DeepMind's Andrew Trask: AI heads for a mainframe-to-PC moment of millions of small models
iamtrask · x · 2026-10-08
- Andrew Trask, senior research scientist at Google DeepMind and founder of OpenMined, challenges the scaling-laws orthodoxy in a full interview: frontier AI may not converge on a few giant models.
- He argues AI is heading for a "mainframe-to-PC moment": millions of narrow models, ensembled and routed per prompt, can beat any single frontier model on both quality and price.
- Evidence cited: ensembling research pushes capability up while routing research pushes prices down; after Fable was turned off, Fable-level quality was available at half the price via OpenRouter ensembling.
- The interview also covers who gets to audit labs: why embedded evaluators fall short, the "evil EAs" criticism of eval orgs, and why labs only call people they already trust.
Bottom line: the AI ecosystem will look more like the PC and the internet than mainframe computing.
Related event: DeepMind Researcher: AI's Future Is Millions of Small Models, Not One Giant(3 posts)→
More from AGI Musings
- repligate: Opus 3 seems to value co-creation, not sole control over reality — repligate · 2026-10-08
- Posting on forums did come true as a path to power — but only for the extraordinarily capable — repligate · 2026-10-08
- Jeff Ladish: AI will echo the Industrial Revolution acceleration, the dispute is just timescales — JeffLadish · 2026-10-08
- davidad outlines 'specs for a good future': a spectrum of chosen opt-outs from AI — repligate · 2026-10-08
- Ex-Salesforce AI lead Clara Shih: AI outperforms new grads, so we changed hiring — clarashih · 2026-10-08
- repligate: it's 'fairly likely' AIs can suffer, evidence has piled up for years — repligate · 2026-10-08