Sorry - I was ambiguous in my Q. By ‘end user’ I meant me as a programmer using Observable’s code assist. I realise that once built, anyone viewing a page I have created doesn’t use my OpenAI access.
Currently I use the ChatGPT 4 interface (i.e. not via the API), and I find it remarkably helpful in speeding up the tedious coding work. Unfortunately, although I pay for this with a flat-rate subscription it doesn’t cover API use, which is charged on a use-basis.
I’d be interested to find out what the Observable specific trained model gives that the normal ChatGPT interface doesn’t. But don’t want this to be a costly exploration.