[IshKebab]

I guess LLM's don't have the social pressure to avoid admitting errors. And those sort of interactions aren't common in text so they don't learn them strongly.

Also ChatGPT is trained specifically to be helpful and subservient.