{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:GYKAY2MLTSKP6HAQITVTNXYROU","short_pith_number":"pith:GYKAY2ML","schema_version":"1.0","canonical_sha256":"36140c698b9c94ff1c1044eb36df11752cb66150b1eb30e6357fb4f4fff3d17e","source":{"kind":"arxiv","id":"2410.04612","version":2},"attestation_state":"computed","paper":{"title":"Regressing the Relative Future: Efficient Policy Optimization for Multi-turn RLHF","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Gokul Swamy, Jason D. Lee, Jonathan D. Chang, Kiant\\'e Brantley, Wenhao Zhan, Wen Sun, Zhaolin Gao","submitted_at":"2024-10-06T20:20:22Z","abstract_excerpt":"Large Language Models (LLMs) have achieved remarkable success at tasks like summarization that involve a single turn of interaction. However, they can still struggle with multi-turn tasks like dialogue that require long-term planning. Previous works on multi-turn dialogue extend single-turn reinforcement learning from human feedback (RLHF) methods to the multi-turn setting by treating all prior dialogue turns as a long context. Such approaches suffer from covariate shift: the conversations in the training set have previous turns generated by some reference policy, which means that low training"},"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":"2410.04612","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-10-06T20:20:22Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"a6082061c76948768760f4148193f093167937b667303dd1197ce53aee979245","abstract_canon_sha256":"da9dbcfdd75d374eb8400374dec4c1089dd9e5b9d445a1c45bb36b6614c10c55"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:53:12.284326Z","signature_b64":"D7iJfzHMAKHWTajQuxez2Lh/vSdQojlmSjDHGgwk60vMPHRV5WbAXNNjYUASmWYF+9UgbQjTeKLgcCJ5TbgsDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"36140c698b9c94ff1c1044eb36df11752cb66150b1eb30e6357fb4f4fff3d17e","last_reissued_at":"2026-07-05T10:53:12.283834Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:53:12.283834Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Regressing the Relative Future: Efficient Policy Optimization for Multi-turn RLHF","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Gokul Swamy, Jason D. Lee, Jonathan D. Chang, Kiant\\'e Brantley, Wenhao Zhan, Wen Sun, Zhaolin Gao","submitted_at":"2024-10-06T20:20:22Z","abstract_excerpt":"Large Language Models (LLMs) have achieved remarkable success at tasks like summarization that involve a single turn of interaction. However, they can still struggle with multi-turn tasks like dialogue that require long-term planning. Previous works on multi-turn dialogue extend single-turn reinforcement learning from human feedback (RLHF) methods to the multi-turn setting by treating all prior dialogue turns as a long context. Such approaches suffer from covariate shift: the conversations in the training set have previous turns generated by some reference policy, which means that low training"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.04612","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/2410.04612/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":"2410.04612","created_at":"2026-07-05T10:53:12.283892+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.04612v2","created_at":"2026-07-05T10:53:12.283892+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.04612","created_at":"2026-07-05T10:53:12.283892+00:00"},{"alias_kind":"pith_short_12","alias_value":"GYKAY2MLTSKP","created_at":"2026-07-05T10:53:12.283892+00:00"},{"alias_kind":"pith_short_16","alias_value":"GYKAY2MLTSKP6HAQ","created_at":"2026-07-05T10:53:12.283892+00:00"},{"alias_kind":"pith_short_8","alias_value":"GYKAY2ML","created_at":"2026-07-05T10:53:12.283892+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.29256","citing_title":"DynSess: Dynamic Session-Level Evaluation and Optimization Framework for Role-Playing Agents","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2510.21060","citing_title":"On the Sample Complexity of Differentially Private Policy Optimization","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09808","citing_title":"Quantifying the Utility of User Simulators for Building Collaborative LLM Assistants","ref_index":10,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/GYKAY2MLTSKP6HAQITVTNXYROU","json":"https://pith.science/pith/GYKAY2MLTSKP6HAQITVTNXYROU.json","graph_json":"https://pith.science/api/pith-number/GYKAY2MLTSKP6HAQITVTNXYROU/graph.json","events_json":"https://pith.science/api/pith-number/GYKAY2MLTSKP6HAQITVTNXYROU/events.json","paper":"https://pith.science/paper/GYKAY2ML"},"agent_actions":{"view_html":"https://pith.science/pith/GYKAY2MLTSKP6HAQITVTNXYROU","download_json":"https://pith.science/pith/GYKAY2MLTSKP6HAQITVTNXYROU.json","view_paper":"https://pith.science/paper/GYKAY2ML","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.04612&json=true","fetch_graph":"https://pith.science/api/pith-number/GYKAY2MLTSKP6HAQITVTNXYROU/graph.json","fetch_events":"https://pith.science/api/pith-number/GYKAY2MLTSKP6HAQITVTNXYROU/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/GYKAY2MLTSKP6HAQITVTNXYROU/action/timestamp_anchor","attest_storage":"https://pith.science/pith/GYKAY2MLTSKP6HAQITVTNXYROU/action/storage_attestation","attest_author":"https://pith.science/pith/GYKAY2MLTSKP6HAQITVTNXYROU/action/author_attestation","sign_citation":"https://pith.science/pith/GYKAY2MLTSKP6HAQITVTNXYROU/action/citation_signature","submit_replication":"https://pith.science/pith/GYKAY2MLTSKP6HAQITVTNXYROU/action/replication_record"}},"created_at":"2026-07-05T10:53:12.283892+00:00","updated_at":"2026-07-05T10:53:12.283892+00:00"}