Scaling and the Road to Human-Level AI (Jared Kaplan, AI Startup School, Y Combinator; speaker-labelled transcript)

Jared Kaplan · 16 Jun 2025

Jared Kaplan did not write this page.

Every claim below was made in this piece, quoted word for word and numbered in the order the piece makes them, so you can read it there rather than take our word for it. The sentence above each quote is our reading of the claim, not their wording. Each quote was checked against a stored copy of the page at build time; where the two differ, the quote is the fact.

Our reading
Scaling results gave conviction that AI would keep getting smarter in a very predictable way.
Mental modelExponential growth

Their wordsThese gave us a lot of conviction to believe that AI was just going to keep getting smarter and smarter in a very predictable way.
Our reading
AI progress comes from a simple, systematic way of improving models, not from researchers suddenly getting smarter.

Their wordsIt's not that AI researchers are really smart or they suddenly got smart. It's that we found a very, very simple way of making AI better systematically, and we're turning that crank.
Our reading
Eventually AI models, perhaps millions working together, may be able to do the work of whole human organizations.

Their wordsEventually, we imagine AI models—or millions of AI models perhaps working together—will be able to do the work that whole human organizations can do.
Our reading
AI models need to arrive with the context of someone who has worked inside an organisation for years, not a blank slate.

Their wordsWe need to train AI models that don't just greet you with a blank slate but can learn to work within companies, organizations, governments as though they have the kind of context that someone who's been working there for years has.
Our reading
Products that do not quite work yet are worth building, because the next model may make them work.

Their wordsThat means that if you build a product that doesn't quite work because Claude 4 is still a little bit too dumb, you could expect that there'll be a Claude 5 coming that will make that product work and deliver a lot of value.
Our reading
For the most advanced tasks, humans will be needed in the loop for now.

Their wordsI think that right now human AI collaboration is going to be the sort of most interesting place because I think that for the most advanced tasks you're really going to need humans in the loop.
Our reading
Using AI to integrate AI into the economy quickly probably has a lot of leverage.

Their wordsI think that probably leveraging AI to integrate AI into parts of the economy as quickly as possible, I expect there's just a lot of leverage there.
Our reading
Asking very naive questions goes a long way in a young field like AI.
Mental modelFirst principles

Their wordsI think generally it's really asking very naive dumb questions that gets you very far.
Our reading
How long a task a model can accomplish depends largely on its ability to notice its own mistakes and correct them.

Their wordsI think that a lot of what determines the horizon length of what models can accomplish is their ability to notice that they're doing something wrong and correct it.

Keep reading

The other pieces are the ones whose claims are worded most like this one's. That measures language, not agreement: they may argue the opposite.