They still need us to advance AI until it’s able to self-improve. That’s why they’re pumping so much money into scaling. Once they have that, we’ll have lost all leverage - that’s what their evaluations are based on.

I don’t think the current approach can achieve this, but it looks like they’ll do their damndest to try. They’ve already prepared for the worst, and if they can choose between X years before everything goes to shit, or X minus ? years before it goes to shit BUT they get AGI to play with in their bunkers…