LLM actions are divorced from that reward function, it's not something they consult or consider. Reward function in that context doesn't make sense.