Scott Alexander, curated
← Back to curation

Why AI Safety Won't Make America Lose The Race With China

Quality
76
Excellent
Claude Shift
56
Moderate
RWI
3
of 10

Summary

Argues that AI-safety regulation (alignment/catastrophe disclosure, not AI-ethics) won't cost the US the AI race with China. Frames the race as three layers -- compute (US ~10x ahead via NVIDIA/TSMC), models (follows compute), applications (China's strength). The actual safety bills (SB53, RAISE) are cheap (~1-2% of a training run) and land on the model layer, where China isn't even competing; they barely touch the application layer that China is betting on. The real lever is export controls: chip smuggling through Singapore/Malaysia turns a 31x flow advantage into 10x, and the loudest anti-safety 'race with China' voices (David Sacks) simultaneously push to let NVIDIA sell advanced chips to China -- which would collapse the US edge from 10-30x to ~2x. That inconsistency reveals the anti-safety position is narrative-driven, not analysis. Goes further: safety work might even SPEED the US up (data-center security vs Chinese espionage; compute-governance chip tags vs smuggling; giving Chinese pro-safety factions cover; pre-empting catastrophe-driven overreaction), and EAs are behind some of the strongest pro-US-supremacy bills.

Why this score

Quality 76 · Excellent. Excellent floor. A detailed, numerate, well-sourced policy intervention whose central move -- the export-controls hypocrisy (people who cry 'race with China' against safety ignore the thing that matters 100x more) -- is a genuine and sharp gotcha, plus the thoughtful 'might even speed us up' turn. Clears Excellent on rigor and reach; held at the floor as a focused-argument essay.

Claude’s paradigm shift 56 · Moderate. Notable. The compute/model/application decomposition and the safety-is-trivial-next-to-export-controls inconsistency are a fresh, non-obvious contribution to the live AI-policy debate.

Real-world impact 3 · Moderate. Moderate. A notable intervention in an active AI-regulation / export-control policy debate; influence stays within the AI-policy professional sphere (no specific legislation verifiably attributable). Niche-professional.