LLMs are not "thinking" machines. The tech is not capable of that, as much as people want to think that reinforcement learning will lead to sentience.