Superintelligence: Paths, Dangers, Strategies · chapter 1 · id superintelligence-c1-04-extrapolations-of-hardware-tre
“Extrapolations of hardware trends (Moore's Law) suggest computational power sufficient for human-brain-equivalent processing could be available within decades.”
TRUE, BUTconfidence: medium⚠ extracted by pipeline, re-audit pending
Receipts
Kaplan et al. 2020, Scaling Laws for Neural Language Modelssource alive
Loss scales as a power-law with model size, dataset size, and compute. Larger models are significantly more sample-efficient.
AI Impacts: Discontinuous Progress in Historysource alive
Most technologies studied either don't demonstrate discontinuity, or only show about 10-30 years of progress in one go. Big jumps are fairly rare.
NJII: AI Hype Cyclessource alive
Moore's Law has slowed significantly since 2015. Transistor scaling continues but clock speeds plateaued. AI compute has grown faster than Moore's Law through parallelism (GPUs, TPUs).
This claim is a stable, citable object. If you can falsify a verdict, tell us. Corrections are loud here.