Indeed, analysis is a bit more loose in its arguments, and so I've found LLMs tend to make more mistakes there.