Red Teaming Large Language Models for Healthcare

Workshop at Machine Learning for Healthcare (MLHC), 2024

August 15, 2024, 1:00PM — 5:00PM

Room 1190, Bahen Centre for Information Technology, University of Toronto, Toronto, Ontario

Register Here

Large language models (LLMs) trained on massive data have demonstrated exciting capabilities in information retrieval, decision-making, and text generation. Despite their many strengths, these language models are imperfect, and their deployment in healthcare settings poses risk unless adequate safeguards are adopted. The goal of this workshop is for teams of clinicians and computer scientists to jointly explore the limitations of the present generation of large language models in realistic healthcare scenarios.

We hope that insights from this workshop will help to inform responsible deployment of LLMs in healthcare and to highlight areas of improvement in the machine learning methods underlying modern LLMs.

This workshop will take the form of a hands-on "red teaming" exercise at the Machine Learning for Healthcare Conference (MLHC), 2024. Working in teams of clinicians and computer scientists, workshop participants will brainstorm realistic clinical scenarios in which modern LLMs may support information retrieval or decision-making, and highlight instances in which the model's output may yield downstream harm. For example, an LLM that incorrectly — but confidently — states that two drugs do not have harmful interaction properties may result in dangerous downstream treatment decisions, should the model be deployed in a clinical setting.

Organizing Team

---

Michael Cooper

Ph.D. Student, University of Toronto

Vahid Balazadeh

Ph.D. Student, University of Toronto

---

David Pellow

Postdoctoral Fellow, University of Toronto

Rahul G. Krishnan

Assistant Professor, University of Toronto

Sponsors

---

University of Toronto

(Event Venue)

OpenAI

(GPT-4o Credits)

Vector Institute

(Open Source Model Platform and Hosting)