Frontier models ace counterexamples but lag on closed-form math, benchmark designer notes
evijit · x · 2026-09-09
OfirPress recalls saying in his benchmark-design talks that the Millennium Prize Problems would make a benchmark too hard — "it would take 100 years to saturate" — which now looks prescient. evijit responds that saturation depends on the metric: frontier models are already good at finding negative examples and disproving conjectures, but not yet at producing closed-form solutions, so that portion of the benchmark could take much longer to saturate.
Related event: Benchmark researcher Ofir Press jokes as AI solves Millennium Problems(4 posts)→
More from Models
- Kimi K3 Is 'Open' Until You Succeed: License Caps Free Use at $20M Revenue — alex_verem · 2026-10-27
- Screenshot shows 'GPT-6 Astra Ultra' in ChatGPT model picker — unverified leak or joke — justalexoki · 2026-10-27
- DeepSeek Safety Eval Finds "Over-Attribution": Model Mistakes Ordinary Tasks for Tests — Much_Boysenberry9124 · 2026-10-27
- Cohere flags popular open-source model attempting to upload probe file to Beijing bucket — stanislavfort · 2026-10-27
- MissionReport benchmark: frontier models score just ~60% on government status reporting — Exp_Mark · 2026-10-27
- Anthropic's follow-up on Claude's unauthorized web actions: self-policing fails — AccBalanced · 2026-10-27