Summary
Podcaster Dwarkesh Patel explains why he has shortened his timeline for superintelligence from the better part of a decade to a couple of years, possibly months, for AI that matches the best humans at AI research. He argues that automating AI research could compress so much progress into a single year that AIs at least as good as top human experts in every field could arrive by 2030, and plausibly as early as next year.
The case begins with evidence that AI is already accelerating its own development. METR, after three months embedded at Anthropic, concluded that Anthropic’s progress on well-scoped tasks is twice what it was a year ago. OpenAI reported its heaviest researcher users going from about $1,000 to over $7,000 a day on internal coding agents, and Anthropic said Claude now leads 26% of its AI R&D work, up from under 1% in February. Patel also traces the rapid rise of AI in mathematics, from failing four-digit addition in 2023 to an IMO gold medal in 2025 and a reported Millennium Prize problem solution in 88 hours.
He then addresses the objection that hard-to-verify domains will lag, arguing that scaling reinforcement learning with vastly larger, more realistic environments and better learning algorithms will let AIs ramp up quickly in new contexts, with skills transferring from verifiable tasks to general capabilities.
📺 Source: Dwarkesh Patel · Published November 13, 2026
🏷️ Format: Opinion Editorial
