> As far as AI safety issues go, the solution is probably to fight fire with fire. Have multiple redundant, independent AIs, and the good AIs can fight the bad AIs, and hopefully, having access to more hardware, the good AIs will win.

And they must be able to run on consumer hardware to guarantee this independence.

I'll take everyone on earth having more capability instead of 3-4 labs controlling said capability with a nonzero chance of said capability all going negative at the same time.