# Dario Amodei: AI Progress Is On Schedule—We're Just Blind to It

*The Builder Weekly* · https://www.youtube.com/watch?v=hc2sfkuIVbc

## Summary

In this interview excerpt, Dario Amodei discusses why he believes AI progress is on track and near the end of the exponential, even though the public largely fails to recognize it. He revisits his 2017 "big blob of compute" hypothesis, explaining that only a few factors—compute, data quantity and quality, training duration, scalable objective functions, and numerical stability—really matter. He argues that RL scaling now mirrors pre-training scaling, and addresses a Richard Sutton-inspired objection that needing bespoke environments suggests a missing core human learning algorithm.

## Key points

- Amodei says the most surprising thing is the lack of public recognition of how close we are to the end of the exponential.
- He still holds the same "big blob of compute" hypothesis he wrote down in 2017, around when GPT-1 came out.
- The hypothesis lists seven things that matter, including raw compute, quantity of data, quality and distribution of data, training duration, and an objective function that can scale to the moon.
- Pre-training scaling laws have continued to deliver gains, and RL now shows the same log-linear scaling behavior on tasks like math contests (AIME).
- The interviewer raises Richard Sutton's view that needing billions of dollars of compute and bespoke environments hints we lack a core human learning algorithm.
- Amodei responds that the puzzle may be genuine but probably doesn't matter, and calls it a red herring to treat RL as different from pre-training in this respect.
- He notes GPT-1 was trained on a narrow distribution of text, reportedly fan fiction, unlike earlier models trained on standard language modeling benchmarks.

## Notes

### Public blindness to AI progress
- Amodei's biggest surprise is the lack of public recognition of how close we are to the end of the exponential.
- He finds it "wild" that people inside and outside the bubble keep discussing the same tired political issues while this is happening.

### The scaling hypothesis today
- Three years ago, pre-training scaling laws were publicly visible across many orders of magnitude of compute.
- Now there is RL scaling, but no publicly known scaling law for it, and it is unclear whether RL teaches skills or meta learning.

### The big blob of compute hypothesis (2017)
- Written around when GPT-1 came out, when the field also included robotics, separate reasoning work, and RL successes like AlphaGo, Dota, and AlphaStar.
- Claims clever techniques and "we need a new method" thinking don't matter much; only a few things do.
- Seven factors listed:
  - Raw compute
  - Quantity of data
  - Quality and distribution of data (needs to be a broad distribution)
  - How long you train for
  - An objective function that can scale to the moon (pre-training is one; RL with objective rewards like math and coding, or subjective rewards like RLHF, is another)
  - Two factors around normalization/conditioning for numerical stability, so the big blob of compute flows in a laminar way
- Amodei says he has seen nothing that contradicts this hypothesis.

### RL scaling
- Pre-training gains have continued; this is now widely reported.
- Other companies have published results showing log-linear improvement on math contests like AIME as RL training time increases.
- Anthropic sees the same scaling across a wide variety of RL tasks, not just math.

### The Sutton objection
- The interviewer paraphrases Richard Sutton: something with the true core of human learning wouldn't need billions of dollars of compute and bespoke environments to learn Excel, PowerPoint, or web browsing, suggesting we're scaling the wrong thing.
- Amodei says this bundles together things that should be thought of differently; the puzzle is genuine but probably doesn't matter.
- He calls it a red herring to say RL is any different from pre-training on this point, noting GPT-1 itself was trained on a narrow distribution (reportedly fan fiction, about a billion words).

## Transcript

[00:00] what the most surprising thing has been. The most surprising thing has been the lack of public recognition of how close we are to the end of the exponential. To me, it is absolutely wild that, you know, you have people, you know, within the bubble and outside the bubble, you know, but but you have people talking about these these, you know, just the same tired old hot button political issues and like, you know, or or around us we're like near the end of the exponential. I I want to understand what the exponential looks like right now because the first question I asked you when we recorded 3 years ago was, you know, what's up with scaling? How what does it work? >> [music] >> Um I have a similar question now, but I

[00:41] feel like it's a more complicated question because at least from the public's point of view, Yes. 3 years ago there were these, you know, well-known public trends [music] where across many orders of magnitude of compute you can see how the loss improves. And now we have RL scaling and there's no publicly known scaling law for it. It's not even clear what exactly the story is of Is this supposed to be teaching the model skills? Is it supposed to be teaching meta learning? Um what [music] is the scaling hypothesis at this point? >> Yeah, so so I have actually the same hypothesis that I had even all the way back in 2017. So in 2017, I think I talked about it last time, but I wrote a doc called the the big blob of compute

