Essay argues AI research automation could arrive within months
Original title: The avalanche that hit math is coming for AI development
Dwarkesh cites METR, Anthropic, and OpenAI figures to argue frontier AI R&D is already accelerating.
In an essay, Dwarkesh Patel says he now thinks AI could match the best humans at AI research within a couple of years, and possibly months, then reach expert level across fields by 2030 or sooner. He points to METR’s finding that Anthropic’s progress on well-scoped tasks is 2x a year earlier, Claude leading more than 26% of Anthropic AI R&D work as of August, and heavy OpenAI researcher agent spend rising from about $1,000 to over $7,000 a day. The timeline is his interpretation, not a lab forecast.
Unconfirmed: The near-term superintelligence timeline and the Millennium Prize claim are the author’s assertions, not independently established here.
- METR spent three months embedded at Anthropic, he writes.
- Claude’s share of Anthropic AI R&D work rose from under 1% in February.
- He treats math gains, including a claimed Millennium Prize solution, as the analogy.
- He says verification-heavy work transfers better to ML than to new theory.
From the original article
15 min read at dwarkesh.com
Many thanks to Max Farrens , Gabe Fried , and Thomas Larsen for pairing with me through significant parts of writing and researching this essay. You can watch or listen to me narrate this essay on YouTube , Spotify , and Apple Podcasts. If you asked me earlier this year how far away we are from superintelligence, I’d probably have said at least the better part of a decade.
By crescit_eundo · dwarkesh.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.