“Dan Houser predicts that achieving the final 5% of human-like quality in LLMs will require 95% of the total development effort, similar to the challenges in perfecting facial animation.”