It’s sad that ChatGPT doesn’t even know about how to use its own API… And it looks like the training data only missed the API change by a month or so!
Seems like a misstep / perhaps miscommunication by the different teams at work. They really should have squeaked in their current API info into the training data before such a big release…
In conclusion:
OpenAI: If possible, please update your training to include the current OpenAI API documentation.
Thank you! 
I don’t mean to dismiss your request, but I also want to sanity check a workaround: could we prototype a rough Assistant version of this in ~1 hour? Has anyone out there tried and would you be willing to share how it went?
Proposed Prototype
Of course that’s not the same as actual training, but I wonder if it would help. It’s also interesting to me for other APIs and SDKs which are under active development. I’ll likely try this for some other platforms that are more stable and I’m already more familiar with.
It’s just… ironic, but not in a funny way (i.e., that it doesn’t know how to work with, itself). And annoying. Yes it can be given instructions, but that is of limited effect and also uses up memory/context resources that could be spent on other things. It still makes sense from a product standpoint to harmonize the model(s) awareness of their own API.
Agreed that it could be a nice and useful addition.
