{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:BFOX3J775OGVS55N6VDJAQGTER","short_pith_number":"pith:BFOX3J77","schema_version":"1.0","canonical_sha256":"095d7da7ffeb8d5977adf5469040d324464b3bb2293468a18d97efd3a782a534","source":{"kind":"arxiv","id":"2505.02363","version":1},"attestation_state":"computed","paper":{"title":"SIMPLEMIX: Frustratingly Simple Mixing of Off- and On-policy Data in Language Model Preference Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Daniel Khashabi, Tianjian Li","submitted_at":"2025-05-05T04:54:44Z","abstract_excerpt":"Aligning language models with human preferences relies on pairwise preference datasets. While some studies suggest that on-policy data consistently outperforms off -policy data for preference learning, others indicate that the advantages of on-policy data may be task-dependent, highlighting the need for a systematic exploration of their interplay.\n  In this work, we show that on-policy and off-policy data offer complementary strengths in preference optimization: on-policy data is particularly effective for reasoning tasks like math and coding, while off-policy data performs better on open-ende"},"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":"2505.02363","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-05-05T04:54:44Z","cross_cats_sorted":[],"title_canon_sha256":"1eef84c2a56aed7a9ff8aa2843cf0faa7ecb901a638c9e845890825ff7b2b69f","abstract_canon_sha256":"c76238232c3465c8ec3dc4a4d6d6ab83cf53b33817053818803ed70e6e235585"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:58:35.941406Z","signature_b64":"oYoxdUT79mPoLcLWwE7sizdNGOZ70/aUBlY2uPmqHxbC5c8wXhLeTz8DpRhcwNMfD4hQHGJBxA1Tv3dlkDQPBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"095d7da7ffeb8d5977adf5469040d324464b3bb2293468a18d97efd3a782a534","last_reissued_at":"2026-07-05T10:58:35.940999Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:58:35.940999Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"SIMPLEMIX: Frustratingly Simple Mixing of Off- and On-policy Data in Language Model Preference Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Daniel Khashabi, Tianjian Li","submitted_at":"2025-05-05T04:54:44Z","abstract_excerpt":"Aligning language models with human preferences relies on pairwise preference datasets. While some studies suggest that on-policy data consistently outperforms off -policy data for preference learning, others indicate that the advantages of on-policy data may be task-dependent, highlighting the need for a systematic exploration of their interplay.\n  In this work, we show that on-policy and off-policy data offer complementary strengths in preference optimization: on-policy data is particularly effective for reasoning tasks like math and coding, while off-policy data performs better on open-ende"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.02363","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/2505.02363/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":"2505.02363","created_at":"2026-07-05T10:58:35.941053+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.02363v1","created_at":"2026-07-05T10:58:35.941053+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.02363","created_at":"2026-07-05T10:58:35.941053+00:00"},{"alias_kind":"pith_short_12","alias_value":"BFOX3J775OGV","created_at":"2026-07-05T10:58:35.941053+00:00"},{"alias_kind":"pith_short_16","alias_value":"BFOX3J775OGVS55N","created_at":"2026-07-05T10:58:35.941053+00:00"},{"alias_kind":"pith_short_8","alias_value":"BFOX3J77","created_at":"2026-07-05T10:58:35.941053+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/BFOX3J775OGVS55N6VDJAQGTER","json":"https://pith.science/pith/BFOX3J775OGVS55N6VDJAQGTER.json","graph_json":"https://pith.science/api/pith-number/BFOX3J775OGVS55N6VDJAQGTER/graph.json","events_json":"https://pith.science/api/pith-number/BFOX3J775OGVS55N6VDJAQGTER/events.json","paper":"https://pith.science/paper/BFOX3J77"},"agent_actions":{"view_html":"https://pith.science/pith/BFOX3J775OGVS55N6VDJAQGTER","download_json":"https://pith.science/pith/BFOX3J775OGVS55N6VDJAQGTER.json","view_paper":"https://pith.science/paper/BFOX3J77","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.02363&json=true","fetch_graph":"https://pith.science/api/pith-number/BFOX3J775OGVS55N6VDJAQGTER/graph.json","fetch_events":"https://pith.science/api/pith-number/BFOX3J775OGVS55N6VDJAQGTER/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/BFOX3J775OGVS55N6VDJAQGTER/action/timestamp_anchor","attest_storage":"https://pith.science/pith/BFOX3J775OGVS55N6VDJAQGTER/action/storage_attestation","attest_author":"https://pith.science/pith/BFOX3J775OGVS55N6VDJAQGTER/action/author_attestation","sign_citation":"https://pith.science/pith/BFOX3J775OGVS55N6VDJAQGTER/action/citation_signature","submit_replication":"https://pith.science/pith/BFOX3J775OGVS55N6VDJAQGTER/action/replication_record"}},"created_at":"2026-07-05T10:58:35.941053+00:00","updated_at":"2026-07-05T10:58:35.941053+00:00"}