A new benchmark called Safe-Psych has been developed to evaluate how large language models (LLMs) handle diagnostic uncertainty in psychiatry. The benchmark, comprising over 1,000 real-world psychiatric notes, simulates incremental evidence disclosure and assigns labels such as DIAGNOSE, CLARIFY, or ABSTAIN at each stage. Evaluations revealed that even advanced LLMs struggle with incomplete clinical information, often diagnosing prematurely and rarely seeking clarification unless prompted. This suggests a significant limitation in current models' ability to recognize when more evidence is needed for safe and accurate clinical decision-making. AI
IMPACT Highlights a critical safety gap in LLMs for healthcare, necessitating further research into calibration and evidence-based decision-making.
RANK_REASON The cluster contains a research paper introducing a new benchmark for evaluating LLMs. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
