February 13, 2026
Genius or Geniu$$?
Dario Amodei – "We are near the end of the exponential"
A 'genius datacenter' in years? Fans cheer, skeptics want receipts
TLDR: Dario Amodei says AI progress is nearing a limit and we’re close to “genius datacenters.” The comments erupt: believers vs skeptics demand receipts, a side quest on Dwarkesh’s sudden rise, and jokes about WeWork-for-robots—making this a showdown over hype, proof, and who’s steering the AI narrative.
Anthropic’s Dario Amodei dropped a spicy claim: we’re “near the end of the exponential,” and just a few years from a country of geniuses in a data center. Cue the comment section going full reality TV. Some hailed Dario as a true believer, others slammed the hype and asked for proof. One skeptic posted failing to understand the exponential again, basically yelling: “calm down.” Another demanded “citation needed” for the idea that “nobody disagrees we’ll hit AGI”—AGI means artificial general intelligence, a machine that can do most human jobs.
Meanwhile, the show’s host Dwarkesh becomes a subplot: is he a brilliant curator or just AI-prepped and mysteriously boosted? A commenter even asked who his “patron” is, while another rushed in with a wholesome defense: “Dwarkesh is a wonderful, phenomenal human.” The tech itself got roast-level scrutiny too—folks debated whether scaling up RL (reinforcement learning, basically teaching AIs with rewards) is real progress or just vibes, and asked the obvious: if AGI is imminent, why not buy more computer power now?
Jokes flew: “We’ll need a new word after ‘genius’,” and “genius datacenter” got WeWork-for-robots memes. Sponsor bingo added fuel—RL task marketplaces, backdoored model bounties, and a debit card for your AI agent? The crowd alternated between awe and side-eye, and that tension is the whole show.
Key Points
- •Dario Amodei asserts AI capability growth has followed a long-running exponential trajectory and believes we are nearing its end.
- •He observes models have advanced from high school to professional/PhD-level tasks, with code capabilities particularly strong.
- •Current RL-era scaling lacks widely known public scaling laws; Amodei references a broader compute-centric view he formed around 2017 (GPT-1 era).
- •The episode examines AI’s economic diffusion, frontier labs’ compute investment (including whether Anthropic is underinvesting), and profitability models.
- •Regulatory impacts and US–China competition in achieving nation-scale AI capability are discussed, with detailed topic timestamps provided.