Dice are more random, too random for believable results. The nice thing with an LLM is that it knows the real-world probability of every answer to a question, or can make a realistic inference of what that probability will be.

It won’t give actual feedback about your thing, but it will give feedback that appears real.