That "try harder" is often a useful LLM prompt is one of their more confounding aspects.
ChatGPT confidently, and incorrectly, identified this bag. When I said, "no, try again," it got it right!
It wants to use less resources, but ends up using more, because it can't tell when it's wrong.
about 1 month ago