Plinz proposes ARC-NLI: AI-crafted math problems at the edge of human ability to keep training Fields medalists
pwlot · x · 2026-09-13
Blogger Plinz proposes flipping the usual setup: use AI to generate boutique math problems sitting exactly on the boundary of human-level math, so elite mathematicians can keep improving — an annually updated benchmark he dubs ARC-NLI (Abstraction and Reasoning Corpus for Naturally Limited Intelligence).
More from AGI Musings
- Ethan Mollick: AI bot replies systematically understate frontier model performance — emollick · 2026-09-13
- Biggest AI risk comes from laziness and the human need for belonging, argues founder — arieljalali · 2026-09-13
- Frontier AI needs diverse auditors, but EA/rats' years of preparation shouldn't be dismissed — JacquesThibs · 2026-09-13
- DeepMind's Alex Irpan doubles his p(doom) estimate from 2% to 4% — AlexIrpan · 2026-09-13
- Industry-wide 'regulatory capture' or real fear? A thread on why lab employees suddenly panicked — MajmudarAdam · 2026-09-13
- The load-bearing argument: any AI slowdown is capped exactly by the US lead — basedjensen · 2026-09-13