daily-hour-news·
🔬Import AI 478: $135K Lean Proofs, ProgramBench Jumps
TL;DR
Jack Clark covers three results: agents formalizing singular-learning math for about $135K in tokens, ProgramBench moving from 0% to 18.5% resolved, and Mirendil-Base, a pretrained LLM built by one researcher for under $10M.
Jack Clark covers three results: agents formalizing singular-learning math for about $135K in tokens, ProgramBench moving from 0% to 18.5% resolved, and Mirendil-Base, a pretrained LLM built by one researcher for under $10M.
Key Points
1
Resolution's agents produced roughly 600K lines of Lean in about a week using ~100B input tokens
Keep reading — it's free
Enter your email to keep reading — plus the best of AI & tech, daily. Free, forever.
Also get
or
Already a member? Sign in
Quick Facts
Import AIJack ClarkProgramBenchautoformalizationLeanMirendilClaude Opus 5.5pretraining
Comments
Subscribe to join the conversation...
Be the first to comment