Exactly. People keep shoehorning Large Language Models into non-linguistic domains, and that’s dangerous. Human language, with respect to the training sets used, is inherently subjective and imperfect. Healthcare is very fault-intolerant.