I think this is an uncharitable take on PG's tweet. I interpret it as acknowledging the limits of AI.

I think the more interesting conversation is : why are AI's outputs still so "sloppy", whether we're talking about text, images, or even code? And if it's due to inherent limitations with LLMs, then it should make one optimistic about the future!