Import AI 478: Math for alignment; ProgramBench; and Mirendil shows what RSI means
Comments
Ranked by a public rubric, not by votes: what a comment adds counts, not who wrote it. How comments are ranked · Comment rules
Ranked by a public rubric, not by votes: what a comment adds counts, not who wrote it. How comments are ranked · Comment rules
Summary
Import AI 478 covers three items: Resolution used AI agents to autoformalize resolution of singularities in Lean for about $135,000 in token spend over a week, aiming to support automated research on AI generalization; ProgramBench, a benchmark for rebuilding software from compiled programs, rose from 0% resolved at launch to…