{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:MRW6BJ2IGHQXBYNXL5QJ2TJPCQ","short_pith_number":"pith:MRW6BJ2I","schema_version":"1.0","canonical_sha256":"646de0a74831e170e1b75f609d4d2f1429f7858cc6538c1545464904b46fafa2","source":{"kind":"arxiv","id":"2605.18721","version":1},"attestation_state":"computed","paper":{"title":"General Preference Reinforcement Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Ahsan Bilal, Andreas Haupt, Arslan Chaudhry, Emily Fox, John M. Cioffi, Muhammad Ahmed Mohsin, Muhammad Umer, Sanmi Koyejo","submitted_at":"2026-05-18T17:50:27Z","abstract_excerpt":"Post-training has split large language model (LLM) alignment into two largely disconnected tracks. Online reinforcement learning (RL) with verifiable rewards drives emergent reasoning on math and code but depends on a programmatic verifier that cannot reach open-ended tasks, while preference optimization handles open-ended generation yet forgoes the continuous exploration that powers online RL. Closing this gap requires a verifier for open-ended quality, but a scalar reward model is the wrong shape for the job. Quality is multi-dimensional, and any scalar score is an incomplete proxy that lets"},"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":"2605.18721","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-05-18T17:50:27Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"ab95ac02e67827161183e5ba84c0409decc3345e06a42faec1c239d2b4d45013","abstract_canon_sha256":"e138757baea20df8eb4cbe2ac9b4a355d14ebd1895c3d9a73c426cab3a17ba30"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:06:16.966829Z","signature_b64":"m7MtiTbHS9QYbwA9sm2lkxfdfa9WOKDDc7BFCD3Qwy4zIg+geIb2960FDcgnmw+BozYCHwSnOXzwmaZTzQm3Ag==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"646de0a74831e170e1b75f609d4d2f1429f7858cc6538c1545464904b46fafa2","last_reissued_at":"2026-05-20T00:06:16.966016Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:06:16.966016Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"General Preference Reinforcement Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Ahsan Bilal, Andreas Haupt, Arslan Chaudhry, Emily Fox, John M. Cioffi, Muhammad Ahmed Mohsin, Muhammad Umer, Sanmi Koyejo","submitted_at":"2026-05-18T17:50:27Z","abstract_excerpt":"Post-training has split large language model (LLM) alignment into two largely disconnected tracks. Online reinforcement learning (RL) with verifiable rewards drives emergent reasoning on math and code but depends on a programmatic verifier that cannot reach open-ended tasks, while preference optimization handles open-ended generation yet forgoes the continuous exploration that powers online RL. Closing this gap requires a verifier for open-ended quality, but a scalar reward model is the wrong shape for the job. Quality is multi-dimensional, and any scalar score is an incomplete proxy that lets"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.18721","kind":"arxiv","version":1},"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/2605.18721/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"claim_evidence","ran_at":"2026-05-20T00:01:59.039302Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"836cd941d4dc2c1550c5cab0f70809201a6dea4f339037c7f7f7e36140c9602b"},"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":"2605.18721","created_at":"2026-05-20T00:06:16.966145+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.18721v1","created_at":"2026-05-20T00:06:16.966145+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.18721","created_at":"2026-05-20T00:06:16.966145+00:00"},{"alias_kind":"pith_short_12","alias_value":"MRW6BJ2IGHQX","created_at":"2026-05-20T00:06:16.966145+00:00"},{"alias_kind":"pith_short_16","alias_value":"MRW6BJ2IGHQXBYNX","created_at":"2026-05-20T00:06:16.966145+00:00"},{"alias_kind":"pith_short_8","alias_value":"MRW6BJ2I","created_at":"2026-05-20T00:06:16.966145+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/MRW6BJ2IGHQXBYNXL5QJ2TJPCQ","json":"https://pith.science/pith/MRW6BJ2IGHQXBYNXL5QJ2TJPCQ.json","graph_json":"https://pith.science/api/pith-number/MRW6BJ2IGHQXBYNXL5QJ2TJPCQ/graph.json","events_json":"https://pith.science/api/pith-number/MRW6BJ2IGHQXBYNXL5QJ2TJPCQ/events.json","paper":"https://pith.science/paper/MRW6BJ2I"},"agent_actions":{"view_html":"https://pith.science/pith/MRW6BJ2IGHQXBYNXL5QJ2TJPCQ","download_json":"https://pith.science/pith/MRW6BJ2IGHQXBYNXL5QJ2TJPCQ.json","view_paper":"https://pith.science/paper/MRW6BJ2I","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.18721&json=true","fetch_graph":"https://pith.science/api/pith-number/MRW6BJ2IGHQXBYNXL5QJ2TJPCQ/graph.json","fetch_events":"https://pith.science/api/pith-number/MRW6BJ2IGHQXBYNXL5QJ2TJPCQ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/MRW6BJ2IGHQXBYNXL5QJ2TJPCQ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/MRW6BJ2IGHQXBYNXL5QJ2TJPCQ/action/storage_attestation","attest_author":"https://pith.science/pith/MRW6BJ2IGHQXBYNXL5QJ2TJPCQ/action/author_attestation","sign_citation":"https://pith.science/pith/MRW6BJ2IGHQXBYNXL5QJ2TJPCQ/action/citation_signature","submit_replication":"https://pith.science/pith/MRW6BJ2IGHQXBYNXL5QJ2TJPCQ/action/replication_record"}},"created_at":"2026-05-20T00:06:16.966145+00:00","updated_at":"2026-05-20T00:06:16.966145+00:00"}