Banning custom AI chips could yield ~10x compute reduction, argues ML engineer
cephaloform · x · 2026-08-23
In a debate over AI bans, @cisfemale argues a full ban is impossible, but banning customized training hardware could achieve roughly a 10x compute reduction if policymakers wanted it. She adds that "ban AI tech" advocates get uncomfortable when shown how simple neural network training is—derivable from middle-school algebra via low-order polynomial fits stacked many times over.
More from AGI Musings
- Max Hodak: AI models and brains share underlying geometry for concepts — saranormous · 2026-08-23
- Criticizing the demand to predict every form of AI stupidity in advance — jd_pressman · 2026-08-23
- I never denied AI extinction risks, but predicting every failure mode is impossible — jd_pressman · 2026-08-23
- Non-researcher view: AGI requires inverting architecture to embrace randomness over determinism — Few_Grass_1054 · 2026-08-23
- P vs NP analogy explains differing views on AI utility — CatAstro_Piyush · 2026-08-23
- 2026 International AI Safety Report: Limited Evidence of AI Manipulation, Positive Learning Impact — flowersslop · 2026-08-23