“Large Language Models (LLMs), despite being trained on significantly more data, possess only a small fraction of total human capabilities.”