Thanks, one follow up point on this. Note that since you have to use the beta endpoint, you cannot also use prompt caching, correct?
This dovetails with my other question about prompt caching-- it would be great to be able to expressly create “context cache” (like the competitor), and be able to refer to it on subsequent API calls.
Personally, I haven’t 
- but this would be an interesting approach to try if you have a use case in mind!
No current plans to bring the ChatGPT voices to API
In terms of controlling the voices, we’ve done a ton to make them steerable., especially on today’s new models. Definitely try prompting them to speak the way you want (accent, speed, excitement, age, emotion, etc…)
Yes, it is definitely possible to do synthetic data generation for DPO, an important thing to keep in mind is to make sure your evaluator for generated samples is consistent across data points. Also, it is possible to fine-tune for multi-turn conversations! Just put the final assistant message in the preferred / non-preferred output.
Thank you for the catch! We’ll fix this right away.
Any update on when the Knowledge Cutoff dates will be updated?
No timeline yet for a PHP SDK, but we’re excited to ultimately support all the most popular languages! We’ll keep you updated, and in the meantime, we recommend community-supported libraries to get started.
It’s still in Beta. There are a few more goodies we’d like to add before declaring it a GA
It’s something we want, but nothing to share just yet. Our biggest focus at the moment is making sure the base audio models are very high quality
Thank for you the great session!
Do you have in plans to add more control over structured outputs (arrays to keep the min-max items as response)?
Is there any plans to have an API error response in case the API is down? Or this not really possible and it would be better for API users to handle this on the backend? For now, I think it just hangs - this was a feature request here in the community a few days ago
Congrats for the release!!
When using o1 with function calling, can it invoke functions as part of its reasoning process, using reasoning tokens to, for example, verify the correctness of its answers (potentially multiple times) within a single API call?
Developer messages are new in o1 as part of our recent work on instruction hierarchy. You can read more about them in the model spec. You don’t have to think too much about the difference between these two, it’s effectively a no op for devs but we wanted to let you know in case any of your messages reference “system messages” so you can update it to reference “developer messages” instead.
Is there more information or maybe a cookbook on the now 3 different versions of fine-tuning and what they’re best at? I wasn’t able to understand from the live stream how preference fine-tuning was different from the OG fine-tuning. I get that you get a desired and undesired example but in what ways does that affect the model different than just giving it a desired response?
Do you have any plans to bring any of the search features to the API? It’s difficult to get real time data and there isn’t a great way that I’ve found to access the depth of information showcased on yesterday’s announcement with a Google, Bing API.
Agreed. Would be so much easier and better than using Perplexity.
Thanks for calling this out - I will both make this change, and revise the process for sharing these updates.
Hi,
Assistant API not available for O1 and not updated for so long is a real disapointment, can you please be more specific as per when you may release it next year? Many of us need to take decisions on how to move on.
Will you improve the RAG?
thanks
My last one— How many days until Tier 4 get o1 API?
Should I just whip out the cc and jump to Tier 5?
We have heard good feedback regarding audio / multi-modal finetuning, but right now we don’t have plans to release audio fine-tuning soon.