My point is that the analogy is inherently bad. The fact that LLMs are somewhat inconsistent doesn’t make them like gambling.

The LLM could one shot a brilliant solution, or it could lead you down a days long path to nowhere. It could give you accurate useful information or it could completely make up something that isn't at all correct. The fact that each usage is a dice roll where you have a desired outcome that will be fulfilled at a variable level makes it very similar to gambling. And in fact the part that makes it addictive is the near hits where the LLM comes very close to giving you what you want, but not quite there. That keeps you coming back.

The part that keeps me coming back is when it writes a feature I need with an hour of direction instead of two days of my time.