{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2023:AD4OU3S57S5Y2GK4FCGKHPM5GD","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"e06fbc5dabdd57615fbf708dac24235e084d3e237bd3abd328f2bc19edfd90ee","cross_cats_sorted":["cs.AI","cs.LG"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2023-09-01T05:53:33Z","title_canon_sha256":"30a71bef573f4df6e33a747f2c8824790bde2c0d53c2ef6779f1df973bc3eb36"},"schema_version":"1.0","source":{"id":"2309.00267","kind":"arxiv","version":3}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2309.00267","created_at":"2026-05-17T23:38:50Z"},{"alias_kind":"arxiv_version","alias_value":"2309.00267v3","created_at":"2026-05-17T23:38:50Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2309.00267","created_at":"2026-05-17T23:38:50Z"},{"alias_kind":"pith_short_12","alias_value":"AD4OU3S57S5Y","created_at":"2026-05-18T12:33:33Z"},{"alias_kind":"pith_short_16","alias_value":"AD4OU3S57S5Y2GK4","created_at":"2026-05-18T12:33:33Z"},{"alias_kind":"pith_short_8","alias_value":"AD4OU3S5","created_at":"2026-05-18T12:33:33Z"}],"graph_snapshots":[{"event_id":"sha256:d1880872e90a02ee00460368c31bc3b1ca2d930b2b6d150d04de2ef0f860e701","target":"graph","created_at":"2026-05-17T23:38:50Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"Across the tasks of summarization, helpful dialogue generation, and harmless dialogue generation, we show that RLAIF achieves comparable performance to RLHF. ... we introduce direct-RLAIF (d-RLAIF) ... which achieves superior performance to canonical RLAIF."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the preferences generated by an off-the-shelf LLM are high-quality enough to serve as a substitute for human preferences in training the reward model."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"RLAIF matches RLHF on summarization and dialogue tasks, with a direct-RLAIF variant achieving superior results by using LLM rewards directly during training."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Reinforcement learning from AI feedback matches human feedback performance for aligning large language models."}],"snapshot_sha256":"4ddbd0027246b56fe7179d811e1ce883159e5134259fd24be22f93395f94c2b3"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"4ea79e99becc6383a8dc457d14fef873224bf8b76f834a6ee459b0fefd95c6cb"},"paper":{"abstract_excerpt":"Reinforcement learning from human feedback (RLHF) has proven effective in aligning large language models (LLMs) with human preferences, but gathering high-quality preference labels is expensive. RL from AI Feedback (RLAIF), introduced in Bai et al., offers a promising alternative that trains the reward model (RM) on preferences generated by an off-the-shelf LLM. Across the tasks of summarization, helpful dialogue generation, and harmless dialogue generation, we show that RLAIF achieves comparable performance to RLHF. Furthermore, we take a step towards \"self-improvement\" by demonstrating that ","authors_text":"Abhinav Rastogi, Colton Bishop, Ethan Hall, Harrison Lee, Hassan Mansoor, Johan Ferret, Kellie Lu, Samrat Phatale, Sushant Prakash, Thomas Mesnard, Victor Carbune","cross_cats":["cs.AI","cs.LG"],"headline":"Reinforcement learning from AI feedback matches human feedback performance for aligning large language models.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2023-09-01T05:53:33Z","title":"RLAIF vs. RLHF: Scaling Reinforcement Learning from Human Feedback with AI Feedback"},"references":{"count":98,"internal_anchors":15,"resolved_work":98,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"E., Fort, S., Lanham, T., Telleen-Lawton, T., Conerly, T., Henighan, T., Hume, T., Bowman, S","work_id":"686a699b-d7b5-4f10-ba6c-e3df30418d80","year":2022},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"D., Dhariwal, P., Neelakantan, A., Shyam, P., Sastry, G., Askell, A., et al","work_id":"fb006848-e99b-481f-9340-c35dbff67d47","year":1901},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":6,"title":"F., Leike, J., Brown, T., Martic, M., Legg, S., and Amodei, D","work_id":"463eef82-ca77-4ce1-8588-f7c3f7abe5a4","year":2017},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":8,"title":"RAFT : Reward ranked finetuning for generative foundation model alignment","work_id":"aa80a8cf-c1b9-4e19-bb61-6c9aea78c183","year":2023},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":9,"title":"Understanding dataset difficulty with V -usable information","work_id":"8b66d63b-71d0-445b-8db1-ceaecd3c2e1f","year":2022}],"snapshot_sha256":"280efcd0accaebc769d975a017e5a7572be47b99c701fdd06e17e462155ec766"},"source":{"id":"2309.00267","kind":"arxiv","version":3},"verdict":{"created_at":"2026-05-15T21:29:18.481786Z","id":"a3f1a224-083d-4a40-8281-981bb014d035","model_set":{"reader":"grok-4.3"},"one_line_summary":"RLAIF matches RLHF on summarization and dialogue tasks, with a direct-RLAIF variant achieving superior results by using LLM rewards directly during training.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Reinforcement learning from AI feedback matches human feedback performance for aligning large language models.","strongest_claim":"Across the tasks of summarization, helpful dialogue generation, and harmless dialogue generation, we show that RLAIF achieves comparable performance to RLHF. ... we introduce direct-RLAIF (d-RLAIF) ... which achieves superior performance to canonical RLAIF.","weakest_assumption":"That the preferences generated by an off-the-shelf LLM are high-quality enough to serve as a substitute for human preferences in training the reward model."}},"verdict_id":"a3f1a224-083d-4a40-8281-981bb014d035"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:46fd3d00a822dfc47773dbafe32daf3ccf3ec0d26c2e6a02ab0a3f17d2c75597","target":"record","created_at":"2026-05-17T23:38:50Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"e06fbc5dabdd57615fbf708dac24235e084d3e237bd3abd328f2bc19edfd90ee","cross_cats_sorted":["cs.AI","cs.LG"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2023-09-01T05:53:33Z","title_canon_sha256":"30a71bef573f4df6e33a747f2c8824790bde2c0d53c2ef6779f1df973bc3eb36"},"schema_version":"1.0","source":{"id":"2309.00267","kind":"arxiv","version":3}},"canonical_sha256":"00f8ea6e5dfcbb8d195c288ca3bd9d30ccec365083e1091ebe19ac2b0a61252f","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"00f8ea6e5dfcbb8d195c288ca3bd9d30ccec365083e1091ebe19ac2b0a61252f","first_computed_at":"2026-05-17T23:38:50.098142Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-17T23:38:50.098142Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"gdaJTPx4/wzV4nWtoCfQEipK/U0PyCM5yZ9HCCKZEjejR3+EW750giZDiWOjTDtKsFZ322HxChSXhVIof+9zAw==","signature_status":"signed_v1","signed_at":"2026-05-17T23:38:50.098761Z","signed_message":"canonical_sha256_bytes"},"source_id":"2309.00267","source_kind":"arxiv","source_version":3}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:46fd3d00a822dfc47773dbafe32daf3ccf3ec0d26c2e6a02ab0a3f17d2c75597","sha256:d1880872e90a02ee00460368c31bc3b1ca2d930b2b6d150d04de2ef0f860e701"],"state_sha256":"23ec69158f02ee21326d9aba879e57f200dc0114ecc60a45c0e02c8b47163998"}