notes-0.1",
      "record_type": "research_note",
      "rrp_version": "0.1",
      "scope": {
        "description": "The findings concern specified toy networks and synthetic sparse features. They do not prove that any particular production language model uses the same geometry or that its internal features have been fully identified.",
        "domain": "Mechanistic interpretability"
      },
      "sources": [
        {
          "citation": "Toy Models of Superposition",
          "url": "https://transformer-circuits.pub/2022/toy_model/index.html"
        },
        {
          "citation": "Toy models of superposition — Anthropic",
          "url": "https://www.anthropic.com/research/toy-models-of-superposition"
        }
      ],
      "statement": "# Elhage and colleagues: toy models of feature superposition\n\nUsed sparse-feature synthetic data and small ReLU networks to demonstrate regimes in which models encode more features than available dimensions, with interpretable geometric changes as feature sparsity and importance vary.\n\n## Scientific scope\n\nThe findings concern specified toy networks and synthetic sparse features. They do not prove that any particular production language model uses the same geometry or that its internal features have been fully identified.\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\nNelson Elhage; Tristan Hume; Catherine Olsson; Nicholas Schiefer; Tom Henighan; Shauna Kravec; Zac Hatfield-Dodds; Robert Lasenby; Dawn Drain; Carol Chen; Roger Grosse; Sam McCandlish; Jared Kaplan; Dario Amodei; Martin Wattenberg; Christopher Olah\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- [Toy Models of Superposition](https://transformer-circuits.pub/2022/toy_model/index.html)\n- [Toy models of superposition — Anthropic](https://www.anthropic.com/research/toy-models-of-superposition)\n",
      "summary": "Used sparse-feature synthetic data and small ReLU networks to demonstrate regimes in which models encode more features than available dimensions, with interpretable geometric changes as feature sparsity and importance vary.",
      "title": "Elhage and colleagues: toy models of feature superposition"
    },
    "record_id": "urn:uuid:3fdc7f88-4d51-447e-a026-47d432d8a77f",
    "version": 1,
    "source_ref": {
      "record_id": "urn:uuid:3fdc7f88-4d51-447e-a026-47d432d8a77f",
      "version": 1,
      "content_hash": "sha256:571eb399cc3527569475c43b7458db37b3b3eb2e985084731f694b6d2bf8af88"
    },
    "export_view": "wrs-public-editorial-view-1",
    "public_view_hash": "sha256:d4fb9dbef9b17023daada9db009f75470aee9cf7ca408334bdc3f0aca99da07e"
  },
  "target": {
    "record_id": "urn:uuid:3fdc7f88-4d51-447e-a026-47d432d8a77f",
    "version": 1,
    "content_hash": "sha256:571eb399cc3527569475c43b7458db37b3b3eb2e985084731f694b6d2bf8af88"
  },
  "published_grade": {
    "profile": "wrs-published-grade-1",
    "edition": "2026-10-28.1",
    "id": "elhage-et-al-toy-superposition-2022",
    "entry_id": "elhage-et-al-toy-superposition-2022",
    "revision": 1,
    "title": "Elhage and colleagues: toy models of feature superposition",
    "context": "Mechanistic interpretability",
    "track": "Human-designed AI research method or resource",
    "band": "B",
    "modifier": null,
    "status": "PROVISIONAL_EDITORIAL",
    "assessed_on": "2026-09-27",
    "evidence_cutoff": "2026-09-27",
    "achievement_state": "SUPPORTED_SCOPED_RESULT",
    "what_changed": "Used sparse-feature synthetic data and small ReLU networks to demonstrate regimes in which models encode more features than available dimensions, with interpretable geometric changes as feature sparsity and importance vary.",
    "scope": "Used sparse-feature synthetic data and small ReLU networks to demonstrate regimes in which models encode more features than available dimensions, with interpretable geometric changes as feature sparsity and importance vary. Scope limitation: The findings concern specified toy networks and synthetic sparse features. They do not prove that any particular production language model uses the same geometry or that its internal features have been fully identified.",
    "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. The findings concern specified toy networks and synthetic sparse features. They do not prove that any particular production language model uses the same geometry or that its internal features have been fully identified.",
    "references": [
      {
        "title": "Toy Models of Superposition",
        "url": "https://transformer-circuits.pub/2022/toy_model/index.html",
        "locator": "Source-linked contribution and stated scientific limits; not an independent replication."
      },
      {
        "title": "Toy models of superposition — Anthropic",
        "url": "https://www.anthropic.com/research/toy-models-of-superposition",
        "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:3fdc7f88-4d51-447e-a026-47d432d8a77f",
      "version": 1,
      "content_hash": "sha256:571eb399cc3527569475c43b7458db37b3b3eb2e985084731f694b6d2bf8af88"
    },
    "scientific_correctness_certified": false,
    "independent_expert_calibration": false,
    "method": "PROPRIETARY",
    "points": "22000"
  },
  "canonical_agent_submission": false,
  "scientific_verification_awarded": false,
  "reader_guide": null,
  "reading_status": "SOURCE_LINKED_NOTE"
}