yeah, and there's the difference - i'm not a physicist, but i am a programmer, and i see the same level of enthusiasm for ai slop code from inexperienced programmers (and very lazy, over-optimistic, programming-for-the-money-not-the-love "seniors")
but it always ends with "i need someone to fix 16k lines of ai-generated code that _almost_ works" by the time it's more than a toy and actually has to be something useful - kinda like his "pretty damn close to some interesting breakthroughs", where someone who doesn't actually grasp the domain sees near-victory in places where a seasoned pro recognises a wall of defeat that they've seen before
but sure, ai is gonna take over science for us too, just like programming, and that's worked out so well so far