900 de-identified electronic medical records (EMRs) from psychiatric inpatients used to evaluate LLM performance in clinical practice. The data consists of real-world clinical resources without synthetic augmentation or paraphrasing to ensure professional-grade benchmarking.
Use Cases
- Evaluate the diagnostic accuracy of LLMs using the de-identified inpatient electronic medical records
- Benchmark clinical reasoning capabilities of models against professional psychiatric practice standards
- Test the safety and reliability of LLM-assisted psychiatric interventions using real-world clinical scenarios
- Analyze model performance on authentic psychiatric inpatient data to identify clinical biases
Strengths
- 900 de-identified electronic medical records (EMRs) from real-world psychiatric inpatients
- Excludes synthetic augmentation or paraphrasing to maintain clinical authenticity
- Official benchmark for evaluating LLM-assisted psychiatric clinical practice
- Sourced from professional psychiatric clinical practice environments