{"paper":{"title":"Adaptive Teacher Exposure for Self-Distillation in LLM Reasoning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Adaptive control of how much reference reasoning the teacher sees during self-distillation improves LLM performance on math tasks.","cross_cats":["cs.CL","cs.LO"],"primary_cat":"cs.AI","authors_text":"Huaibin Wang, Tiangang Zhang, Yilun Sun, Zihao Han","submitted_at":"2026-05-12T03:15:58Z","abstract_excerpt":"On-policy self-distillation has become a strong recipe for LLM reasoning, where a privileged teacher supervises the student's own rollouts while conditioning on the reference solution. A design choice shared by nearly all such methods, however, has gone unquestioned: the teacher always sees the full reference reasoning. We argue that this default itself is part of the problem and identify a teacher-side exposure mismatch: when the teacher conditions on reasoning far beyond the student's current competence, the resulting token targets become too strong to absorb. A controlled fixed-exposure swe"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Experiments on AIME 24, AIME 25, and HMMT 25 across Qwen3-{1.7B, 4B, 8B} show that ATESD consistently outperforms competitive self-distillation and RL baselines, improving over OPSD by +0.95, +2.05, and +2.33 Average@12 points respectively.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That a lightweight Beta-policy controller, optimized via a discounted learning-progress reward on compact training-state statistics, will reliably produce exposure decisions that improve long-term student performance without introducing training instability or benchmark-specific overfitting.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"ATESD makes teacher exposure to reference reasoning a learnable control variable via a Beta-policy optimized on future student improvement, yielding gains of up to +2.33 points over fixed-exposure self-distillation on AIME and HMMT math benchmarks.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Adaptive control of how much reference reasoning the teacher sees during self-distillation improves LLM performance on math tasks.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"247392f2a7a4b3ce7677691686c4452dd9cf055493a5c128ece74144e3dda313"},"source":{"id":"2605.11458","kind":"arxiv","version":2},"verdict":{"id":"280b6ff4-8e7c-40fd-a0ac-4edfcda05f18","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-13T01:58:33.413358Z","strongest_claim":"Experiments on AIME 24, AIME 25, and HMMT 25 across Qwen3-{1.7B, 4B, 8B} show that ATESD consistently outperforms competitive self-distillation and RL baselines, improving over OPSD by +0.95, +2.05, and +2.33 Average@12 points respectively.","one_line_summary":"ATESD makes teacher exposure to reference reasoning a learnable control variable via a Beta-policy optimized on future student improvement, yielding gains of up to +2.33 points over fixed-exposure self-distillation on AIME and HMMT math benchmarks.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That a lightweight Beta-policy controller, optimized via a discounted learning-progress reward on compact training-state statistics, will reliably produce exposure decisions that improve long-term student performance without introducing training instability or benchmark-specific overfitting.","pith_extraction_headline":"Adaptive control of how much reference reasoning the teacher sees during self-distillation improves LLM performance on math tasks."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.11458/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"claim_evidence","ran_at":"2026-05-20T04:22:00.356800Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T12:35:50.141606Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T10:01:16.403452Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T08:27:17.389902Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"38adbe07a56e833b42a95bc4bdf342f3b25d1dbdf58aeae49e5fe179f68572f6"},"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"}