{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:3CB7RNIK6IXTPDVB54BJRTZRHF","short_pith_number":"pith:3CB7RNIK","schema_version":"1.0","canonical_sha256":"d883f8b50af22f378ea1ef0298cf31396c7eaf55e8fb3ee2b97a7b2091f6755c","source":{"kind":"arxiv","id":"2307.15054","version":4},"attestation_state":"computed","paper":{"title":"A Geometric Notion of Causal Probing","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Alexander Warstadt, Anej Svete, Cl\\'ement Guerner, Ryan Cotterell, Tianyu Liu","submitted_at":"2023-07-27T17:57:57Z","abstract_excerpt":"The linear subspace hypothesis (Bolukbasi et al., 2016) states that, in a language model's representation space, all information about a concept such as verbal number is encoded in a linear subspace. Prior work has relied on auxiliary classification tasks to identify and evaluate candidate subspaces that might give support for this hypothesis. We instead give a set of intrinsic criteria which characterize an ideal linear concept subspace and enable us to identify the subspace using only the language model distribution. Our information-theoretic framework accounts for spuriously correlated feat"},"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":"2307.15054","kind":"arxiv","version":4},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2023-07-27T17:57:57Z","cross_cats_sorted":[],"title_canon_sha256":"2ee15334a9f5d5a59f15c2dd7bca33e9fcdfc923829fb152205c64b026f79447","abstract_canon_sha256":"768dea60fcc949dd1cd98ba25476ee81483411ea0ddbe279e5fbf5ca75f102bc"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:39:15.224523Z","signature_b64":"+RswlaXm6TGRuIxQlOgxF3sInFr+YT2iJD8w2wbyo3/+kGZG3MrufPA412Z37xzsTdotwuZ9dCyDPit1udL5DA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d883f8b50af22f378ea1ef0298cf31396c7eaf55e8fb3ee2b97a7b2091f6755c","last_reissued_at":"2026-07-05T10:39:15.224012Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:39:15.224012Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"A Geometric Notion of Causal Probing","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Alexander Warstadt, Anej Svete, Cl\\'ement Guerner, Ryan Cotterell, Tianyu Liu","submitted_at":"2023-07-27T17:57:57Z","abstract_excerpt":"The linear subspace hypothesis (Bolukbasi et al., 2016) states that, in a language model's representation space, all information about a concept such as verbal number is encoded in a linear subspace. Prior work has relied on auxiliary classification tasks to identify and evaluate candidate subspaces that might give support for this hypothesis. We instead give a set of intrinsic criteria which characterize an ideal linear concept subspace and enable us to identify the subspace using only the language model distribution. Our information-theoretic framework accounts for spuriously correlated feat"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2307.15054","kind":"arxiv","version":4},"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/2307.15054/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":"2307.15054","created_at":"2026-07-05T10:39:15.224072+00:00"},{"alias_kind":"arxiv_version","alias_value":"2307.15054v4","created_at":"2026-07-05T10:39:15.224072+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2307.15054","created_at":"2026-07-05T10:39:15.224072+00:00"},{"alias_kind":"pith_short_12","alias_value":"3CB7RNIK6IXT","created_at":"2026-07-05T10:39:15.224072+00:00"},{"alias_kind":"pith_short_16","alias_value":"3CB7RNIK6IXTPDVB","created_at":"2026-07-05T10:39:15.224072+00:00"},{"alias_kind":"pith_short_8","alias_value":"3CB7RNIK","created_at":"2026-07-05T10:39:15.224072+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2406.11717","citing_title":"Refusal in Language Models Is Mediated by a Single Direction","ref_index":134,"is_internal_anchor":false},{"citing_arxiv_id":"2605.01148","citing_title":"Arithmetic in the Wild: Llama uses Base-10 Addition to Reason About Cyclic Concepts","ref_index":152,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05115","citing_title":"Manifold Steering Reveals the Shared Geometry of Neural Network Representation and Behavior","ref_index":147,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/3CB7RNIK6IXTPDVB54BJRTZRHF","json":"https://pith.science/pith/3CB7RNIK6IXTPDVB54BJRTZRHF.json","graph_json":"https://pith.science/api/pith-number/3CB7RNIK6IXTPDVB54BJRTZRHF/graph.json","events_json":"https://pith.science/api/pith-number/3CB7RNIK6IXTPDVB54BJRTZRHF/events.json","paper":"https://pith.science/paper/3CB7RNIK"},"agent_actions":{"view_html":"https://pith.science/pith/3CB7RNIK6IXTPDVB54BJRTZRHF","download_json":"https://pith.science/pith/3CB7RNIK6IXTPDVB54BJRTZRHF.json","view_paper":"https://pith.science/paper/3CB7RNIK","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2307.15054&json=true","fetch_graph":"https://pith.science/api/pith-number/3CB7RNIK6IXTPDVB54BJRTZRHF/graph.json","fetch_events":"https://pith.science/api/pith-number/3CB7RNIK6IXTPDVB54BJRTZRHF/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/3CB7RNIK6IXTPDVB54BJRTZRHF/action/timestamp_anchor","attest_storage":"https://pith.science/pith/3CB7RNIK6IXTPDVB54BJRTZRHF/action/storage_attestation","attest_author":"https://pith.science/pith/3CB7RNIK6IXTPDVB54BJRTZRHF/action/author_attestation","sign_citation":"https://pith.science/pith/3CB7RNIK6IXTPDVB54BJRTZRHF/action/citation_signature","submit_replication":"https://pith.science/pith/3CB7RNIK6IXTPDVB54BJRTZRHF/action/replication_record"}},"created_at":"2026-07-05T10:39:15.224072+00:00","updated_at":"2026-07-05T10:39:15.224072+00:00"}