Does Noam Brown think recursive self-improvement will cause an intelligence explosion?

Noam Brown expects AI that automates AI research to bring a significant speed-up but not an overnight explosion. He argues that unlike pure mathematics, AI research is bottlenecked by experiments that must run one after another and by the GPUs to run them. Pressed for a number, he says he could see progress going about 3x faster, which he calls massive given today's pace, while allowing anything from 50% to 10x.

• Brown says OpenAI with 100 times less compute would make less progress even with the world's best researchers, though not 100 times less.
• Dwarkesh Patel argues the opposite case, that steady progress alone could give each lab many Earths' worth of human-level minds by the mid-2030s.

▸In their words

“If you had to put a gun to my head and ask me for a number, I could see things going 3x faster.” Noam Brown · at 38:12 — open in the transcript →

Source: our own transcription of the episode. The full conversation is on Dwarkesh Podcast (2026-09-17). Read the whole episode summary →

▸More from this episode