[01:18] hypothesis. And and you know, it it wasn't about the scaling of language models in particular. When I when I wrote it, GPT-1 had had just come out, right? So that was, you know, one among many things, right? There [music] was back in those days there was robotics, people tried to work on reasoning as a separate thing from language models, there was [music] scaling of the kind of RL that happened that, you know, kind of happened in AlphaGo and uh you know, that that happened at Dota at OpenAI and um you know, people remember StarCraft at DeepMind, you know, AlphaStar. Um so uh it was written as a more general document. And and the specific thing I said was the following, that and you know, it's it's

[01:57] very, you know, it sudden put out the bit bit of a lesson a couple years later. Um uh but, you know, the the hypothesis is is basically the same. So so what it says is all the cleverness, all the techniques, all all the kind of we need a new method to to do something like that doesn't matter very much. There are only a few things that matter and I think I listed seven of them. One is like how much raw compute you have. The other is the quantity of data that you have. Then the third is kind of the quality and distribution of data, right? It needs to be a broad broad distribution of data. The fourth is, I think, how long you train for. Um the fifth is you need an objective function that can scale to the moon. So the

[02:40] pre-training objective function is one such objective function, right? An another objective function is, you know, the the kind of RL objective function that says like you have a goal, you're going to go out and reach the goal. Within that, of course, there's objective rewards like, you know, like you see in math and coding and there's more subjective rewards like you see in RL from human feedback or kind of higher-order higher-order versions of that. And and then the sixth and seventh were things around kind of like normalization or conditioning like, you know, just getting the numerical stability so that kind of the big blob of compute flows in this laminar way instead of instead of running into

[03:18] problems. So that was the hypothesis and it's a hypothesis I still hold. I I don't think I've seen very much that is not in line with that hypothesis. And so >> [music] >> the pre-training scaling laws were one example of what of of of of kind of what we see there. And indeed those have continued going. Like, you know, uh you know, I think I think now it's been it's been widely reported like, you know, we feel good about pre-training. Like pre-training is continuing to give us gains. What has changed is that now we're also seeing the same thing for RL, right? So we're seeing a pre-training phase and then we're seeing like an RL phase on top of that. Um and with RL it's it's actually just the same like, you know, even even other

[04:03] companies have have published um uh um like um you know, in some of their in some of their releases have published things that say, "Look, you know, we trained the model on math contests, you know, AIME or or the kind of other things and, you know, how well how well the model does is log-linear and how long we've trained it." And we see that as well and it's not just math contest, it's a wide variety of RL tasks. And so we're seeing the same scaling >> [music] >> in RL that we saw for pre-training. Um you mentioned Rich Richard Sutton and the bitter lesson. I interviewed him uh last year and he is actually very non-LLM-pilled. And if I'm if I I don't know if this is his perspective, but one way to paraphrase

[04:48] this objection is something like, "Look, something which possesses the true core of human learning would not require all these billions of dollars of data and compute and these bespoke environments to learn how to use Excel or how does an account, you know, how to how to use PowerPoint, how to navigate a web browser. And the fact that we have to build in these skills using these RL environments hints that we're actually lacking this core human learning algorithm. Uh and so we're scaling the wrong thing. And so yeah, that that that is a good question. Why are we doing all this RL scaling if we do think there's something that's going to be human-like in its ability to learn on the fly? Yeah, yeah. So I think

[05:28] I think this kind of puts together several things that should be kind of thought of thought of differently. Yeah. I think there is a genuine puzzle here, but it it may not matter. Um in fact, I would guess it probably it probably doesn't matter. So let's take the RL out of it for a second cuz I actually think RL and it's a red herring to say that RL is any different from pre-training in this matter. Um so if we if we look at pre-training scaling, um it it was very interesting back in, you know, 2017 when Alec Radford was doing GPT-1. If you look at the models before GPT-1, they were trained on these data sets that didn't represent a wide, you know, distribution of text, right? You had like, you know,

[06:13] these very standard, you know, kind of language modeling benchmarks. And GPT-1 itself was trained on a bunch of I think it was fan fiction, actually. Um but, you know, it was it was like literary, you know, it was like literary text, which is a very small fraction of the text that you get. And what we found with that, you know, and in those days it was like a billion words or something. So small data sets and represented a pretty narrow distribution, right? Like a narrow >> [music]