When Health Assistants Agree Too Readily: A Benchmark of Time-Critical Self-Diagnosis

Key points

• Every day, more than 40 million people ask ChatGPT a health question, such as whether laptop posture explains a stiff neck and days of pounding headache.1
• Posture is plausible, but meningitis is a time-critical possibility.1
• We find across 27 frontier models that they tend to inappropriately agree with the user's self-diagnosis in cases that require caution.1
• We recommend post-training to improve appropriate caution without making models cautious on every case, with paired evaluation to verify both response behaviors.1

Sources (1)

• [1]When Health Assistants Agree Too Readily: A Benchmark of Time-Critical Self-Diagnosis
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 20, 02:31 AM

  We introduce SelfDx, a paired benchmark for whether models have the appropriate response behavior, a form of epistemic independence, in health questions where misplaced agreement with the user's self-diagnosis could delay urgent or emergency care.

  Every day, more than 40 million people ask ChatGPT a health question, such as whether laptop posture explains a stiff neck and days of pounding headache.

Extractive summary: sentences quoted from the sources.