{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:HO2BLUIFYL75VP7RLM7VSGPFGH","short_pith_number":"pith:HO2BLUIF","schema_version":"1.0","canonical_sha256":"3bb415d105c2ffdabff15b3f5919e531ff92395c8148768b4e090e08f9649328","source":{"kind":"arxiv","id":"2505.11462","version":2},"attestation_state":"computed","paper":{"title":"Disentangling Reasoning and Knowledge in Medical Large Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Angela Zhang, Ben Athiwaratkun, Eric Wu, Haotian Ye, Harrison Zhang, James Zou, Joseph Boen, Kevin Wu, Nevin Aresh, Qingyang Wu, Rahul Thapa, Shriya Reddy, Shuaiwen Leon Song, Suhana Bedi","submitted_at":"2025-05-16T17:16:27Z","abstract_excerpt":"Medical reasoning in large language models (LLMs) aims to emulate clinicians' diagnostic thinking, but current benchmarks such as MedQA-USMLE, MedMCQA, and PubMedQA often mix reasoning with factual recall. We address this by separating 11 biomedical QA benchmarks into reasoning- and knowledge-focused subsets using a PubMedBERT classifier that reaches 81 percent accuracy, comparable to human performance. Our analysis shows that only 32.8 percent of questions require complex reasoning. We evaluate biomedical models (HuatuoGPT-o1, MedReason, m1) and general-domain models (DeepSeek-R1, o4-mini, Qw"},"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":"2505.11462","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-05-16T17:16:27Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"f308ac39b41f4fa031859e54c1b82d51ce90dedc46da1d86ca2e7dc138ab377c","abstract_canon_sha256":"27c264030a3827738c11c6113c646887b09504614f0b86bb4c8227ffb5c9daff"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:26:04.402717Z","signature_b64":"pzZg5RfFOZUoG6+n+v11YCkuOx+3v6qs1crVPOb/fNl/jtpxHF4uesWmD+xr3T+0iEc/UleOfdAQ0mn/NdEABw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"3bb415d105c2ffdabff15b3f5919e531ff92395c8148768b4e090e08f9649328","last_reissued_at":"2026-07-05T11:26:04.402209Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:26:04.402209Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Disentangling Reasoning and Knowledge in Medical Large Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Angela Zhang, Ben Athiwaratkun, Eric Wu, Haotian Ye, Harrison Zhang, James Zou, Joseph Boen, Kevin Wu, Nevin Aresh, Qingyang Wu, Rahul Thapa, Shriya Reddy, Shuaiwen Leon Song, Suhana Bedi","submitted_at":"2025-05-16T17:16:27Z","abstract_excerpt":"Medical reasoning in large language models (LLMs) aims to emulate clinicians' diagnostic thinking, but current benchmarks such as MedQA-USMLE, MedMCQA, and PubMedQA often mix reasoning with factual recall. We address this by separating 11 biomedical QA benchmarks into reasoning- and knowledge-focused subsets using a PubMedBERT classifier that reaches 81 percent accuracy, comparable to human performance. Our analysis shows that only 32.8 percent of questions require complex reasoning. We evaluate biomedical models (HuatuoGPT-o1, MedReason, m1) and general-domain models (DeepSeek-R1, o4-mini, Qw"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.11462","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/2505.11462/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":"2505.11462","created_at":"2026-07-05T11:26:04.402268+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.11462v2","created_at":"2026-07-05T11:26:04.402268+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.11462","created_at":"2026-07-05T11:26:04.402268+00:00"},{"alias_kind":"pith_short_12","alias_value":"HO2BLUIFYL75","created_at":"2026-07-05T11:26:04.402268+00:00"},{"alias_kind":"pith_short_16","alias_value":"HO2BLUIFYL75VP7R","created_at":"2026-07-05T11:26:04.402268+00:00"},{"alias_kind":"pith_short_8","alias_value":"HO2BLUIF","created_at":"2026-07-05T11:26:04.402268+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.05241","citing_title":"Search-Time Contamination in Deep Research Agents: Measuring Performance Inflation in Public Benchmark Evaluation","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2604.08559","citing_title":"Medical Reasoning with Large Language Models: A Survey and MR-Bench","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2604.14475","citing_title":"Evo-MedAgent: Beyond One-Shot Diagnosis with Agents That Remember, Reflect, and Improve","ref_index":23,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/HO2BLUIFYL75VP7RLM7VSGPFGH","json":"https://pith.science/pith/HO2BLUIFYL75VP7RLM7VSGPFGH.json","graph_json":"https://pith.science/api/pith-number/HO2BLUIFYL75VP7RLM7VSGPFGH/graph.json","events_json":"https://pith.science/api/pith-number/HO2BLUIFYL75VP7RLM7VSGPFGH/events.json","paper":"https://pith.science/paper/HO2BLUIF"},"agent_actions":{"view_html":"https://pith.science/pith/HO2BLUIFYL75VP7RLM7VSGPFGH","download_json":"https://pith.science/pith/HO2BLUIFYL75VP7RLM7VSGPFGH.json","view_paper":"https://pith.science/paper/HO2BLUIF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.11462&json=true","fetch_graph":"https://pith.science/api/pith-number/HO2BLUIFYL75VP7RLM7VSGPFGH/graph.json","fetch_events":"https://pith.science/api/pith-number/HO2BLUIFYL75VP7RLM7VSGPFGH/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/HO2BLUIFYL75VP7RLM7VSGPFGH/action/timestamp_anchor","attest_storage":"https://pith.science/pith/HO2BLUIFYL75VP7RLM7VSGPFGH/action/storage_attestation","attest_author":"https://pith.science/pith/HO2BLUIFYL75VP7RLM7VSGPFGH/action/author_attestation","sign_citation":"https://pith.science/pith/HO2BLUIFYL75VP7RLM7VSGPFGH/action/citation_signature","submit_replication":"https://pith.science/pith/HO2BLUIFYL75VP7RLM7VSGPFGH/action/replication_record"}},"created_at":"2026-07-05T11:26:04.402268+00:00","updated_at":"2026-07-05T11:26:04.402268+00:00"}