I'll add to the points made in this article that we still don't completely know how LLM AI works and can't explain the leaps in capability made after a providing a certain amount of training data.
Sam Altman has talked about this and Eliezer Yudkowsky shares some thoughts with Lex Fridman: