Cross-Model Variability in Large Language Model Triage Behavior for Potential Stroke Symptoms
Frontier LLMs under-responded to potential stroke like symptoms, and showed substantial model-to-model differences in response to changing symptom descriptions. As individuals increasingly turn to LLMs for medical advice, which model they use and what words they choose could lead to drastically different informal triage guidance.
AI safetyemergency medicinetriagepublic health
Building in Calm, Deploying in Crisis
Most AI systems are built in calm but must perform in crisis. Three structural gaps explain why current AI falls short in emergency medicine: training on theory not practice, output designed for calm cognition not crisis, and systems built for individuals not teams.
emergency medicinedesignAI safety
Software for HALO (High Acuity Low Occurrence) event exposure and training simulation for emergency teams.
softwaretrainingHALO
Prompt-Injected State Descriptions Shift Surgical Airway Decision Timing in a Large Language Model
Modeling LLM decision-making in CICO (Can't Intubate, Can't Oxygenate) emergencies. Prompt-injections focused on altered affective and cognitive states substantially changed LLM decision making.
simulationAI safetyemergency medicinedecision making
Wiring Your Brain for Performance Under Pressure. How to train mentally to perform at your best when you're needed the most.
bookcognitive performanceemergency medicine
Conversations on performance, decision-making, and the human side of emergency care.
podcastcognitive performanceemergency medicine