upgrade to gpt-3.5-turbo

[chartpath]

Fixes #19

This worked on one of my projects.

[gregnr]

Thanks @chartpath. For this to work we would also need to update the client right? Switching to something like useChat().

[chartpath]

@gregnr Yes probably if you want full bidirectional chat like the API supports under the hood. My project is sticking with uni-directional Q&A for now, so it still works for that but allows us to use the gpt-3.5-turbo model.

EDIT: I'd also like to add that not all users want a full chat and there are many other frameworks out there which provide this. Would love if this project stayed as a minimal seed for starting new ones without imposing too many user-facing features beyond the excellent POC for vector search + context injection that it is. 🙌

[gregnr]

Thanks for explaining - I have no problem keeping the single Q&A format. I assumed the JSON response format from the Chat API was different and incompatible with useCompletion(). I could be wrong - have you confirmed the frontend continues to work as-is with chat API + useCompletion()?

[chartpath]

@gregnr Good point, but yes I can confirm that it still works, give it a shot!

Maybe that OpenAIStream() is just more flexible than expected.

Can't tell if this comment means it falls back to a "chat of 1 message" when id doesn't get passed in: https://github.com/vercel-labs/ai/blob/e86c2ecc1a810626167bfe5a827f7da1d14889c8/packages/core/shared/types.ts#L142-L144

[gregnr]

Thanks @chartpath!

[musashi-garami]

Thanks @chartpath ! Appreciated