Superintelligence: Paths, Dangers, Strategies · chapter 4 · id superintelligence-c4-02-a-hardware-overhang-could-exis
“A 'hardware overhang' could exist -- if AGI algorithms are discovered, existing computational infrastructure could immediately support superintelligent performance, enabling a sudden capability jump.”
needs contextconfidence: medium⚠ extracted by pipeline, re-audit pending
Receipts
Wikipedia: AlexNet / Deep Learning Revolution (2012)source alive
Neural network algorithms existed for decades. The combination of GPUs + large datasets + algorithmic refinements produced a sudden performance jump circa 2012 (AlexNet). A genuine narrow-AI example of hardware overhang.
Hoffmann et al. 2022, Training Compute-Optimal Large Language Models (Chinchilla)source alive
More capable systems require more compute, not just better algorithms. Each generation demands more, not less. This directly contradicts the overhang thesis as applied to current paradigms.
Kaplan et al. 2020, Scaling Laws for Neural Language Modelssource alive
Scaling laws show capability improves as a power law of compute. Linear capability gains require exponential compute increases -- the opposite of an intelligence explosion.
This claim is a stable, citable object. If you can falsify a verdict, tell us — corrections are loud here.