Future LLMs that is.

AlphaGO has already shown to come up with a completely and unknown playstyle, from training against itself.
I do believe these models can go beyond what is already known.