Yudkowsky, like Altman, isn't a great public speaker*: too close to the in-group, not aware enough of how the words he uses are understood by people who don't already Get It. Ironically for AI safety, I think Yudkowsky and Altman are, if not on the same page, extending the metaphor they're on the same chapter.

Some random people with a gun and a Molotov aren't even the same (metaphorical) book.

* likely still better than me though, even on this specific measure. But even being the ten thousandth best speaker on the planet, out of 8 billion, leaves you at a huge disadvantage compared to the best.

> bombing data centers

At the risk of demonstrating the exact mistake I've just accused Yudkowsky and Altman of:

With B-52s, not as a DIY job with home-made Molotovs.

If you start with the claim "AI has the potential to cause as much harm as nuclear weapons", and "the USA already uses B-52s to enforce the non-proliferation treaty", this follows naturally.

If you're not willing to call on your representative to sign a binding international treaty to stop data centres that forcefully, talking about "stopping AI" or "pausing AI" seems hollow, because even if your government agrees to not build data centres near you as a result of low-grade domestic terrorism, in the absence of a credible threat to use a B-52 on someone else's sovereign territory there's nothing that you and your flaming rag in a bottle of petroleum distillate can do about them being built outside your country, in exactly the same way and for exactly the same reasons that German public opposition to nuclear weapons completely failed to influence North Korea.