Great cadence of features lately! Please add logprobs to chat endpoints! Getting per-token uncertainty helps build prompts, calibrate confidence, and tune logit biases. Thank you!

1) Just ensuring API performance is consistent, will be great. In several instances there is a big lag in API output 2) Plus the downtime seems to be much higher than what is reported - pls sort this 3) If you have one really fast customer support channel that'll be great

Fewer outages perhaps

automatic token management

Ability to willingly trade cpu/response time in exchange for being able to submit more characters. I would also pay extra for the ability to upload documents. Speed is great, but if it’s a trade off between substance, accuracy, and volume, I’ll willingly pay more and wait longer

Word-level timestamps in output for Whisper API. Has already been merged to main in Whisper repo.

I want the next generate API could Offer some privacy reserved fine tune method on private data.

allow the model to think (print to hidden area it uses to generate actual reply) :D I know this might be more suited for a year of work, not month, but still

Ability to fine-tune gpt-3.5-turbo and text-davinci-003 on our data

Fine-tuning on gpt3.5-turbo 🚀 🚀 🚀