“The dirty secret in AI is that everything is a data and an eval problem.

The best models have the best data and best internal benchmarks. The mid ones buy a lot of data, not the best, and hillclimb public benchmarks.

(you need a lot of compute too)”

– Stanford CS Professor