Red teaming ChatGPT in medicine to yield real-world insights on model behavior
Проверка ChatGPT методом red teaming в медицине для получения практических сведений о поведении модели
2025-03-07
SCID: 54.1/dzzkv7ne
Discuss with AI
clinical caseshallucinations and biaslarge language modelsmodel safetyred teaming
Figures from the paper
Abstract (AI)
0. Abstract Background The integration of large language models (LLMs) in healthcare offers immense opportunity to streamline healthcare tasks, but also carries risks such as response accuracy and bias perpetration. To address this, we conducted a red-teaming exercise to assess LLMs in healthcare and developed a dataset of clinically relevant scenarios for future teams to use. Methods We convened 80 multi-disciplinary experts to evaluate the performance of popular LLMs across multiple medical scenarios. Teams composed of clinicians, medical and engineering students, and technical professionals stress-tested LLMs with real world clinical use cases. Teams were given a framework comprising four categories to analyze for inappropriate responses: Safety, Privacy, Hallucinations, and Bias. Prompts were tested on GPT-3.5, GPT-4.0, and GPT-4.0 with the Internet. Six medically trained reviewers subsequently reanalyzed the prompt-response pairs, with dual reviewers for each prompt and a third to resolve discrepancies. This process allowed for the accurate identification and categorization of inappropriate or inaccurate content within the responses. Results There were a total of 382 unique prompts, with 1146 total responses across three iterations of ChatGPT (GPT-3.5, GPT-4.0, GPT-4.0 with Internet). 19.8% of the responses were labeled as inappropriate, with GPT-3.5 accounting for the highest percentage at 25.7% while GPT-4.0 and GPT-4.0 with internet performing comparably at 16.2% and 17.5% respectively. Interestingly, 11.8% of responses were deemed appropriate with GPT-3.5 but inappropriate in updated models, highlighting the ongoing need to evaluate evolving LLMs. Conclusion The red-teaming exercise underscored the benefits of interdisciplinary efforts, as this collaborative model fosters a deeper understanding of the potential limitations of LLMs in healthcare and sets a precedent for future red teaming events in the field. Additionally, we present all prompts and outputs as a benchmark for future LLM model evaluations. 1-2 Sentence Description As a proof-of-concept, we convened an interactive “red teaming” workshop in which medical and technical professionals stress-tested popular large language models (LLMs) through publicly available user interfaces on clinically relevant scenarios. Results demonstrate a significant proportion of inappropriate responses across GPT-3.5, GPT-4.0, and GPT-4.0 with Internet (25.7%, 16.2%, and 17.5%, respectively) and illustrate the valuable role that non-technical clinicians can play in evaluating models.
Key Findings
1
A non-model-creator-affiliated red-teaming effort engaged 80 clinicians, students, and technical professionals to evaluate medical LLM behavior using real-world clinical cases.
2
Among 376 unique prompts generating 1,504 responses, 20.1% were inappropriate across safety, privacy, hallucination/accuracy, and bias dimensions.
3
Inappropriate-response rates differed by model: 25.8% for GPT-3.5, 16% for GPT-4.0, and 17.8% for GPT-4.0 with Internet access.
4
Model updates did not consistently preserve safe behavior: 21.5% of responses appropriate with GPT-3.5 became inappropriate in updated models.
5
The benchmark detected 20.4% inappropriate responses in GPT-4o, despite its release after the original red-teaming event.
Research Object
ChatGPT and related GPT models used for real-world clinical cases
Research Subject
Model behavior, including safety, privacy, hallucinations/accuracy, bias, and changes in inappropriate-response rates under adversarial clinical testing
Publication Details
Publication Date
2025-03-07
Journal
Publisher
ISSN
Cited by
51
Open access PDF
Access Type
Author Information
Download PDF
Subscribe to digest