Sorry, but anybody who still says "LLMs just predict text" is embarrassing themselves at this point It is, as OpenAI's roon says, "categorically wrong"