AI accelerates cyber, not math or AI itself; SPADE auto-generates training environments; Hawkeye writes better GPU kernels

Import AI 470: No rights for machines; automating environment generation with SPADE; and building better GPU kernels with Hawkeye

METR finds LLMs dramatically accelerate cyber vulnerability discovery, mildly boost math, and barely speed up AI research itself. SPADE lets a 30B model alternate between designing executable environments and solving them, gaining +8.1 on games and +5.3 on tool-use tasks. The post doesn't disclose Hawkeye's specific performance numbers, only that well-documented unit tests help agents write better GPU kernels.