{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:CA27LYLPCGEFFCW2HWFLZMHWYB","short_pith_number":"pith:CA27LYLP","schema_version":"1.0","canonical_sha256":"1035f5e16f1188528ada3d8abcb0f6c0714a3db1e764d97dc2efd91605cf8b26","source":{"kind":"arxiv","id":"2407.16637","version":2},"attestation_state":"computed","paper":{"title":"Course-Correction: Safety Alignment Using Synthetic Preferences","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Haiqin Weng, Han Qiu, Renjie Gu, Rongwu Xu, Tianwei Zhang, Wei Xu, Yan Liu, Yishuo Cai, Zhenhong Zhou","submitted_at":"2024-07-23T16:54:28Z","abstract_excerpt":"The risk of harmful content generated by large language models (LLMs) becomes a critical concern. This paper presents a systematic study on assessing and improving LLMs' capability to perform the task of \\textbf{course-correction}, \\ie, the model can steer away from generating harmful content autonomously. To start with, we introduce the \\textsc{C$^2$-Eval} benchmark for quantitative assessment and analyze 10 popular LLMs, revealing varying proficiency of current safety-tuned LLMs in course-correction. To improve, we propose fine-tuning LLMs with preference learning, emphasizing the preference"},"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":"2407.16637","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.CL","submitted_at":"2024-07-23T16:54:28Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"f4245f96ff66064442d091a2505b5a3ad69f600440462d661edce87f7988555a","abstract_canon_sha256":"59d0268e9542f9123c66502fcd164522800ef65cc3472d3fe5fe48ce2199537e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:26:29.319459Z","signature_b64":"lpoOVWj0LVuXEfo2c6iZA3maAy9NqeVSM3VfZ9G79+TbAfJUDDPtjJfq1t5Z5iyYpHCid7lqqJSOMG9gCVVyBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"1035f5e16f1188528ada3d8abcb0f6c0714a3db1e764d97dc2efd91605cf8b26","last_reissued_at":"2026-07-05T09:26:29.318968Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:26:29.318968Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Course-Correction: Safety Alignment Using Synthetic Preferences","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Haiqin Weng, Han Qiu, Renjie Gu, Rongwu Xu, Tianwei Zhang, Wei Xu, Yan Liu, Yishuo Cai, Zhenhong Zhou","submitted_at":"2024-07-23T16:54:28Z","abstract_excerpt":"The risk of harmful content generated by large language models (LLMs) becomes a critical concern. This paper presents a systematic study on assessing and improving LLMs' capability to perform the task of \\textbf{course-correction}, \\ie, the model can steer away from generating harmful content autonomously. To start with, we introduce the \\textsc{C$^2$-Eval} benchmark for quantitative assessment and analyze 10 popular LLMs, revealing varying proficiency of current safety-tuned LLMs in course-correction. To improve, we propose fine-tuning LLMs with preference learning, emphasizing the preference"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2407.16637","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/2407.16637/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":"2407.16637","created_at":"2026-07-05T09:26:29.319026+00:00"},{"alias_kind":"arxiv_version","alias_value":"2407.16637v2","created_at":"2026-07-05T09:26:29.319026+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2407.16637","created_at":"2026-07-05T09:26:29.319026+00:00"},{"alias_kind":"pith_short_12","alias_value":"CA27LYLPCGEF","created_at":"2026-07-05T09:26:29.319026+00:00"},{"alias_kind":"pith_short_16","alias_value":"CA27LYLPCGEFFCW2","created_at":"2026-07-05T09:26:29.319026+00:00"},{"alias_kind":"pith_short_8","alias_value":"CA27LYLP","created_at":"2026-07-05T09:26:29.319026+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2505.24341","citing_title":"Exploring Multimodal Challenges in Toxic Chinese Detection: Taxonomy, Benchmark, and Findings","ref_index":34,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/CA27LYLPCGEFFCW2HWFLZMHWYB","json":"https://pith.science/pith/CA27LYLPCGEFFCW2HWFLZMHWYB.json","graph_json":"https://pith.science/api/pith-number/CA27LYLPCGEFFCW2HWFLZMHWYB/graph.json","events_json":"https://pith.science/api/pith-number/CA27LYLPCGEFFCW2HWFLZMHWYB/events.json","paper":"https://pith.science/paper/CA27LYLP"},"agent_actions":{"view_html":"https://pith.science/pith/CA27LYLPCGEFFCW2HWFLZMHWYB","download_json":"https://pith.science/pith/CA27LYLPCGEFFCW2HWFLZMHWYB.json","view_paper":"https://pith.science/paper/CA27LYLP","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2407.16637&json=true","fetch_graph":"https://pith.science/api/pith-number/CA27LYLPCGEFFCW2HWFLZMHWYB/graph.json","fetch_events":"https://pith.science/api/pith-number/CA27LYLPCGEFFCW2HWFLZMHWYB/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/CA27LYLPCGEFFCW2HWFLZMHWYB/action/timestamp_anchor","attest_storage":"https://pith.science/pith/CA27LYLPCGEFFCW2HWFLZMHWYB/action/storage_attestation","attest_author":"https://pith.science/pith/CA27LYLPCGEFFCW2HWFLZMHWYB/action/author_attestation","sign_citation":"https://pith.science/pith/CA27LYLPCGEFFCW2HWFLZMHWYB/action/citation_signature","submit_replication":"https://pith.science/pith/CA27LYLPCGEFFCW2HWFLZMHWYB/action/replication_record"}},"created_at":"2026-07-05T09:26:29.319026+00:00","updated_at":"2026-07-05T09:26:29.319026+00:00"}