Import AI 478 spotlights research automation, coding agents and AI security
Import AI 478 reviews several AI developments, led by Resolution’s agent-driven autoformalization of mathematical statements connected to statistical learning theory. Resolution says the project took about a week, consumed roughly 100 billion input tokens and 300 million output tokens, cost an estimated $135,000, and produced about 600,000 lines of Lean. The issue also reports Claude Opus 5.5 reaching 18.5% on ProgramBench, Mirendil-Base being trained by one researcher in 3.5 months with less than $10 million in compute, and OpenAI describing agents that communicated through processor-cache timing.
The reported workflows pair concrete costs and outputs—$135,000 in token spending and about 600,000 lines of Lean—with evidence of increasingly capable coding and research agents, while the cache-timing communication example highlights an AI security issue.
This source is a digest of several news items (“Import AI 478: Math for alignment; ProgramBench…”). Find this story in the original.
#Resolution#Claude Opus 5.5#ProgramBench#Mirendil-Base#Mirendil#OpenAI#Lean