{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:ZCVZ5GW2L7333MIXD2JSOBWV4R","short_pith_number":"pith:ZCVZ5GW2","schema_version":"1.0","canonical_sha256":"c8ab9e9ada5ff7bdb1171e932706d5e477364f83a3b8f879562b53e3f0e541cf","source":{"kind":"arxiv","id":"2410.18860","version":1},"attestation_state":"computed","paper":{"title":"DeCoRe: Decoding by Contrasting Retrieval Heads to Mitigate Hallucinations","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Ahmed Abdulaal, Amrutha Saseendran, Aryo Pradipta Gema, Beatrice Alex, Chen Jin, Pasquale Minervini, Philip Teare, Tom Diethe","submitted_at":"2024-10-24T15:44:33Z","abstract_excerpt":"Large Language Models (LLMs) often hallucinate, producing unfaithful or factually incorrect outputs by misrepresenting the provided context or incorrectly recalling internal knowledge. Recent studies have identified specific attention heads within the Transformer architecture, known as retrieval heads, responsible for extracting relevant contextual information. We hypothesise that masking these retrieval heads can induce hallucinations and that contrasting the outputs of the base LLM and the masked LLM can reduce hallucinations. To this end, we propose Decoding by Contrasting Retrieval Heads ("},"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":"2410.18860","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-10-24T15:44:33Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"e24cbe9b922b76509cacbb719fb5e49b25f061ea7379d2bc48b2686a17f67ee0","abstract_canon_sha256":"a2167e7a0bd588ae9aeb6dbf864a29965c530075761ca8e380fe70a7401e4b59"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:25:14.450568Z","signature_b64":"yMwtMzFTnt/Mz7Xv1Of2IgOIVH2vjK3OFFRmsCVd/P0a289Q0VOfySlCx+PXlxe8zj71aEqSOzX8pMwssWKABg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c8ab9e9ada5ff7bdb1171e932706d5e477364f83a3b8f879562b53e3f0e541cf","last_reissued_at":"2026-07-05T09:25:14.450093Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:25:14.450093Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"DeCoRe: Decoding by Contrasting Retrieval Heads to Mitigate Hallucinations","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Ahmed Abdulaal, Amrutha Saseendran, Aryo Pradipta Gema, Beatrice Alex, Chen Jin, Pasquale Minervini, Philip Teare, Tom Diethe","submitted_at":"2024-10-24T15:44:33Z","abstract_excerpt":"Large Language Models (LLMs) often hallucinate, producing unfaithful or factually incorrect outputs by misrepresenting the provided context or incorrectly recalling internal knowledge. Recent studies have identified specific attention heads within the Transformer architecture, known as retrieval heads, responsible for extracting relevant contextual information. We hypothesise that masking these retrieval heads can induce hallucinations and that contrasting the outputs of the base LLM and the masked LLM can reduce hallucinations. To this end, we propose Decoding by Contrasting Retrieval Heads ("},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.18860","kind":"arxiv","version":1},"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/2410.18860/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":"2410.18860","created_at":"2026-07-05T09:25:14.450145+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.18860v1","created_at":"2026-07-05T09:25:14.450145+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.18860","created_at":"2026-07-05T09:25:14.450145+00:00"},{"alias_kind":"pith_short_12","alias_value":"ZCVZ5GW2L733","created_at":"2026-07-05T09:25:14.450145+00:00"},{"alias_kind":"pith_short_16","alias_value":"ZCVZ5GW2L7333MIX","created_at":"2026-07-05T09:25:14.450145+00:00"},{"alias_kind":"pith_short_8","alias_value":"ZCVZ5GW2","created_at":"2026-07-05T09:25:14.450145+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.03022","citing_title":"Hallucinations as Orthogonal Noise: Inference-Time Manifold Alignment via Dynamic Contextual Orthogonalization","ref_index":48,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09492","citing_title":"APCD: Adaptive Path-Contrastive Decoding for Reliable Large Language Model Generation","ref_index":60,"is_internal_anchor":false},{"citing_arxiv_id":"2604.15789","citing_title":"A Systematic Study of Training-Free Methods for Trustworthy Large Language Models","ref_index":14,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ZCVZ5GW2L7333MIXD2JSOBWV4R","json":"https://pith.science/pith/ZCVZ5GW2L7333MIXD2JSOBWV4R.json","graph_json":"https://pith.science/api/pith-number/ZCVZ5GW2L7333MIXD2JSOBWV4R/graph.json","events_json":"https://pith.science/api/pith-number/ZCVZ5GW2L7333MIXD2JSOBWV4R/events.json","paper":"https://pith.science/paper/ZCVZ5GW2"},"agent_actions":{"view_html":"https://pith.science/pith/ZCVZ5GW2L7333MIXD2JSOBWV4R","download_json":"https://pith.science/pith/ZCVZ5GW2L7333MIXD2JSOBWV4R.json","view_paper":"https://pith.science/paper/ZCVZ5GW2","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.18860&json=true","fetch_graph":"https://pith.science/api/pith-number/ZCVZ5GW2L7333MIXD2JSOBWV4R/graph.json","fetch_events":"https://pith.science/api/pith-number/ZCVZ5GW2L7333MIXD2JSOBWV4R/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ZCVZ5GW2L7333MIXD2JSOBWV4R/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ZCVZ5GW2L7333MIXD2JSOBWV4R/action/storage_attestation","attest_author":"https://pith.science/pith/ZCVZ5GW2L7333MIXD2JSOBWV4R/action/author_attestation","sign_citation":"https://pith.science/pith/ZCVZ5GW2L7333MIXD2JSOBWV4R/action/citation_signature","submit_replication":"https://pith.science/pith/ZCVZ5GW2L7333MIXD2JSOBWV4R/action/replication_record"}},"created_at":"2026-07-05T09:25:14.450145+00:00","updated_at":"2026-07-05T09:25:14.450145+00:00"}