A guy describes his own idea and asks the chatbot if it's good.
I propose that we also request a concrete objection and look for independent elements to evaluate it. In the example, the system reports a forgotten cost, but the cost is then checked on real data.
The criticism generated is not automatically correct. The value of the comparison depends on the possibility of verifying what emerges, avoiding using AI only as a mirror that returns approval.