Amen. And LLMs have a bias for producing text. And the more questions I ask it, for example to solve more edge cases, it quickly escalates on the existing suggestion rather than find a new approach.