{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:SKH5THI64JH4L5276KGXB7YPWT","short_pith_number":"pith:SKH5THI6","schema_version":"1.0","canonical_sha256":"928fd99d1ee24fc5f75ff28d70ff0fb4d6702a68bee720e9b626e6c4c42dc154","source":{"kind":"arxiv","id":"2504.07527","version":2},"attestation_state":"computed","paper":{"title":"Supervised Optimism Correction: Be Confident When LLMs Are Sure","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Dacheng Tao, Fei Huang, Junjie Zhang, Rushuai Yang, Shunyu Liu, Ting-En Lin, Yi Chen, Yongbin Li","submitted_at":"2025-04-10T07:50:03Z","abstract_excerpt":"In this work, we establish a novel theoretical connection between supervised fine-tuning and offline reinforcement learning under the token-level Markov decision process, revealing that large language models indeed learn an implicit $Q$-function for inference. Through this theoretical lens, we demonstrate that the widely used beam search method suffers from unacceptable over-optimism, where inference errors are inevitably amplified due to inflated $Q$-value estimations of suboptimal steps. To address this limitation, we propose Supervised Optimism Correction(SOC), which introduces a simple yet"},"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":"2504.07527","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-04-10T07:50:03Z","cross_cats_sorted":[],"title_canon_sha256":"eceecbf4b2dbe02dea068890ad4b9359f7acd7ce695f95c65d2d8f56888b358b","abstract_canon_sha256":"e0c44422d2e70ce317396412655aca523bc8b1e87b1a11dc5112550ff8348dd6"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:13:23.892481Z","signature_b64":"F1oqJ+2Mcjmq+m1BpR9S8wXzt2mUJJ4w/VrH4aGo5YMxTUbPJ9fAXZUZDTjgGgZ0vg7RIBTcRQ+RiU4fC4POCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"928fd99d1ee24fc5f75ff28d70ff0fb4d6702a68bee720e9b626e6c4c42dc154","last_reissued_at":"2026-07-05T11:13:23.891765Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:13:23.891765Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Supervised Optimism Correction: Be Confident When LLMs Are Sure","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Dacheng Tao, Fei Huang, Junjie Zhang, Rushuai Yang, Shunyu Liu, Ting-En Lin, Yi Chen, Yongbin Li","submitted_at":"2025-04-10T07:50:03Z","abstract_excerpt":"In this work, we establish a novel theoretical connection between supervised fine-tuning and offline reinforcement learning under the token-level Markov decision process, revealing that large language models indeed learn an implicit $Q$-function for inference. Through this theoretical lens, we demonstrate that the widely used beam search method suffers from unacceptable over-optimism, where inference errors are inevitably amplified due to inflated $Q$-value estimations of suboptimal steps. To address this limitation, we propose Supervised Optimism Correction(SOC), which introduces a simple yet"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.07527","kind":"arxiv","version":2},"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/2504.07527/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":"2504.07527","created_at":"2026-07-05T11:13:23.891862+00:00"},{"alias_kind":"arxiv_version","alias_value":"2504.07527v2","created_at":"2026-07-05T11:13:23.891862+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.07527","created_at":"2026-07-05T11:13:23.891862+00:00"},{"alias_kind":"pith_short_12","alias_value":"SKH5THI64JH4","created_at":"2026-07-05T11:13:23.891862+00:00"},{"alias_kind":"pith_short_16","alias_value":"SKH5THI64JH4L527","created_at":"2026-07-05T11:13:23.891862+00:00"},{"alias_kind":"pith_short_8","alias_value":"SKH5THI6","created_at":"2026-07-05T11:13:23.891862+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/SKH5THI64JH4L5276KGXB7YPWT","json":"https://pith.science/pith/SKH5THI64JH4L5276KGXB7YPWT.json","graph_json":"https://pith.science/api/pith-number/SKH5THI64JH4L5276KGXB7YPWT/graph.json","events_json":"https://pith.science/api/pith-number/SKH5THI64JH4L5276KGXB7YPWT/events.json","paper":"https://pith.science/paper/SKH5THI6"},"agent_actions":{"view_html":"https://pith.science/pith/SKH5THI64JH4L5276KGXB7YPWT","download_json":"https://pith.science/pith/SKH5THI64JH4L5276KGXB7YPWT.json","view_paper":"https://pith.science/paper/SKH5THI6","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2504.07527&json=true","fetch_graph":"https://pith.science/api/pith-number/SKH5THI64JH4L5276KGXB7YPWT/graph.json","fetch_events":"https://pith.science/api/pith-number/SKH5THI64JH4L5276KGXB7YPWT/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/SKH5THI64JH4L5276KGXB7YPWT/action/timestamp_anchor","attest_storage":"https://pith.science/pith/SKH5THI64JH4L5276KGXB7YPWT/action/storage_attestation","attest_author":"https://pith.science/pith/SKH5THI64JH4L5276KGXB7YPWT/action/author_attestation","sign_citation":"https://pith.science/pith/SKH5THI64JH4L5276KGXB7YPWT/action/citation_signature","submit_replication":"https://pith.science/pith/SKH5THI64JH4L5276KGXB7YPWT/action/replication_record"}},"created_at":"2026-07-05T11:13:23.891862+00:00","updated_at":"2026-07-05T11:13:23.891862+00:00"}