Yep it completely ignores my Custom Instructions now, it was very obvious when they switched to Turbo.
What I don’t understand is that GPT-4 Turbo does much better than anything else in lmsys arena, which is a blind head-to-head test, so it should be very fair:
I guess the prompts people are entering in the arena aren’t representative of the kind of prompts we’re using in the actual chat interface?