Our dev platform team is planning their work for the next month, and wants to make sure we're building what people want. So if you're an OpenAI API user — what should we add/fix/improve to make your experience better?

Mar 8, 2023 · 5:10 PM UTC

353
54
7
520
Replying to @gdb
Very simple but helpful change - ability to label API keys in the API platform portal. Totally forget what key is for what app. Especially when limited to 5
1
Replying to @gdb
Some level of session ID/context tracking on the OpenAI end.
2
Replying to @gdb
It was somewhat easier to build on the non-chat apis, since gpt-3.5-turbo is very "aware" that it is a language model and is vocal about it. Needs a lot of prompting and response cleaning to create a service that is not supposed to appear as a chat bot. Back to Davinci.
2
4
Replying to @gdb
logprobs for whisper
1
Replying to @gdb
Image edits API should accept any image not only PNG with RGBA format.
1
Replying to @gdb
Whisper is too slow compared to some of the other services currently available for shorter audio clips. It's hard to implement it in real time use cases. As well as conversation id for chat API would make it so much better.
1
1
Replying to @gdb
Domain specific fine-tuning of GPT3.5. The ability to fine-tune on domain specific text(as GPT3 supports) in unsupervised way plus supervised fine-tuning for domain specific query/response dataset. (Kinda proxy for human in the loop RL).
1
Replying to @gdb
A truncate parameter for the embedding api would be super useful. tiktoken is not trivial to port to other languages.
1