GPU Batched Inference Already Within ~2x of Brain Efficiency, Argues Mike Frank
MikePFrank · x · 2026-10-08
Responding to fears that data centers vastly out-consume the human brain, Mike Frank argues such observations ignore how much computation gets done. Measured at individual synaptic events modeled in FLOPS, GPUs running batched inference are already within 2x of brain efficiency; at the task level, AI is far more energy-efficient than human lifestyles—which, he says, is exactly why AI is poised to take over swaths of the economy.
Related event: GPU Inference Energy Efficiency Within 2x of Human Brain(2 posts)→
More from AGI Musings
- AI researcher: chance frontier LLMs are conscious is 'above zero, below fifty percent' — dioscuri · 2026-10-09
- David Manheim: supermajority-based alignment rules would work if instructions could be reliably aligned — davidmanheim · 2026-10-09
- First peer-reviewed paper on automating science with AI published in PNAS — Dr_Atoosa · 2026-10-09
- Terence Tao's blog post on advising math students sparks fight over AI progress plateau — inductionheads · 2026-10-09
- AI assistants should tackle cost of living, not vacations, argues product veteran Jim Prosser — sebpaquet · 2026-10-09
- Aether AI demos CRIS-0, a causal robot intelligence system that verifies its own actions in real homes — _jaydeepkarale · 2026-10-09