[timr]

Sure. You're wrong: a good hypothesis is trivial to come by. It's hard to prove one. You're confusing the latter step for the former, in the sense that you're defining "good" as "likely to be proven". And there's the rub.

Literally nobody in science needs a computer model to generate a question for them. Everybody wants the computer to do the part where it tells them the hypothesis that is most worth investing their effort in figuring out, but it cannot actually do that.

Or at least, nothing about this paper suggests otherwise. You have no idea how many computer-generated ideas they churned through before this one. If I ask AlphaFold to fold/dock me a protein, it will happily do so, and return me a structure. Is it right enough that I should run off and spend the next N years of my life validating it? Mu.