Dwarkesh: The avalanche that hit math is coming for AI — superintelligence by 2030

ColinWPLewis · x · 2026-11-14

Dwarkesh Patel published a long essay explaining why he has dramatically pulled forward his superintelligence timeline. He admits that earlier this year he would have said at least the better part of a decade, but his views shifted fundamentally over the past three months.

Key arguments:

He now believes we may be only a couple of years — possibly a couple of months — from AIs as good as the best humans at AI research.
Automating AI research will compress enormous progress into a single year, so by 2030 (plausibly as early as next year) we'll have AIs at least as good as the best human experts in every field.
The argument starts from AI's rapid progress in verifiable domains like math, which he sees spreading into AI development itself like an avalanche.
Key evidence: METR, an independent evaluation org embedded inside Anthropic for three months, concluded Anthropic's progress rate on well-scoped tasks is 2x what it was a year ago, with a majority of that acceleration attributed to AI itself.

Colin Lewis shared the essay, agreeing with the claim about expert-level AI in every field possibly arriving as early as next year.

Original post →

Dwarkesh: The avalanche that hit math is coming for AI — superintelligence by 2030

More from AGI Musings