{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:GBFIUZKOSW43JXPVKKAI3TWJOW","short_pith_number":"pith:GBFIUZKO","schema_version":"1.0","canonical_sha256":"304a8a654e95b9b4ddf552808dcec975a017b781eeb8e26f86b5a4a46f62bf58","source":{"kind":"arxiv","id":"2408.00137","version":2},"attestation_state":"computed","paper":{"title":"Correcting Negative Bias in Large Language Models through Negative Attention Score Alignment","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Bongkyu Hwang, Hoyoung Kang, Jongyoon Song, Junhwa Choi, Sangwon Yu, Seongho Joe, Sooah Cho, Sungroh Yoon, Taehee Lee, Youngjune L. Gwon","submitted_at":"2024-07-31T19:50:57Z","abstract_excerpt":"A binary decision task, like yes-no questions or answer verification, reflects a significant real-world scenario such as where users look for confirmation about the correctness of their decisions on specific issues. In this work, we observe that language models exhibit a negative bias in the binary decisions of complex reasoning tasks. Based on our observations and the rationale about attention-based model dynamics, we propose a negative attention score (NAS) to systematically and quantitatively formulate negative bias. Based on NAS, we identify attention heads that attend to negative tokens p"},"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":"2408.00137","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-07-31T19:50:57Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"420d2b7a1b082ef17507e81021c87a9ce50dc009deb656786f2b4c2f634cb4a8","abstract_canon_sha256":"e6a25c53d59cf79f94e179bf0762107389b1009b4b1db70c37f6178db7f2cd03"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:55:22.540097Z","signature_b64":"/hRrBJSUh2ujjEiDPmtjhZu8X62eULVDckg2DL9+DrjXCDNIVXW/zGXmvwybJLIQ69YZzh2SdS1QutnsBValAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"304a8a654e95b9b4ddf552808dcec975a017b781eeb8e26f86b5a4a46f62bf58","last_reissued_at":"2026-07-05T10:55:22.539482Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:55:22.539482Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Correcting Negative Bias in Large Language Models through Negative Attention Score Alignment","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Bongkyu Hwang, Hoyoung Kang, Jongyoon Song, Junhwa Choi, Sangwon Yu, Seongho Joe, Sooah Cho, Sungroh Yoon, Taehee Lee, Youngjune L. Gwon","submitted_at":"2024-07-31T19:50:57Z","abstract_excerpt":"A binary decision task, like yes-no questions or answer verification, reflects a significant real-world scenario such as where users look for confirmation about the correctness of their decisions on specific issues. In this work, we observe that language models exhibit a negative bias in the binary decisions of complex reasoning tasks. Based on our observations and the rationale about attention-based model dynamics, we propose a negative attention score (NAS) to systematically and quantitatively formulate negative bias. Based on NAS, we identify attention heads that attend to negative tokens p"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2408.00137","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/2408.00137/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":"2408.00137","created_at":"2026-07-05T10:55:22.539564+00:00"},{"alias_kind":"arxiv_version","alias_value":"2408.00137v2","created_at":"2026-07-05T10:55:22.539564+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2408.00137","created_at":"2026-07-05T10:55:22.539564+00:00"},{"alias_kind":"pith_short_12","alias_value":"GBFIUZKOSW43","created_at":"2026-07-05T10:55:22.539564+00:00"},{"alias_kind":"pith_short_16","alias_value":"GBFIUZKOSW43JXPV","created_at":"2026-07-05T10:55:22.539564+00:00"},{"alias_kind":"pith_short_8","alias_value":"GBFIUZKO","created_at":"2026-07-05T10:55:22.539564+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.22714","citing_title":"AMEL: Accumulated Message Effects on LLM Judgments","ref_index":33,"is_internal_anchor":false},{"citing_arxiv_id":"2605.22714","citing_title":"AMEL: Accumulated Message Effects on LLM Judgments","ref_index":33,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/GBFIUZKOSW43JXPVKKAI3TWJOW","json":"https://pith.science/pith/GBFIUZKOSW43JXPVKKAI3TWJOW.json","graph_json":"https://pith.science/api/pith-number/GBFIUZKOSW43JXPVKKAI3TWJOW/graph.json","events_json":"https://pith.science/api/pith-number/GBFIUZKOSW43JXPVKKAI3TWJOW/events.json","paper":"https://pith.science/paper/GBFIUZKO"},"agent_actions":{"view_html":"https://pith.science/pith/GBFIUZKOSW43JXPVKKAI3TWJOW","download_json":"https://pith.science/pith/GBFIUZKOSW43JXPVKKAI3TWJOW.json","view_paper":"https://pith.science/paper/GBFIUZKO","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2408.00137&json=true","fetch_graph":"https://pith.science/api/pith-number/GBFIUZKOSW43JXPVKKAI3TWJOW/graph.json","fetch_events":"https://pith.science/api/pith-number/GBFIUZKOSW43JXPVKKAI3TWJOW/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/GBFIUZKOSW43JXPVKKAI3TWJOW/action/timestamp_anchor","attest_storage":"https://pith.science/pith/GBFIUZKOSW43JXPVKKAI3TWJOW/action/storage_attestation","attest_author":"https://pith.science/pith/GBFIUZKOSW43JXPVKKAI3TWJOW/action/author_attestation","sign_citation":"https://pith.science/pith/GBFIUZKOSW43JXPVKKAI3TWJOW/action/citation_signature","submit_replication":"https://pith.science/pith/GBFIUZKOSW43JXPVKKAI3TWJOW/action/replication_record"}},"created_at":"2026-07-05T10:55:22.539564+00:00","updated_at":"2026-07-05T10:55:22.539564+00:00"}