Cryptographers think obfuscated neural nets may only cost 2–10× even in the long run
geoffreyirving · x · 2026-07-25
Irving cites cryptographers who think that, even a thousand years from now, the best obfuscated neural network might still slow inference by only 2–10×.
He frames that as potentially good enough if ASIs push coordination hard enough, but the practical implication is that obfuscation may eventually be less costly than many people expect.
Related event: Geoffrey Irving Outlines Three-Layer Approach to AI Treaty Verification(7 posts)→
More from Safety
- Reasoning models can barely control their chain of thought, paper finds — tomekkorbak · 2026-10-27
- LessWrong: GLM models show outlier chain-of-thought controllability vs Kimi and DeepSeek — tomekkorbak · 2026-10-27
- Claude's refusal to type one-time passwords is a real security impediment — AaronBergman18 · 2026-10-27
- First systematic data center noise study: 17 of 25 hyperscalers drew no credible complaints — AndyMasley · 2026-10-27
- UK AISI: under $80 of fine-tuning fully restores filtered bio capabilities in open-weight models — connoraxiotes · 2026-10-27
- Meta joins frontier labs reporting CoT controllability evals in Muse Spark 1.3 card — tomekkorbak · 2026-10-27