{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:VRUKGRLGEF5PL5K7NMPK2XIFLK","short_pith_number":"pith:VRUKGRLG","schema_version":"1.0","canonical_sha256":"ac68a34566217af5f55f6b1ead5d055a8e4e154574e9e561da9d11fc3dce71e3","source":{"kind":"arxiv","id":"2504.14492","version":2},"attestation_state":"computed","paper":{"title":"FairSteer: Inference Time Debiasing for LLMs with Dynamic Activation Steering","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Luqi Gong, Ruizhe Chen, Xiaotang Gai, Yan Zhang, Yichen Li, Zhiting Fan, Zuozhu Liu","submitted_at":"2025-04-20T04:57:00Z","abstract_excerpt":"Large language models (LLMs) are prone to capturing biases from training corpus, leading to potential negative social impacts. Existing prompt-based debiasing methods exhibit instability due to their sensitivity to prompt changes, while fine-tuning-based techniques incur substantial computational overhead and catastrophic forgetting. In this paper, we propose FairSteer, a novel inference-time debiasing framework without requiring customized prompt design or model retraining. Motivated by the linear representation hypothesis, our preliminary investigation demonstrates that fairness-related 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":"2504.14492","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-04-20T04:57:00Z","cross_cats_sorted":[],"title_canon_sha256":"5b157a3709d1d05357d5412b243088a80fa92173edf06ff605519a982b704eac","abstract_canon_sha256":"2012518ecec543eda54474c651ed091cf4c49089be8cb49b4c06ff70a84a3b10"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:32:28.364926Z","signature_b64":"f488NOKfMVKU0dEjtRBxKlAJ95ARjHNqFfiCgDQoh13La2nn351ZZfhw6L8nFlGvQS+urLLVrjs1jJuWrcZGAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"ac68a34566217af5f55f6b1ead5d055a8e4e154574e9e561da9d11fc3dce71e3","last_reissued_at":"2026-07-05T11:32:28.364425Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:32:28.364425Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"FairSteer: Inference Time Debiasing for LLMs with Dynamic Activation Steering","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Luqi Gong, Ruizhe Chen, Xiaotang Gai, Yan Zhang, Yichen Li, Zhiting Fan, Zuozhu Liu","submitted_at":"2025-04-20T04:57:00Z","abstract_excerpt":"Large language models (LLMs) are prone to capturing biases from training corpus, leading to potential negative social impacts. Existing prompt-based debiasing methods exhibit instability due to their sensitivity to prompt changes, while fine-tuning-based techniques incur substantial computational overhead and catastrophic forgetting. In this paper, we propose FairSteer, a novel inference-time debiasing framework without requiring customized prompt design or model retraining. Motivated by the linear representation hypothesis, our preliminary investigation demonstrates that fairness-related feat"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.14492","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/2504.14492/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":"2504.14492","created_at":"2026-07-05T11:32:28.364487+00:00"},{"alias_kind":"arxiv_version","alias_value":"2504.14492v2","created_at":"2026-07-05T11:32:28.364487+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.14492","created_at":"2026-07-05T11:32:28.364487+00:00"},{"alias_kind":"pith_short_12","alias_value":"VRUKGRLGEF5P","created_at":"2026-07-05T11:32:28.364487+00:00"},{"alias_kind":"pith_short_16","alias_value":"VRUKGRLGEF5PL5K7","created_at":"2026-07-05T11:32:28.364487+00:00"},{"alias_kind":"pith_short_8","alias_value":"VRUKGRLG","created_at":"2026-07-05T11:32:28.364487+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.23040","citing_title":"Steered Generation via Gradient-Based Optimization on Sparse Query Features","ref_index":26,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/VRUKGRLGEF5PL5K7NMPK2XIFLK","json":"https://pith.science/pith/VRUKGRLGEF5PL5K7NMPK2XIFLK.json","graph_json":"https://pith.science/api/pith-number/VRUKGRLGEF5PL5K7NMPK2XIFLK/graph.json","events_json":"https://pith.science/api/pith-number/VRUKGRLGEF5PL5K7NMPK2XIFLK/events.json","paper":"https://pith.science/paper/VRUKGRLG"},"agent_actions":{"view_html":"https://pith.science/pith/VRUKGRLGEF5PL5K7NMPK2XIFLK","download_json":"https://pith.science/pith/VRUKGRLGEF5PL5K7NMPK2XIFLK.json","view_paper":"https://pith.science/paper/VRUKGRLG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2504.14492&json=true","fetch_graph":"https://pith.science/api/pith-number/VRUKGRLGEF5PL5K7NMPK2XIFLK/graph.json","fetch_events":"https://pith.science/api/pith-number/VRUKGRLGEF5PL5K7NMPK2XIFLK/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/VRUKGRLGEF5PL5K7NMPK2XIFLK/action/timestamp_anchor","attest_storage":"https://pith.science/pith/VRUKGRLGEF5PL5K7NMPK2XIFLK/action/storage_attestation","attest_author":"https://pith.science/pith/VRUKGRLGEF5PL5K7NMPK2XIFLK/action/author_attestation","sign_citation":"https://pith.science/pith/VRUKGRLGEF5PL5K7NMPK2XIFLK/action/citation_signature","submit_replication":"https://pith.science/pith/VRUKGRLGEF5PL5K7NMPK2XIFLK/action/replication_record"}},"created_at":"2026-07-05T11:32:28.364487+00:00","updated_at":"2026-07-05T11:32:28.364487+00:00"}