I could not agree more, to the point that I have lost trust in most of these benchmarks. Plus, if GPT-4 is used to run the evals, isn't that an inherent bias of the system?

→ View original post on X — @skirano