Your experience matches mine. Everything I have asked is usually _horribly_ wrong, and even asking things in a different order makes it completely change its responses, even for otherwise binary questions. Even the "snippet" part of a Google search with the same prompt normally contains enough information to contradict it...
I'll also note that when there was hype around Stable Diffusion, one of the images shared around was that of an astronaut riding a horse. If you actually run Stable Diffusion with its default tuning and ask for that prompt, you will get 6 images, of which 5 of them are outright disasters (horses with 6 legs and going downhill from there), and then the 6th image, the only one which could possibly pass as a decent result, is the one that everyone shared and reshared and hyped. Usually other prompts give even more terrible results where there are 0 passable images without extensive tuning. Stable Diffusion now is acknowledged to actually be crap --despite the hype-- and I supposedly need to try the next best thing, whatever that is. But I find myself facing the same situation with ChatGPT 3.5, and now with ChatGPT 4, despite the fact there is no "next best thing", and I don't even know how they could even possible try to fix the problem of it being just wrong.
I'll also note that when there was hype around Stable Diffusion, one of the images shared around was that of an astronaut riding a horse. If you actually run Stable Diffusion with its default tuning and ask for that prompt, you will get 6 images, of which 5 of them are outright disasters (horses with 6 legs and going downhill from there), and then the 6th image, the only one which could possibly pass as a decent result, is the one that everyone shared and reshared and hyped. Usually other prompts give even more terrible results where there are 0 passable images without extensive tuning. Stable Diffusion now is acknowledged to actually be crap --despite the hype-- and I supposedly need to try the next best thing, whatever that is. But I find myself facing the same situation with ChatGPT 3.5, and now with ChatGPT 4, despite the fact there is no "next best thing", and I don't even know how they could even possible try to fix the problem of it being just wrong.