{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:KPETSIRAXMZMTZJDCSQQOTIZFZ","short_pith_number":"pith:KPETSIRA","schema_version":"1.0","canonical_sha256":"53c9392220bb32c9e52314a1074d192e5982f94f9f08f9ea8f2e31bafb37cc1a","source":{"kind":"arxiv","id":"2404.09824","version":1},"attestation_state":"computed","paper":{"title":"Impact of Preference Noise on the Alignment Performance of Generative Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Dana Alon, Donald Metzler, Yang Gao","submitted_at":"2024-04-15T14:21:53Z","abstract_excerpt":"A key requirement in developing Generative Language Models (GLMs) is to have their values aligned with human values. Preference-based alignment is a widely used paradigm for this purpose, in which preferences over generation pairs are first elicited from human annotators or AI systems, and then fed into some alignment techniques, e.g., Direct Preference Optimization. However, a substantial percent (20 - 40%) of the preference pairs used in GLM alignment are noisy, and it remains unclear how the noise affects the alignment performance and how to mitigate its negative impact. In this paper, we 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":"2404.09824","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-04-15T14:21:53Z","cross_cats_sorted":[],"title_canon_sha256":"b0471ea742761a6f4342277d73621871db37d90197a27f7a2ca355077f74f42e","abstract_canon_sha256":"aa13bf4a0b4f60b75a129f071954ae66dfd9734b0308213b641c2952ccbd9462"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:08:12.222295Z","signature_b64":"AbmyVWvL8oGqt4mMOVN997HJ55Lt6Dw1bkZsJMXn9PCcKQyphcrmm7H1ZuqhoTFg9HEbJlFiXHu8Kg862BP/Dg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"53c9392220bb32c9e52314a1074d192e5982f94f9f08f9ea8f2e31bafb37cc1a","last_reissued_at":"2026-07-05T08:08:12.221932Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:08:12.221932Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Impact of Preference Noise on the Alignment Performance of Generative Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Dana Alon, Donald Metzler, Yang Gao","submitted_at":"2024-04-15T14:21:53Z","abstract_excerpt":"A key requirement in developing Generative Language Models (GLMs) is to have their values aligned with human values. Preference-based alignment is a widely used paradigm for this purpose, in which preferences over generation pairs are first elicited from human annotators or AI systems, and then fed into some alignment techniques, e.g., Direct Preference Optimization. However, a substantial percent (20 - 40%) of the preference pairs used in GLM alignment are noisy, and it remains unclear how the noise affects the alignment performance and how to mitigate its negative impact. In this paper, we p"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2404.09824","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/2404.09824/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":"2404.09824","created_at":"2026-07-05T08:08:12.221988+00:00"},{"alias_kind":"arxiv_version","alias_value":"2404.09824v1","created_at":"2026-07-05T08:08:12.221988+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2404.09824","created_at":"2026-07-05T08:08:12.221988+00:00"},{"alias_kind":"pith_short_12","alias_value":"KPETSIRAXMZM","created_at":"2026-07-05T08:08:12.221988+00:00"},{"alias_kind":"pith_short_16","alias_value":"KPETSIRAXMZMTZJD","created_at":"2026-07-05T08:08:12.221988+00:00"},{"alias_kind":"pith_short_8","alias_value":"KPETSIRA","created_at":"2026-07-05T08:08:12.221988+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":6,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2607.00531","citing_title":"Active-GRPO: Adaptive Imitation and Self-Improving Reasoning for Molecular Optimization","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2502.06387","citing_title":"How Humans Help LLMs: Assessing and Incentivizing Human Preference Annotators","ref_index":37,"is_internal_anchor":false},{"citing_arxiv_id":"2602.07340","citing_title":"Revisiting Robustness for LLM Safety Alignment via Selective Geometry Control","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2508.04149","citing_title":"Difficulty-Based Preference Data Selection by DPO Implicit Reward Gap","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2505.19134","citing_title":"Incentivizing High-Quality Human Annotations with Golden Questions","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2510.13830","citing_title":"Users as Annotators: LLM Preference Learning from Comparison Mode","ref_index":10,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/KPETSIRAXMZMTZJDCSQQOTIZFZ","json":"https://pith.science/pith/KPETSIRAXMZMTZJDCSQQOTIZFZ.json","graph_json":"https://pith.science/api/pith-number/KPETSIRAXMZMTZJDCSQQOTIZFZ/graph.json","events_json":"https://pith.science/api/pith-number/KPETSIRAXMZMTZJDCSQQOTIZFZ/events.json","paper":"https://pith.science/paper/KPETSIRA"},"agent_actions":{"view_html":"https://pith.science/pith/KPETSIRAXMZMTZJDCSQQOTIZFZ","download_json":"https://pith.science/pith/KPETSIRAXMZMTZJDCSQQOTIZFZ.json","view_paper":"https://pith.science/paper/KPETSIRA","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2404.09824&json=true","fetch_graph":"https://pith.science/api/pith-number/KPETSIRAXMZMTZJDCSQQOTIZFZ/graph.json","fetch_events":"https://pith.science/api/pith-number/KPETSIRAXMZMTZJDCSQQOTIZFZ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/KPETSIRAXMZMTZJDCSQQOTIZFZ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/KPETSIRAXMZMTZJDCSQQOTIZFZ/action/storage_attestation","attest_author":"https://pith.science/pith/KPETSIRAXMZMTZJDCSQQOTIZFZ/action/author_attestation","sign_citation":"https://pith.science/pith/KPETSIRAXMZMTZJDCSQQOTIZFZ/action/citation_signature","submit_replication":"https://pith.science/pith/KPETSIRAXMZMTZJDCSQQOTIZFZ/action/replication_record"}},"created_at":"2026-07-05T08:08:12.221988+00:00","updated_at":"2026-07-05T08:08:12.221988+00:00"}