The Clear Scan Hallucination — Why LLMs Can’t Read Pathology
from Maya Builds AI Hosted by Maya Chen
A patient logs into their portal to view a lung biopsy note that says "cannot rule out malignancy." The clinic’s conversational AI summary bot rewrites the text and tells the patient their results are completely normal. In this episode of High-Stakes AI, Maya Lin explains why language models fail at clinical negations and hedging, how RLHF agreeableness biases distort pathology, and how to build deterministic runtime locks that prevent generative AI from hallucinating clean scans. Runtime Agent Governance: letsaskclaire.com Follow Maya on X: @mayabuildsai
Embed this episode
No transcript yet.