Still not trained on current OpenAI API specifications

After reading some others similar experiences it seems that this might be a broad code generation related issue which is specific to GPT-4o. I’ve tried asking for the same code discussed below but with GPT-4 and GPT-3.5 and both of them produce code that is either completely in line with the latest OpenAI API spec or extremely close to it (GPT-3.5 only got a few property names incorrect).

I’m going to agree with what seems to be the consensus that GPT-4o is simply trash at code in general.

Originally posted:
Why aren’t any of the models ever updated to include knowledge of the current OpenAI API specs? When I ask it for anything in code for OpenAI API’s it’s still giving me outdated junk for davinci …

Is this an intentional move to control the progression of AI by impeding average people from simply asking for some AI code and getting exactly that in an up to date and working form?

This is starting to seem intentional because you keep them up to date on everything else, they also are more than capable of finding this info using it’s web browsing capabilities but nope it still thinks it’s best to provide tis obsolete junk for the v1 completions API and plays completely oblivious to the existence of assistants at all let alone v2…

Same thing for other features like vector stores, when I ask it for related code for that it prefers pinecone or another third party depedency rather than just using the new vector stores introduced to OpenAI API