Discussion about this post

User's avatar
Michael Simmons's avatar

Great summary.

I listened to the original podcast where Noah Smith talked about this and then read a bunch of later analysis.

It makes sense on a theoretical level, but it also feels really hard to estimate what the cost of compute will be in the future. On the one hand, there is almost unlimited need for more compute if the AI is smart enough.

On the other hand, we know that AI right now is incredibly inefficient compared to the human brain. And that all the AI companies have been making AI more efficient at a drastic pace.

What frameworks do you use to think about the price of compute in the future?

Expand full comment
Solomon Maxwell's avatar

Nobody serious is saying we won’t. The question is how many, and how economies and governments will have to change so a huge spike in unemployment won’t lead to mass poverty.

Expand full comment
3 more comments...

No posts