Comment by a_c

While LLM models are bad at games, they are perfectly capable of writing a RL agent to train on the game itself.