What are we scaling?

We are scaling Dwarkesh beard

I really appreciate that you're giving your honest opinions even though you have so many AI sponsorships

We are scaling uploads boys

In one year Dwarkesh went from totally AGI pilled to pretty much an AI skeptic. Appreciate the intellectual honesty.

I think at this point we are just confused about the term AGI and we should retire the word. The next milestone is models that can be deployed at enterprises and be at least as useful as most human employees. That may or may not require continual learning since many coders think Claude Opus 4.5 is already there without it. That said, Dwarkesh, you might want to talk to your roommate Sholto, he just predicted that continual learning will be solved "satisfactorily" in 2026.

Good on you for putting your sincere thoughts out there. Merry Christmas

We’re in the trough of disillusionment as defined in the Gartner hype cycle

Really thoughtful insights. I did my graduate school on human learning and computational models of it. I no longer work on such research, but one takeaway was that human learning is quick, persistent, and heavily relies on generalization. In fact, most of the time, the primary type of errors to correct for human learners are over-generalization errors (think how kids say words like "gooses"). Humans are much more efficient learners in that we require fewer examples to start generalizing (while making errors), and that we are really tuned to semantic and structural similarities between tasks. These are very different from the approaches of RL in nature.

We were used to seeing yearly reviews from youtubers for movies, tech gadgets, and now it’s AI. What a time to be alive?!

This is a very humbling and grounding view on where we do stand with AI.