Hacker News Reader

The benchmarks are what allows the AI researchers to improve the models. The AI doesn't get better on its own.

AI should be free to evaluate itself free from arbitrary human constraints like benchmarks imo.