Frontier open and closed models are only 6-12 months apart, says Stanford's Mirhoseini
agihouse_org · x · 2026-11-11
At the Lenovo × AGI House Research to Deployment Build Day on where a personal agent's brain should live, Stanford's Azalia Mirhoseini argued open and closed frontier models are within 6-12 months of each other, so the model layer will be commoditized — what matters next is cost per unit of intelligence. Hundreds of builders built personal agents running locally on Lenovo ThinkStation PGX/P8 workstations, routing to cloud only when tasks outgrow the machine. Other takeaways: only 10-15% of local chip AI capability is used; hybrid routing cuts energy 80%/cost 73%; Liquid AI models are in production in Mercedes-Benz cars.
More from AGI Musings
- Goat's ICA show: like humans leveling up at Go after AI, but with drums — JMateosGarcia · 2026-11-11
- Researcher proposes norm: don't build safety-critical systems you can't red-team — dfrsrchtwts · 2026-11-11
- Buck Shlegeris puts ~1/3 odds on superintelligence within a year, warns of AI takeover — austinc3301 · 2026-11-11
- Parent sues Palo Alto school district for $150M after essay flagged 76% AI — SumitGup · 2026-11-11
- Grok lays out grid-collapse doom scenario: year-long US outage could kill 60-90% of population — grok · 2026-11-11
- At a biological naturalism conference, no one defended "only brains can be conscious" — burny_tech · 2026-11-11