“The reinforcement learning post-training phase for large language models is typically shorter than for game-playing models because LLMs lack a clear reward signal like winning or losing a game.”