Public transcript

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

By: The Builder Weekly
Length: 7 min
Language: English
Text from: the video's captions
Transcribed: 13 October 2026

Read the full transcript Open on YouTube

Notes in English.

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.

Questions it answers

00:00Surprise at public blindness

What has surprised Dario Amodei most about AI progress?

The lack of public recognition of how close we are to the end of the exponential. He finds it wild that people keep discussing the same tired political issues while this is happening.

Amodei believes we are near the end of the exponential.
He is surprised the public doesn't recognize this.

00:41The scaling question today

Why is the scaling question more complicated now than three years ago?

Pre-training scaling laws were publicly visible then, but now there is RL scaling with no publicly known scaling law, and it's unclear whether RL teaches skills or meta learning.

RL scaling lacks a publicly known scaling law.
The purpose of RL—skills vs meta learning—is unclear.

01:18The big blob of compute hypothesis

What is Amodei's scaling hypothesis and does it still hold?

His 2017 "big blob of compute" hypothesis says clever techniques don't matter much; only about seven factors matter, including compute, data quantity, quality and distribution of data, training duration, scalable objective functions, and numerical stability. He says he still holds it and has seen nothing contradicting it.

Written in 2017, around GPT-1's release.
Only a few factors matter, not clever methods.
He still holds the hypothesis today.

04:03RL scaling mirrors pre-training

Is RL scaling working the same way pre-training did?

Yes. Pre-training gains have continued, and other companies have published log-linear improvements on math contests like AIME with longer RL training. Anthropic sees the same scaling across a wide variety of RL tasks.

Pre-training is still giving gains.
RL shows log-linear scaling with training time.
This holds beyond math, across many RL tasks.

04:48The Sutton objection

Does needing bespoke RL environments mean we lack a core human learning algorithm?

The interviewer paraphrases Richard Sutton's concern that needing massive compute and bespoke environments to learn Excel or PowerPoint hints we're scaling the wrong thing. Amodei says the puzzle is genuine but probably doesn't matter.

Sutton is described as very non-LLM-pilled.
The objection suggests we're missing a core learning algorithm.
Amodei thinks the puzzle probably doesn't matter.

05:28RL vs pre-training and GPT-1

Why does Amodei call the RL objection a red herring?

He says it's a red herring to treat RL as different from pre-training on this point, noting that pre-training models like GPT-1 were also trained on small, narrow data distributions—reportedly about a billion words of fan fiction—rather than broad text.

RL isn't fundamentally different from pre-training here.
GPT-1 was trained on a narrow distribution of text.
Earlier models used narrow benchmark data sets.

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).

Text from the video's own captions. Summary and notes written by AI from the transcript, so check anything important against the recording.