What question did this study set out to answer?

To measure the frequency and severity of harmful errors generated by large language models and to evaluate physician-AI collaboration during medical consultations.

December 1, 2025Open Access

First, do NOHARM: a medical safety benchmark and randomized study of physician and AI teaming on clinical consultations

View Full Paper

Ask AI

Bookmark

To measure the frequency and severity of harmful errors generated by large language models and to evaluate physician-AI collaboration during medical consultations.
Evaluated 20 generalist large language models and 4 retrieval-augmented generation clinical tools using the NOHARM benchmark, containing 1,100 consultation tasks across 10 specialties with 12,747 expert annotations.
Conducted a randomized trial of 101 U.S.-licensed generalist physicians comparing clinical consultation performance using AI assistance versus conventional resources.
AI-generated recommendations carried potential for severe harm in up to 24.6% of consultation cases, with errors of omission accounting for over 80% of severe errors.
Specialized clinical AI tools outperformed generalist models, and multi-agent AI teaming improved generalist model performance.
AI assistance improved physician performance compared to conventional resources, but physicians frequently omitted valid AI advice and scored lower than standalone AI systems.

Discussion

No takes yet. Share an insight, caveat, or question.

Implication

Randomized trial reveals AI assistance enhances clinical consultation accuracy in generalist physicians, indicating unrealized potential and safety risks in human-AI teaming.

Cite This Study

Wu et al. (2025) studied this question.

synapsesocial.com/papers/6ab3534a664f6c052eeb190c https://doi.org/10.48550/arxiv.2512.01241

View Full Paper

Ask AI

Bookmark

Implication

Randomized trial reveals AI assistance enhances clinical consultation accuracy in generalist physicians, indicating unrealized potential and safety risks in human-AI teaming.

First, do NOHARM: a medical safety benchmark and randomized study of physician and AI teaming on clinical consultations

Key Points

Discussion

Implication

Cite This Study

First, do NOHARM: a medical safety benchmark and randomized study of physician and AI teaming on clinical consultations

Key Points

Discussion

Implication

Cite This Study

Keep reading

Keep reading