LLMs are plateauing and the gap between closed vs. open is almost closed!! If you are look at MMLU open-source is caught up to closed source and we are seeing the LLMs plateau It's time to move on to different benchmarks that measure LLM capabilities on hard problems The key