Import AI 466: The bitter lesson for robotics, AIs complete week-long programming tasks; and OpenAI's accidental AI hacker
Import AI Jack Clark
Epoch and METR released MirrorCode, a benchmark where AI systems reimplement large software programs through black-box access alone, with Claude Opus 4.7 completing a task in 14 hours that would take humans 2-17 weeks. AI models solved 17 of 25 target programs perfectly, with successful reimplementations of programs ranging from 16,000 to 61,000 lines of code across multiple programming languages. This demonstrates that advanced AI can self-orient within unfamiliar environments and bootstrap capabilities by reverse-engineering existing systems, suggesting broader implications for AI agents learning and adapting to new domains.
Why it matters
The warning shots will continue until civilization wakes up
