Person
1 entry · 28 January 2022
Asking a model to show its working improved reasoning benchmarks sharply, with no retraining — the seed of the later reasoning models.
Ideas & essays · Benchmarks & progress