Also what effect will these extra instructions have on output?

Every token of context can drastically change the output. That's a big issue right now with Claude and their long conversation reminders.

You mean, for instance, if you ask it to insert an advert into content, can it do so, based on its training set, without changing the wider content into advertorial?

It's a really good point. And it has some horrifying potential outcomes for advertisers.

Exactly.

Am I going to come back from coding with a function named BurgerKing or something? Lol