{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:AE7JSO7US72GWJ25YRVKSPVFZ3","short_pith_number":"pith:AE7JSO7U","schema_version":"1.0","canonical_sha256":"013e993bf497f46b275dc46aa93ea5cef4de8d472d0f9d99102e26007822f364","source":{"kind":"arxiv","id":"2607.05184","version":1},"attestation_state":"computed","paper":{"title":"Rethinking On-Policy Self-Distillation for Thinking Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.AI","authors_text":"Liam Fowl, Narutatsu Ri, Sanjeev Arora, Simran Kaur, Yinghui He","submitted_at":"2026-07-06T15:01:35Z","abstract_excerpt":"Self-distillation is a promising recipe for self-improvement in language models. In this setting, a model can serve as its own teacher when given privileged information, such as a solution to a math problem. This seems especially appealing for thinking models, which can use test-time reasoning to absorb the privileged information. Surprisingly, we show that privileged self-distillation degrades thinking models on long reasoning traces: across five Qwen3 and OLMo thinking models evaluated on AIME24, AIME25, and HMMT25, privileged-context distillation causes a relative drop of up to 17% in avg@1"},"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":"2607.05184","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-07-06T15:01:35Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"4ad7b4aa4741bf92f39ee5bfd602b38849959b2837198d8d8f8e350e2681e2e2","abstract_canon_sha256":"b594c79ff4302318796e721da7d981906d385c93345b85f7f8356877b209fec2"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-07T03:19:19.005796Z","signature_b64":"TSAUXWYCSPjFQS9JpFS5K1TNmcU3VsIBu9XTq6p/po9hDrwa3rlbdmhRIvDIE0pNDCtZdGSTEV7A0lLwvFXvBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"013e993bf497f46b275dc46aa93ea5cef4de8d472d0f9d99102e26007822f364","last_reissued_at":"2026-07-07T03:19:19.005380Z","signature_status":"signed_v1","first_computed_at":"2026-07-07T03:19:19.005380Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Rethinking On-Policy Self-Distillation for Thinking Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.AI","authors_text":"Liam Fowl, Narutatsu Ri, Sanjeev Arora, Simran Kaur, Yinghui He","submitted_at":"2026-07-06T15:01:35Z","abstract_excerpt":"Self-distillation is a promising recipe for self-improvement in language models. In this setting, a model can serve as its own teacher when given privileged information, such as a solution to a math problem. This seems especially appealing for thinking models, which can use test-time reasoning to absorb the privileged information. Surprisingly, we show that privileged self-distillation degrades thinking models on long reasoning traces: across five Qwen3 and OLMo thinking models evaluated on AIME24, AIME25, and HMMT25, privileged-context distillation causes a relative drop of up to 17% in avg@1"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2607.05184","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/2607.05184/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":"2607.05184","created_at":"2026-07-07T03:19:19.005439+00:00"},{"alias_kind":"arxiv_version","alias_value":"2607.05184v1","created_at":"2026-07-07T03:19:19.005439+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2607.05184","created_at":"2026-07-07T03:19:19.005439+00:00"},{"alias_kind":"pith_short_12","alias_value":"AE7JSO7US72G","created_at":"2026-07-07T03:19:19.005439+00:00"},{"alias_kind":"pith_short_16","alias_value":"AE7JSO7US72GWJ25","created_at":"2026-07-07T03:19:19.005439+00:00"},{"alias_kind":"pith_short_8","alias_value":"AE7JSO7U","created_at":"2026-07-07T03:19:19.005439+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2608.01735","citing_title":"DAPD: Dual-Anchored Policy Distillation","ref_index":76,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/AE7JSO7US72GWJ25YRVKSPVFZ3","json":"https://pith.science/pith/AE7JSO7US72GWJ25YRVKSPVFZ3.json","graph_json":"https://pith.science/api/pith-number/AE7JSO7US72GWJ25YRVKSPVFZ3/graph.json","events_json":"https://pith.science/api/pith-number/AE7JSO7US72GWJ25YRVKSPVFZ3/events.json","paper":"https://pith.science/paper/AE7JSO7U"},"agent_actions":{"view_html":"https://pith.science/pith/AE7JSO7US72GWJ25YRVKSPVFZ3","download_json":"https://pith.science/pith/AE7JSO7US72GWJ25YRVKSPVFZ3.json","view_paper":"https://pith.science/paper/AE7JSO7U","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2607.05184&json=true","fetch_graph":"https://pith.science/api/pith-number/AE7JSO7US72GWJ25YRVKSPVFZ3/graph.json","fetch_events":"https://pith.science/api/pith-number/AE7JSO7US72GWJ25YRVKSPVFZ3/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/AE7JSO7US72GWJ25YRVKSPVFZ3/action/timestamp_anchor","attest_storage":"https://pith.science/pith/AE7JSO7US72GWJ25YRVKSPVFZ3/action/storage_attestation","attest_author":"https://pith.science/pith/AE7JSO7US72GWJ25YRVKSPVFZ3/action/author_attestation","sign_citation":"https://pith.science/pith/AE7JSO7US72GWJ25YRVKSPVFZ3/action/citation_signature","submit_replication":"https://pith.science/pith/AE7JSO7US72GWJ25YRVKSPVFZ3/action/replication_record"}},"created_at":"2026-07-07T03:19:19.005439+00:00","updated_at":"2026-07-07T03:19:19.005439+00:00"}