{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:ABS7DS6TK4YNNAN5AXCRUT3IDR","short_pith_number":"pith:ABS7DS6T","schema_version":"1.0","canonical_sha256":"0065f1cbd35730d681bd05c51a4f681c6fe628e5bf5dce5115cddaac5b791253","source":{"kind":"arxiv","id":"2403.18167","version":2},"attestation_state":"computed","paper":{"title":"Mechanistic Understanding and Mitigation of Language Model Non-Factual Hallucinations","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Jackie Chi Kit Cheung, Lei Yu, Meng Cao, Yue Dong","submitted_at":"2024-03-27T00:23:03Z","abstract_excerpt":"State-of-the-art language models (LMs) sometimes generate non-factual hallucinations that misalign with world knowledge. To explore the mechanistic causes of these hallucinations, we create diagnostic datasets with subject-relation queries and adapt interpretability methods to trace hallucinations through internal model representations. We discover two general and distinct mechanistic causes of hallucinations shared across LMs (Llama-2, Pythia, GPT-J): 1) knowledge enrichment hallucinations: insufficient subject attribute knowledge in lower layer MLPs, and 2) answer extraction hallucinations: "},"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":"2403.18167","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-03-27T00:23:03Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"1ff9ce30343a92845d739a60a128d9cfd3edce86229a20de67c72c6d30599ef4","abstract_canon_sha256":"516489274130027c36f3bbb9007695600295e58e7ab25d3e3f5b1978c24cc0e8"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:33:01.454131Z","signature_b64":"d5PFyxMpszD4eGex4v65+ZeX+wIxjVVur1rrt0eoHny03EsJHwuN6x6QnSASIrKbvJO35WQ5YpqOrbNYXFhBCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0065f1cbd35730d681bd05c51a4f681c6fe628e5bf5dce5115cddaac5b791253","last_reissued_at":"2026-07-05T08:33:01.453582Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:33:01.453582Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Mechanistic Understanding and Mitigation of Language Model Non-Factual Hallucinations","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Jackie Chi Kit Cheung, Lei Yu, Meng Cao, Yue Dong","submitted_at":"2024-03-27T00:23:03Z","abstract_excerpt":"State-of-the-art language models (LMs) sometimes generate non-factual hallucinations that misalign with world knowledge. To explore the mechanistic causes of these hallucinations, we create diagnostic datasets with subject-relation queries and adapt interpretability methods to trace hallucinations through internal model representations. We discover two general and distinct mechanistic causes of hallucinations shared across LMs (Llama-2, Pythia, GPT-J): 1) knowledge enrichment hallucinations: insufficient subject attribute knowledge in lower layer MLPs, and 2) answer extraction hallucinations: "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2403.18167","kind":"arxiv","version":2},"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/2403.18167/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":"2403.18167","created_at":"2026-07-05T08:33:01.453645+00:00"},{"alias_kind":"arxiv_version","alias_value":"2403.18167v2","created_at":"2026-07-05T08:33:01.453645+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2403.18167","created_at":"2026-07-05T08:33:01.453645+00:00"},{"alias_kind":"pith_short_12","alias_value":"ABS7DS6TK4YN","created_at":"2026-07-05T08:33:01.453645+00:00"},{"alias_kind":"pith_short_16","alias_value":"ABS7DS6TK4YNNAN5","created_at":"2026-07-05T08:33:01.453645+00:00"},{"alias_kind":"pith_short_8","alias_value":"ABS7DS6T","created_at":"2026-07-05T08:33:01.453645+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.07209","citing_title":"Hallucination Detection via Activations of Open-Weight Proxy Analyzers","ref_index":8,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ABS7DS6TK4YNNAN5AXCRUT3IDR","json":"https://pith.science/pith/ABS7DS6TK4YNNAN5AXCRUT3IDR.json","graph_json":"https://pith.science/api/pith-number/ABS7DS6TK4YNNAN5AXCRUT3IDR/graph.json","events_json":"https://pith.science/api/pith-number/ABS7DS6TK4YNNAN5AXCRUT3IDR/events.json","paper":"https://pith.science/paper/ABS7DS6T"},"agent_actions":{"view_html":"https://pith.science/pith/ABS7DS6TK4YNNAN5AXCRUT3IDR","download_json":"https://pith.science/pith/ABS7DS6TK4YNNAN5AXCRUT3IDR.json","view_paper":"https://pith.science/paper/ABS7DS6T","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2403.18167&json=true","fetch_graph":"https://pith.science/api/pith-number/ABS7DS6TK4YNNAN5AXCRUT3IDR/graph.json","fetch_events":"https://pith.science/api/pith-number/ABS7DS6TK4YNNAN5AXCRUT3IDR/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ABS7DS6TK4YNNAN5AXCRUT3IDR/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ABS7DS6TK4YNNAN5AXCRUT3IDR/action/storage_attestation","attest_author":"https://pith.science/pith/ABS7DS6TK4YNNAN5AXCRUT3IDR/action/author_attestation","sign_citation":"https://pith.science/pith/ABS7DS6TK4YNNAN5AXCRUT3IDR/action/citation_signature","submit_replication":"https://pith.science/pith/ABS7DS6TK4YNNAN5AXCRUT3IDR/action/replication_record"}},"created_at":"2026-07-05T08:33:01.453645+00:00","updated_at":"2026-07-05T08:33:01.453645+00:00"}