26-author study: a hardwired frontier AI training pause is technically feasible
wfithian · x · 2026-10-20
A 26-author interdisciplinary working group led by UC Berkeley's William Fithian argues governments could implement a "hardwired pause" of frontier AI training: phase out advanced training chips in favor of specialized inference-only chips that can't practically train new models, enabling a lasting, verifiable international agreement. A WSJ poll of 1,500 registered voters shows 63% support pausing AI development. Authors span Oxford China Policy Lab, Princeton, Stanford and more; co-authorship doesn't imply agreement with every claim.
More from Infra
- Fused linear DPO loss merged into NVIDIA TileGym: 7.2x less memory, 1.27x faster — blelbach · 2026-10-20
- If GitHub goes down, how does GitHub push code to GitHub to bring GitHub back up? — VeryWellVersed · 2026-10-20
- Meta: Co-Design for AMD GPUs Delivers Up to 1.5x Training Throughput vs Direct NVIDIA Ports — burny_tech · 2026-10-20
- Dev rewrites opencode server in Rust: 100MB vs 2.4GB RAM, session search 30,000x faster — microbass · 2026-10-20
- Muon optimizer tweak with a polynomial grid saves ~0.5% of LLM training compute — burny_tech · 2026-10-20
- Tech Fund estimates $414B in annualized AI revenue by year-end, ex-China — FinanceYF5 · 2026-10-20