I say this not because I think his prediction as stated here is necessarily wrong or unreasonable, but because I myself might want to make investment decisions based upon this prediction, and translating a prediction about the future into the correct executions today is not trivial.
Without addressing his argument about AGI-from-LLMs - because I don't have any better information myself than listening to Sutskever on Dwarkesh's podcast - I am somewhat skeptical that the current market price of AI-related assets is actually pricing in a "60-80%" chance of AGI from LLMs specifically, rather than all the useful applications of LLMs that are not AGI. But this isn't a prediction I'm very confident in myself.