More recently, I find ChatGPT to become increasingly unreliable. It makes up almost every second answer, forgets context, or is just downright wrong. Maybe I am used these days more and more to dump huge texts for context into the prompt, as aistudio allows me. Maybe ChatGPT isn't as good as with such information. Gemini/Aistudio will stay on track even with 300k tokens consumed, it just needs a little nudge here and there.

FWIW, I found things improved greatly once I turned off the memory feature of ChatGPT. My guess is that a lot of tokens were going towards trying to follow instructions from past conversations.