Model Evals
Testing models on real tasks and reporting what the numbers do and don't show.
I asked seven models to sort 1,000 payees into US tax categories. Jev and d1 tied at the top, and d1 cost the least.
See all writing.
Testing models on real tasks and reporting what the numbers do and don't show.
I asked seven models to sort 1,000 payees into US tax categories. Jev and d1 tied at the top, and d1 cost the least.
See all writing.