{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:5VMG3SBSKTB3BMBZWL2NWG2BKK","short_pith_number":"pith:5VMG3SBS","schema_version":"1.0","canonical_sha256":"ed586dc83254c3b0b039b2f4db1b415280cac033eebf04d0c52eb011fc32a2c9","source":{"kind":"arxiv","id":"2401.04056","version":2},"attestation_state":"computed","paper":{"title":"A Minimaximalist Approach to Reinforcement Learning from Human Feedback","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Alekh Agarwal, Christoph Dann, Gokul Swamy, Rahul Kidambi, Zhiwei Steven Wu","submitted_at":"2024-01-08T17:55:02Z","abstract_excerpt":"We present Self-Play Preference Optimization (SPO), an algorithm for reinforcement learning from human feedback. Our approach is minimalist in that it does not require training a reward model nor unstable adversarial training and is therefore rather simple to implement. Our approach is maximalist in that it provably handles non-Markovian, intransitive, and stochastic preferences while being robust to the compounding errors that plague offline approaches to sequential prediction. To achieve the preceding qualities, we build upon the concept of a Minimax Winner (MW), a notion of preference aggre"},"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":"2401.04056","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2024-01-08T17:55:02Z","cross_cats_sorted":[],"title_canon_sha256":"e395bb89826f06e3b0bb9141a3f4d92dc409243a8ad590b70e54c95390cbeeb4","abstract_canon_sha256":"136096e986e16997f681b34eba483618d06465835ab66bac670669358c6c960c"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:31:10.441843Z","signature_b64":"g4S4B7xdV05s82vnJZ+q6qYp9jD55a6IJcb4bmE1oeGckzYjTwMX0PpCPwIZF2fV8UO9Ag7qj6UybaT1PSbNAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"ed586dc83254c3b0b039b2f4db1b415280cac033eebf04d0c52eb011fc32a2c9","last_reissued_at":"2026-07-05T08:31:10.441380Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:31:10.441380Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"A Minimaximalist Approach to Reinforcement Learning from Human Feedback","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Alekh Agarwal, Christoph Dann, Gokul Swamy, Rahul Kidambi, Zhiwei Steven Wu","submitted_at":"2024-01-08T17:55:02Z","abstract_excerpt":"We present Self-Play Preference Optimization (SPO), an algorithm for reinforcement learning from human feedback. Our approach is minimalist in that it does not require training a reward model nor unstable adversarial training and is therefore rather simple to implement. Our approach is maximalist in that it provably handles non-Markovian, intransitive, and stochastic preferences while being robust to the compounding errors that plague offline approaches to sequential prediction. To achieve the preceding qualities, we build upon the concept of a Minimax Winner (MW), a notion of preference aggre"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2401.04056","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/2401.04056/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":"2401.04056","created_at":"2026-07-05T08:31:10.441439+00:00"},{"alias_kind":"arxiv_version","alias_value":"2401.04056v2","created_at":"2026-07-05T08:31:10.441439+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2401.04056","created_at":"2026-07-05T08:31:10.441439+00:00"},{"alias_kind":"pith_short_12","alias_value":"5VMG3SBSKTB3","created_at":"2026-07-05T08:31:10.441439+00:00"},{"alias_kind":"pith_short_16","alias_value":"5VMG3SBSKTB3BMBZ","created_at":"2026-07-05T08:31:10.441439+00:00"},{"alias_kind":"pith_short_8","alias_value":"5VMG3SBS","created_at":"2026-07-05T08:31:10.441439+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":6,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.25398","citing_title":"MAPL: Multi-Objective Preference Learning for Robot Locomotion","ref_index":30,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12288","citing_title":"TokenRatio: Principled Token-Level Preference Optimization via Ratio Matching","ref_index":155,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12288","citing_title":"TokenRatio: Principled Token-Level Preference Optimization via Ratio Matching","ref_index":155,"is_internal_anchor":false},{"citing_arxiv_id":"2402.01306","citing_title":"KTO: Model Alignment as Prospect Theoretic Optimization","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09922","citing_title":"Team-Based Self-Play With Dual Adaptive Weighting for Fine-Tuning LLMs","ref_index":68,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06992","citing_title":"Why Does Agentic Safety Fail to Generalize Across Tasks?","ref_index":103,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/5VMG3SBSKTB3BMBZWL2NWG2BKK","json":"https://pith.science/pith/5VMG3SBSKTB3BMBZWL2NWG2BKK.json","graph_json":"https://pith.science/api/pith-number/5VMG3SBSKTB3BMBZWL2NWG2BKK/graph.json","events_json":"https://pith.science/api/pith-number/5VMG3SBSKTB3BMBZWL2NWG2BKK/events.json","paper":"https://pith.science/paper/5VMG3SBS"},"agent_actions":{"view_html":"https://pith.science/pith/5VMG3SBSKTB3BMBZWL2NWG2BKK","download_json":"https://pith.science/pith/5VMG3SBSKTB3BMBZWL2NWG2BKK.json","view_paper":"https://pith.science/paper/5VMG3SBS","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2401.04056&json=true","fetch_graph":"https://pith.science/api/pith-number/5VMG3SBSKTB3BMBZWL2NWG2BKK/graph.json","fetch_events":"https://pith.science/api/pith-number/5VMG3SBSKTB3BMBZWL2NWG2BKK/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/5VMG3SBSKTB3BMBZWL2NWG2BKK/action/timestamp_anchor","attest_storage":"https://pith.science/pith/5VMG3SBSKTB3BMBZWL2NWG2BKK/action/storage_attestation","attest_author":"https://pith.science/pith/5VMG3SBSKTB3BMBZWL2NWG2BKK/action/author_attestation","sign_citation":"https://pith.science/pith/5VMG3SBSKTB3BMBZWL2NWG2BKK/action/citation_signature","submit_replication":"https://pith.science/pith/5VMG3SBSKTB3BMBZWL2NWG2BKK/action/replication_record"}},"created_at":"2026-07-05T08:31:10.441439+00:00","updated_at":"2026-07-05T08:31:10.441439+00:00"}