--- format: "aidr-story-markdown/v1" id: "2d2a70b7b4865bd070793c98dafa6c0a48f38be23271b9e0b0bcfcb179ba2870" canonical_url: "https://aidr.today/2d2a70b7?lang=en" title: "Researchers Launch Synthetic Hospital Benchmark of 1,268 Patients That Fool Physicians" lang: "en" requested_lang: "en" available_langs: ["en","vi"] translation_fallback: null fallback_fields: [] published_at: "2026-09-25T16:36:16.000Z" category: "Research" topics: ["open-source"] source_urls: ["https://huggingnews.com/ai/researchers-launch-synthetic-hospital-benchmark-of-1268-patients-that-fo-0ea3a8a4","https://x.com/Tim_Dettmers/status/2103499499742048292","https://marketbrief.now/ai/researchers-launch-synthetic-hospital-benchmark-of-1268-patients-that-fo-0ea3a8a4"] summary: "A new open source electronic health record provides an anonymized way to train reinforcement learning models on longitudinal patient data. Tim Dettmers and his team released the Synthetic Hospital benchmark, which consists of 1,268 patients and 5,602 encounters. The researchers verified that the records contain zero protected health information, allowing the dataset to be distributed publicly without privacy restrictions. The benchmark's data quality is high enough that physicians were unable to reliably differentiate the synthetic charts from actual patient records. Because the simulation includes a verifiable ground truth, the authors describe it as an ideal tool for reinforcement learning and for testing AI models in medical settings." --- # Researchers Launch Synthetic Hospital Benchmark of 1,268 Patients That Fool Physicians > [Open the canonical story]() **Published:** 2026-09-25T16:36:16.000Z **Category:** Research **Topics:** open\-source ## Summary A new open source electronic health record provides an anonymized way to train reinforcement learning models on longitudinal patient data\. Tim Dettmers and his team released the Synthetic Hospital benchmark, which consists of 1,268 patients and 5,602 encounters\. The researchers verified that the records contain zero protected health information, allowing the dataset to be distributed publicly without privacy restrictions\. The benchmark's data quality is high enough that physicians were unable to reliably differentiate the synthetic charts from actual patient records\. Because the simulation includes a verifiable ground truth, the authors describe it as an ideal tool for reinforcement learning and for testing AI models in medical settings\. ## Sources - [Story source]() - [Story source]() - [Story source]()