The counter-case on compute demand: slop gets optimized, robots run distilled on-device policies
teortaxesTex · x · 2026-11-21
In a debate over whether silicon demand is limitless, the author argues AI slop will get optimized since human sensory bandwidth is finite, robots will run distilled on-device policies rather than large models like Astra, and workloads like cancer drug discovery are one-time and heavily amortized. Still, he expects the upstream semiconductor chain to be very successful, with plenty of silicon being used.
Related event: Compute Overbuild Debate Flips as Efficiency Gains Threaten Demand(3 posts)→
More from AGI Musings
- AI eats the low-hanging fruit, so the next famous unsolved problems will be harder — yaroslavvb · 2026-11-21
- At least six labs will autonomously build RH-proving tech within a year, predicts author — _aidan_clark_ · 2026-11-21
- Paul Christiano: AI labs should stop advancing until they fix security — Dwarkesh Patel · 2026-11-21
- Prediction: half a dozen labs will autonomously build RH-proving tech within a year — _aidan_clark_ · 2026-11-21
- Blogger declares AI bubble already burst: bubble stocks down 50%, some down over 90% — oran_ge · 2026-11-21
- Machines consumed 7.8x more tokens than humans by September — infra must be rebuilt for agents — AccBalanced · 2026-11-21