“Aidan Gomez believes the performance gains from purely scaling up the size of LLMs, data, and compute are beginning to saturate.”