“Ilya Sutskever believes it is unusual that Large Language Models (LLMs) do not utilize value functions in their training.”