LLMs Think about unrelated data when recommending health-related treatment plans Researchers uncover nonclinical information in client messages — like typos, added white House, and colorful language — cuts down the accuracy of an AI product. Study total story → Far more information