{"paper":{"title":"Large Language Models Lack Temporal Awareness of Medical Knowledge","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Large language models lack awareness of when medical knowledge applies in time.","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Anil Vullikanti, Fangyuan Chen, Guangzhi Xiong, Mengxuan Hu, Qiao Jin, Qingyu Chen, Yifan Peng, Zhiyong Lu, Zihan Guan","submitted_at":"2026-05-13T06:04:40Z","abstract_excerpt":"The existing methods for evaluating the medical knowledge of Large Language Models (LLMs) are largely based on atemporal examination-style benchmarks, while in reality, medical knowledge is inherently dynamic and continuously evolves as new evidence emerges and treatments are approved. Consequently, evaluating medical knowledge without a temporal context may provide an incomplete assessment of whether LLMs can accurately reason about time-specific medical knowledge. Moreover, most medical data are historical, requiring the models not only to recall the correct knowledge, but also to know when "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"LLMs lack temporal awareness in medical knowledge: performance on up-to-date knowledge declines gradually rather than showing sharp cutoff, historical knowledge accuracy is only 25.37%-53.89% of up-to-date, and models exhibit temporally inconsistent behaviors.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the selected evolving medical guidelines in TempoMed-Bench are representative of temporal medical knowledge changes and that model outputs reflect internal parametric knowledge rather than prompt or retrieval artifacts.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"LLMs lack temporal awareness of medical knowledge, showing gradual performance decline on up-to-date facts, much lower accuracy on historical knowledge (25-54% relative), and inconsistent year-to-year predictions.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Large language models lack awareness of when medical knowledge applies in time.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"602eff3df8c72203971b150a13131afa5f490c080a913fa420471453c1513e84"},"source":{"id":"2605.13045","kind":"arxiv","version":1},"verdict":{"id":"cbfd81a9-b9e3-4dc0-850b-e5c48e60962b","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T20:13:56.264058Z","strongest_claim":"LLMs lack temporal awareness in medical knowledge: performance on up-to-date knowledge declines gradually rather than showing sharp cutoff, historical knowledge accuracy is only 25.37%-53.89% of up-to-date, and models exhibit temporally inconsistent behaviors.","one_line_summary":"LLMs lack temporal awareness of medical knowledge, showing gradual performance decline on up-to-date facts, much lower accuracy on historical knowledge (25-54% relative), and inconsistent year-to-year predictions.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the selected evolving medical guidelines in TempoMed-Bench are representative of temporal medical knowledge changes and that model outputs reflect internal parametric knowledge rather than prompt or retrieval artifacts.","pith_extraction_headline":"Large language models lack awareness of when medical knowledge applies in time."},"references":{"count":62,"sample":[{"doi":"","year":2025,"title":"gpt-oss-120b & gpt-oss-20b Model Card","work_id":"178c1f7e-4f19-4392-a45d-45a6dfa88ead","ref_index":1,"cited_arxiv_id":"2508.10925","is_internal_anchor":true},{"doi":"","year":2025,"title":"The distracting effect: Understanding irrelevant passages in rag","work_id":"153de318-71f7-44de-ac1d-aadcea5188a7","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"HealthBench: Evaluating Large Language Models Towards Improved Human Health","work_id":"5d613c2c-158f-488c-9981-fc9b82a5e093","ref_index":3,"cited_arxiv_id":"2505.08775","is_internal_anchor":true},{"doi":"","year":2024,"title":"Jae Hyun Bae, Ji-Hee Haam, Eonju Jeon, Seo Young Kang, SuJin Song, Cheol-Young Park, Hyuktae Kwon, Committee of Clinical Practice Guidelines, et al. 2024 clinical practice guidelines for the diagnosis","work_id":"57553244-8629-4d81-9c16-ff85672051bd","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Diagnostic accuracy of a large language model in pediatric case studies.JAMA pediatrics, 178(3):313–315, 2024","work_id":"0f440dca-ddff-46fd-b787-68d504d8f466","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":62,"snapshot_sha256":"405ff4665692bbd05ec0cf42b89e10498dd979ac1efc99bf20b40a0059a61b37","internal_anchors":9},"formal_canon":{"evidence_count":2,"snapshot_sha256":"abd01da47dcf8afd97d71c568c2e7cadb181d6e5c9d026a9c39c159c06af52ed"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}