Medical large language models are vulnerable to data-poisoning attacks.

The adoption of large language models (LLMs) in healthcare demands a careful analysis of their potential to spread false medical knowledge. Because LLMs ingest massive volumes of data from the open Internet during training, they are potentially exposed to unverified medical knowledge that may include deliberately planted misinformation. Here, we perform a threat assessment that simulates a data-poisoning attack against The Pile, a popular dataset used for LLM development.

A Patient Interview-Based Needs Assessment and Evaluation of Experiences with Periviable Pregnancies.

Given the complex ethical and emotional nature of births during the periviable period for both health care providers and families, this investigation sought to identify strategies for improved counseling of pregnant patients facing preterm birth at the cusp of viability at a tertiary care center in Hawai'i. As part of a larger quality improvement project on periviability counseling, 10 patients were interviewed during either individual or small focus groups using a progression of hypothetical scenarios. Interviews were analyzed independently by 3 investigators to identify themes of patient experience and potential areas for improvement when counseling patients who are carrying periviable pregnancies.