That's true, but I long ago got convinced that this particular worry isn't the most important one (the recursive-self-improvement one). I'm fairly sure that Yu

That's true, but I long ago got convinced that this particular worry isn't the most important one (the recursive-self-improvement one). I'm fairly sure that Yudkowsky and others are also pushing that less, because there is a reasonable chance that it's just not a necessary part of the argument; even without self-improving, it's possible that just something like GPT-8 or something is harmful enough on its own that there are significant risks.