What's new in OpenAI Assistants API v2? Changes to Cost, Quality, and Speed

In this video, I cover everything new in the OpenAI Assistants API V2 and walk you through what it means for you. You'll get a peek under the hood to see the differences between how things used to work, how they work now, and how you can leverage the new controls to improve the performance of your assistant chatbots while paying less.
00:00 Intro
01:25 Previous Architecture
02:51 New Vector Store Object
04:03 Previous Context Retrieval Flow (RAG)
05:23 New Context Retrieval Flow (RAG)
08:30 New Fine-tuning capabilities
10:10 New History Management
12:18 New Streaming Responses
13:30 Next Steps
Looking for guidance to implement or scale AI for your business?
Book a call with me: cal.com/borna/discovery
Connect on LinkedIn: / balmasi

[@Paul_Marek]

Great overview. Thx! Happy to find you.

[@Fedgery007]

Thanks for sharing! You’re really good at explaining things for newbies!

[@suitedadvice]

thank you for the interesting video!!!

[@DouglasCamargoLeal]

Excellent explanation, thank you! It would be great to have a video explaining how to integrate assistant v2 into botpress so that it can be used as a card in any part of the flow, in the same way it works in typebot.

[@show-me-the-data]

I have a video on that (as well as a template)! --> th-cam.com/video/6zS0LToLOiA/w-d-xo.html
You just have to change the header from V1 to V2 in the API call card.

[@KevinRamirez-vw5kv]

Great explanation and exactly what I was looking for.

[@MeinDeutschkurs]

Great! I think I should try this method with flowise and llama3.

[@show-me-the-data]

Great idea, let me know if flowise supports something like reranking, cause I haven't had a chance to try!

[@MeinDeutschkurs]

@@show-me-the-dataSorry for the late reply. I have no Idea. But I tried to use several models to value/rank content. The greatest troubles are to find a ranking system that is functional on every topic and/or type of content and to have a format, which is then constantly being used. Especially when using ollama onboard models.

[@user-mv3yw4zs4q]

I would like to know the cost of such requests. In the playground when I ask a question and the answer is extracted from the document it takes about 10,000 tokens.

[@marpertube]

Please update the Botpress template with the OpenAI API V2. Thanks

[@show-me-the-data]

Done! I sent an email to everyone who downloaded it and updated the download flow from the other youtube video: th-cam.com/video/6zS0LToLOiA/w-d-xo.html

[@marpertube]

@@show-me-the-data Thanks!

[@RafiqLightwala]

How to hook it up with botpress?

[@show-me-the-data]

I explain that in this video here: th-cam.com/video/6zS0LToLOiA/w-d-xo.html

[@RafiqLightwala]

@@show-me-the-data Since it was three months back, I thought it covered only v1. Does it work for v2 api as well?