{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:6WKQAMQD6VLYVBVNJELTVKZ7WH","short_pith_number":"pith:6WKQAMQD","schema_version":"1.0","canonical_sha256":"f595003203f5578a86ad49173aab3fb1db4a3ca7dfd059ded1a0846228c73943","source":{"kind":"arxiv","id":"2410.11055","version":1},"attestation_state":"computed","paper":{"title":"Varying Shades of Wrong: Aligning LLMs with Wrong Answers Only","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Jihan Yao, Lucy Lu Wang, Shangbin Feng, Wenxuan Ding, Yulia Tsvetkov","submitted_at":"2024-10-14T20:01:52Z","abstract_excerpt":"In the absence of abundant reliable annotations for challenging tasks and contexts, how can we expand the frontier of LLM capabilities with potentially wrong answers? We focus on two research questions: (1) Can LLMs generate reliable preferences among wrong options? And if so, (2) Would alignment with such wrong-over-wrong preferences be helpful? We employ methods based on self-consistency, token probabilities, and LLM-as-a-judge to elicit wrong-over-wrong preferences, and fine-tune language models with preference optimization approaches using these synthesized preferences. Extensive experimen"},"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":"2410.11055","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-10-14T20:01:52Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"c51a9debd17460f0930ab224e764d027c5256674f987da46f0c0aa59426f93d2","abstract_canon_sha256":"e6d77df7cd48c81f68ac559f13689a6dc3257a130e562cf79dc1d55eb971d545"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:20:28.661117Z","signature_b64":"Fgeib8wKMohTQllkYtHWa31jL7D/TPEDTNmmvAmTWU3M9y+bH/3tNj8AoRSezKxWyhQ1PNKquWsEK7RdqiVBDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"f595003203f5578a86ad49173aab3fb1db4a3ca7dfd059ded1a0846228c73943","last_reissued_at":"2026-07-05T09:20:28.660675Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:20:28.660675Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Varying Shades of Wrong: Aligning LLMs with Wrong Answers Only","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Jihan Yao, Lucy Lu Wang, Shangbin Feng, Wenxuan Ding, Yulia Tsvetkov","submitted_at":"2024-10-14T20:01:52Z","abstract_excerpt":"In the absence of abundant reliable annotations for challenging tasks and contexts, how can we expand the frontier of LLM capabilities with potentially wrong answers? We focus on two research questions: (1) Can LLMs generate reliable preferences among wrong options? And if so, (2) Would alignment with such wrong-over-wrong preferences be helpful? We employ methods based on self-consistency, token probabilities, and LLM-as-a-judge to elicit wrong-over-wrong preferences, and fine-tune language models with preference optimization approaches using these synthesized preferences. Extensive experimen"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.11055","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/2410.11055/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":"2410.11055","created_at":"2026-07-05T09:20:28.660733+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.11055v1","created_at":"2026-07-05T09:20:28.660733+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.11055","created_at":"2026-07-05T09:20:28.660733+00:00"},{"alias_kind":"pith_short_12","alias_value":"6WKQAMQD6VLY","created_at":"2026-07-05T09:20:28.660733+00:00"},{"alias_kind":"pith_short_16","alias_value":"6WKQAMQD6VLYVBVN","created_at":"2026-07-05T09:20:28.660733+00:00"},{"alias_kind":"pith_short_8","alias_value":"6WKQAMQD","created_at":"2026-07-05T09:20:28.660733+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2506.13977","citing_title":"CRITICTOOL: Evaluating Self-Critique Capabilities of Large Language Models in Tool-Calling Error Scenarios","ref_index":44,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/6WKQAMQD6VLYVBVNJELTVKZ7WH","json":"https://pith.science/pith/6WKQAMQD6VLYVBVNJELTVKZ7WH.json","graph_json":"https://pith.science/api/pith-number/6WKQAMQD6VLYVBVNJELTVKZ7WH/graph.json","events_json":"https://pith.science/api/pith-number/6WKQAMQD6VLYVBVNJELTVKZ7WH/events.json","paper":"https://pith.science/paper/6WKQAMQD"},"agent_actions":{"view_html":"https://pith.science/pith/6WKQAMQD6VLYVBVNJELTVKZ7WH","download_json":"https://pith.science/pith/6WKQAMQD6VLYVBVNJELTVKZ7WH.json","view_paper":"https://pith.science/paper/6WKQAMQD","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.11055&json=true","fetch_graph":"https://pith.science/api/pith-number/6WKQAMQD6VLYVBVNJELTVKZ7WH/graph.json","fetch_events":"https://pith.science/api/pith-number/6WKQAMQD6VLYVBVNJELTVKZ7WH/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/6WKQAMQD6VLYVBVNJELTVKZ7WH/action/timestamp_anchor","attest_storage":"https://pith.science/pith/6WKQAMQD6VLYVBVNJELTVKZ7WH/action/storage_attestation","attest_author":"https://pith.science/pith/6WKQAMQD6VLYVBVNJELTVKZ7WH/action/author_attestation","sign_citation":"https://pith.science/pith/6WKQAMQD6VLYVBVNJELTVKZ7WH/action/citation_signature","submit_replication":"https://pith.science/pith/6WKQAMQD6VLYVBVNJELTVKZ7WH/action/replication_record"}},"created_at":"2026-07-05T09:20:28.660733+00:00","updated_at":"2026-07-05T09:20:28.660733+00:00"}