> pointing to a looming plateau Latest LLM's have already been declining in performance. It's not a plateau and it's not looming. We don't know what the hell

> pointing to a looming plateau Latest LLM's have already been declining in performance. It's not a plateau and it's not looming. We don't know what the hell we are doing. Adding more training data and parameters only degrades LLMs after a certain threshold.