How come openai models get their own prompt wrong. for the last year + they have been referencing the old calls. Also when working with code the models tend to replace correct calls with the old broken.
Anyone else gets this? I should not have to put in instructions. the ai should at least know its own api inside out 