{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:ZD3TXDFRDE2DFYFAYQ7XDWCQCU","short_pith_number":"pith:ZD3TXDFR","schema_version":"1.0","canonical_sha256":"c8f73b8cb1193432e0a0c43f71d85015056ad33dca81b2e955f6709740da3c86","source":{"kind":"arxiv","id":"2605.17314","version":1},"attestation_state":"computed","paper":{"title":"Weak-to-Strong Elicitation via Mismatched Wrong Drafts","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Wei Deng","submitted_at":"2026-05-17T08:12:51Z","abstract_excerpt":"We consider whether off-policy experience from a smaller, weaker model can elicit capability in a stronger learner that on-policy RL fine-tuning (e.g., GRPO) does not reach. We find that injecting mathematically wrong drafts from a smaller but more domain-trained model -- mismatched to the current problem -- into a stronger learner's GRPO context consistently outperforms standard on-policy GRPO on held-out MATH-500 and out-of-distribution AIME 2025/2026. Concretely, we use Mathstral-7B as the learner, Qwen2.5-Math-1.5B as the draft model, 8.8K Level 3--5 MATH problems (with MATH-500 held out),"},"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":"2605.17314","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-05-17T08:12:51Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"f919b4ff0c669173d7777f1632f70a7533f9a1c41be723509d60d1911e3974db","abstract_canon_sha256":"94afb21904af58104ff3468cedd65611840ffd10fce1594d35f8bc6bfb872009"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:03:51.689581Z","signature_b64":"FqC1dlH/u2CRVPeFpauqVWBeSTAJqsCPKpmtU+6jGSi5cQYdfMZZddoJnerx9jbq7knRA1ZpfN4WPhtVbr9BBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c8f73b8cb1193432e0a0c43f71d85015056ad33dca81b2e955f6709740da3c86","last_reissued_at":"2026-05-20T00:03:51.688613Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:03:51.688613Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Weak-to-Strong Elicitation via Mismatched Wrong Drafts","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Wei Deng","submitted_at":"2026-05-17T08:12:51Z","abstract_excerpt":"We consider whether off-policy experience from a smaller, weaker model can elicit capability in a stronger learner that on-policy RL fine-tuning (e.g., GRPO) does not reach. We find that injecting mathematically wrong drafts from a smaller but more domain-trained model -- mismatched to the current problem -- into a stronger learner's GRPO context consistently outperforms standard on-policy GRPO on held-out MATH-500 and out-of-distribution AIME 2025/2026. Concretely, we use Mathstral-7B as the learner, Qwen2.5-Math-1.5B as the draft model, 8.8K Level 3--5 MATH problems (with MATH-500 held out),"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.17314","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/2605.17314/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"claim_evidence","ran_at":"2026-05-19T22:01:57.783512Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T21:33:23.752665Z","status":"skipped","version":"1.0.0","findings_count":0}],"snapshot_sha256":"036dbb531853c838a93669c16361e747791775b5965e1e568668f977dab6fdb7"},"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":"2605.17314","created_at":"2026-05-20T00:03:51.688778+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.17314v1","created_at":"2026-05-20T00:03:51.688778+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.17314","created_at":"2026-05-20T00:03:51.688778+00:00"},{"alias_kind":"pith_short_12","alias_value":"ZD3TXDFRDE2D","created_at":"2026-05-20T00:03:51.688778+00:00"},{"alias_kind":"pith_short_16","alias_value":"ZD3TXDFRDE2DFYFA","created_at":"2026-05-20T00:03:51.688778+00:00"},{"alias_kind":"pith_short_8","alias_value":"ZD3TXDFR","created_at":"2026-05-20T00:03:51.688778+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/ZD3TXDFRDE2DFYFAYQ7XDWCQCU","json":"https://pith.science/pith/ZD3TXDFRDE2DFYFAYQ7XDWCQCU.json","graph_json":"https://pith.science/api/pith-number/ZD3TXDFRDE2DFYFAYQ7XDWCQCU/graph.json","events_json":"https://pith.science/api/pith-number/ZD3TXDFRDE2DFYFAYQ7XDWCQCU/events.json","paper":"https://pith.science/paper/ZD3TXDFR"},"agent_actions":{"view_html":"https://pith.science/pith/ZD3TXDFRDE2DFYFAYQ7XDWCQCU","download_json":"https://pith.science/pith/ZD3TXDFRDE2DFYFAYQ7XDWCQCU.json","view_paper":"https://pith.science/paper/ZD3TXDFR","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.17314&json=true","fetch_graph":"https://pith.science/api/pith-number/ZD3TXDFRDE2DFYFAYQ7XDWCQCU/graph.json","fetch_events":"https://pith.science/api/pith-number/ZD3TXDFRDE2DFYFAYQ7XDWCQCU/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ZD3TXDFRDE2DFYFAYQ7XDWCQCU/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ZD3TXDFRDE2DFYFAYQ7XDWCQCU/action/storage_attestation","attest_author":"https://pith.science/pith/ZD3TXDFRDE2DFYFAYQ7XDWCQCU/action/author_attestation","sign_citation":"https://pith.science/pith/ZD3TXDFRDE2DFYFAYQ7XDWCQCU/action/citation_signature","submit_replication":"https://pith.science/pith/ZD3TXDFRDE2DFYFAYQ7XDWCQCU/action/replication_record"}},"created_at":"2026-05-20T00:03:51.688778+00:00","updated_at":"2026-05-20T00:03:51.688778+00:00"}