{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:AVTCXNBDPZQISE4EUZQOGUZ33L","short_pith_number":"pith:AVTCXNBD","schema_version":"1.0","canonical_sha256":"05662bb4237e60891384a660e3533bdaecae3d085ec519627651a61451b86686","source":{"kind":"arxiv","id":"2501.12735","version":3},"attestation_state":"computed","paper":{"title":"Online Preference Alignment for Language Models via Count-based Exploration","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Chenjia Bai, Kang Xu, Qiaosheng Zhang, Shuang Qiu, Xuelong Li, Yang Zhang","submitted_at":"2025-01-22T09:12:09Z","abstract_excerpt":"Reinforcement Learning from Human Feedback (RLHF) has shown great potential in fine-tuning Large Language Models (LLMs) to align with human preferences. Existing methods perform preference alignment from a fixed dataset, which can be limited in data coverage, and the resulting reward model is hard to generalize in out-of-distribution responses. Thus, online RLHF is more desirable to empower the LLM to explore outside the support of the initial dataset by iteratively collecting the prompt-response pairs. In this paper, we study the fundamental problem in online RLHF, i.e. \\emph{how to explore} "},"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":"2501.12735","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-01-22T09:12:09Z","cross_cats_sorted":[],"title_canon_sha256":"431a30181d6d15ef5b1191220937b0ffc36ca9ea08ca5f5878d4c7600941240c","abstract_canon_sha256":"be9c73548f9809473bf4ee8956632a097fae515bc8c85356b121bfd0a5534f7b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:10:48.099384Z","signature_b64":"l0niKyVG5KX3P3rT1AcilDQSjtfBITjSyXlRSjc6G0/Wyor6yIQqh4hq5D+anrpoARLHRZWAxEs2fWGD7oRuDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"05662bb4237e60891384a660e3533bdaecae3d085ec519627651a61451b86686","last_reissued_at":"2026-07-05T10:10:48.098897Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:10:48.098897Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Online Preference Alignment for Language Models via Count-based Exploration","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Chenjia Bai, Kang Xu, Qiaosheng Zhang, Shuang Qiu, Xuelong Li, Yang Zhang","submitted_at":"2025-01-22T09:12:09Z","abstract_excerpt":"Reinforcement Learning from Human Feedback (RLHF) has shown great potential in fine-tuning Large Language Models (LLMs) to align with human preferences. Existing methods perform preference alignment from a fixed dataset, which can be limited in data coverage, and the resulting reward model is hard to generalize in out-of-distribution responses. Thus, online RLHF is more desirable to empower the LLM to explore outside the support of the initial dataset by iteratively collecting the prompt-response pairs. In this paper, we study the fundamental problem in online RLHF, i.e. \\emph{how to explore} "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2501.12735","kind":"arxiv","version":3},"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/2501.12735/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":"2501.12735","created_at":"2026-07-05T10:10:48.098954+00:00"},{"alias_kind":"arxiv_version","alias_value":"2501.12735v3","created_at":"2026-07-05T10:10:48.098954+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2501.12735","created_at":"2026-07-05T10:10:48.098954+00:00"},{"alias_kind":"pith_short_12","alias_value":"AVTCXNBDPZQI","created_at":"2026-07-05T10:10:48.098954+00:00"},{"alias_kind":"pith_short_16","alias_value":"AVTCXNBDPZQISE4E","created_at":"2026-07-05T10:10:48.098954+00:00"},{"alias_kind":"pith_short_8","alias_value":"AVTCXNBD","created_at":"2026-07-05T10:10:48.098954+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.10768","citing_title":"N-GRPO: Embedding-Level Neighbor Mixing for Enhanced Policy Optimization","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2606.06080","citing_title":"On Advantage Estimates for Max@K Policy Gradients","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2605.31455","citing_title":"DRIFT: Decoupled Rollouts and Importance-Weighted Fine-Tuning for Efficient Multi-Turn Optimization","ref_index":1,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/AVTCXNBDPZQISE4EUZQOGUZ33L","json":"https://pith.science/pith/AVTCXNBDPZQISE4EUZQOGUZ33L.json","graph_json":"https://pith.science/api/pith-number/AVTCXNBDPZQISE4EUZQOGUZ33L/graph.json","events_json":"https://pith.science/api/pith-number/AVTCXNBDPZQISE4EUZQOGUZ33L/events.json","paper":"https://pith.science/paper/AVTCXNBD"},"agent_actions":{"view_html":"https://pith.science/pith/AVTCXNBDPZQISE4EUZQOGUZ33L","download_json":"https://pith.science/pith/AVTCXNBDPZQISE4EUZQOGUZ33L.json","view_paper":"https://pith.science/paper/AVTCXNBD","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2501.12735&json=true","fetch_graph":"https://pith.science/api/pith-number/AVTCXNBDPZQISE4EUZQOGUZ33L/graph.json","fetch_events":"https://pith.science/api/pith-number/AVTCXNBDPZQISE4EUZQOGUZ33L/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/AVTCXNBDPZQISE4EUZQOGUZ33L/action/timestamp_anchor","attest_storage":"https://pith.science/pith/AVTCXNBDPZQISE4EUZQOGUZ33L/action/storage_attestation","attest_author":"https://pith.science/pith/AVTCXNBDPZQISE4EUZQOGUZ33L/action/author_attestation","sign_citation":"https://pith.science/pith/AVTCXNBDPZQISE4EUZQOGUZ33L/action/citation_signature","submit_replication":"https://pith.science/pith/AVTCXNBDPZQISE4EUZQOGUZ33L/action/replication_record"}},"created_at":"2026-07-05T10:10:48.098954+00:00","updated_at":"2026-07-05T10:10:48.098954+00:00"}