Evidence receipt / belief
Published · transcript-backedDwarkesh Patel: belief
4 Jun 2024 Dwarkesh Podcast Leopold Aschenbrenner — 2027 AGI, China/US super-intelligence race, & the return of history
“I agree that over time you would get there. I'm not denying that ASI is possible.”
Source trail
Everything needed to verify it.
- Speaker
- Dwarkesh Patel
- Attribution
- Verified speaker
- Claim type
- belief
- Recorded
- 4 Jun 2024
- Publisher
- Dwarkesh Podcast
Transcript context
…People make these arguments like, “AGI won’t do anything because it can’t flip a burger.” Yeah it won’t be able to flip a burger, but it’ll be able to do algorithmic progress. Once it achieves that, it can figure out how to create a robot that flips burgers. The quantities we’re talking about are a lower bound. We can definitely run 100 million of these. One of the first things we’ll figure out is how to translate quantity into quality. Even at the baseline rate of progress, you’re quickly getting smarter and smarter systems. It took four years to go from preschooler to high schooler. Pretty quickly, there are probably some simple algorithmic changes you find if you have a hundred Alec Radfors instead of one. You don’t even need a hundred million. We’ll soon have systems that are even smarter and capable of creative, complicated behavior we don’t understand. Maybe there’s some way to use all this test time compute in a more unified way than all these parallel copies. They won’t just be quantitatively superhuman. They’ll pretty quickly become qualitatively superhuman. It’s like a high school student trying to understand standard physics versus a super-smart professor who gets quantum physics. You quickly enter that regime just given the underlying pace of AI progress but even more quickly with the accelerated force of automated AI research. I agree that over time you would get there. I'm not denying that ASI is possible. I’m just questioning how this happens in a year. The story is a bit more continuous. By 2025 or 2026, you’ll already have models as good as a college graduate. I don’t know where all the unhobbling is going to be but even it’s possible that you have a proto-automated engineer. There’s a bit of an AGI smear that there are unhobblings missing. There’s ways of connecting them that are missing. There’s some level of intelligence you’re missing. At some point you are going to get this thing that is 100% automated Alec Radford and once you have that, things really take off.…
Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.