points
I've seen different benchmarks come to different conclusions
Benchmarking these models must be an incredibly complex and difficult problem
How can a lay person know which benchmarks actually have good signal?