Comment by OGWhales

I've found this a really useful strategy in many situations when working with LLMS. It seems odd that it works, since one one think its ability to give a good reply to such a question means it already "understands" your intent in the first place, but that's just projecting human ability onto LLMS. I would guess this technique is similar to how reasoning modes seems to improve output quality, though I may misunderstand how reasoning modes work.