AI Summary: Researchers from Boston Children’s Hospital, Harvard University, and OpenAI utilized the OpenAI o3 Deep Research model to analyze clinical and genomic data from 376 previously unsolved rare disease cases. The model generated evidence-linked hypotheses, leading to the establishment of diagnoses in 18 cases, representing a 4.8% increase in diagnostic yield after prior expert analysis. This study, published in NEJM AI, demonstrates the potential of AI-assisted workflows to facilitate the reanalysis of genetic data as scientific knowledge evolves, highlighting the importance of periodic expert review in uncovering previously overlooked diagnoses. The model served as a reasoning layer, connecting clinical features and genetic evidence to support human review rather than making clinical decisions independently.
Improving health intelligence in ChatGPT
AI Summary: The article discusses advancements in ChatGPT's health-related capabilities with the introduction of GPT-5.5 Instant, which shows significant improvements in recognizing urgent care needs, contextual understanding, and clarity in communication. Evaluations using HealthBench and comparisons with physician-written responses indicate that GPT-5.5 Instant performs at a level comparable to leading models, with higher ratings in accuracy and fewer failure modes than both older models and physician responses. Additionally, a 71% reduction in flagged factuality issues in health responses over the past two months highlights the model's enhanced reliability. These improvements are attributed to collaboration with a global network of physicians who help define and measure effective health communication.