Which ones? The often made claim is that LLMs, or maybe some other architecture for AI, will be better than humans at most/all cognitive jobs. Why would one hire a human instead?