Everyone in my network is tingling today… May 13, 2024. We are expecting a large announcement from OpenAI, the creators of ChatGPT. If you have been following, OpenAI tends to keep building onto the ecosystem and make model enhancements (base, turbo, and a .5 release). Over the past year or so, they added more API endpoints for developers to make apps on top of their foundational Generative AI models, added memory to ChatGPT, relocated the whole web app to chatgpt.com (silently), and added custom GPTs as a store. I expect more of the same… more of the ecosystem being built out so they can capitalize fully on all of the traffic to ChatGPT.
Here’s what we know so far…
- The scheduled May 13, 2024 Event is called “ChatGPT and GPT-4 Updates.”
- OpenAI social media manager on X says, “We’ll be streaming live on http://openai.com at 10AM PT Monday, May 13 to demo some ChatGPT and GPT-4 updates.”
- The OpenAI website has a “Spring Updates” call-to-action that downloads a calendar invite for the live event.
I will keep this post updated as I learn more.
Update: OpenAI is streaming now at 1 pm EST.
Update: OpenAI Spring Update Live Blog
- New web UI and desktop app – complete refresh of the web app and dedicated desktop app
- GPT-4o – New flagship model, GPT-4 level intelligence, but it is much faster
- GPT-4o free access – bringing the new model to free users without a sign-up flow
Other Notes:
- “o” means omnimodel.
- Omnimodel means bringing multiple modes together.
- GPT Store is expanding to free users. Around 1M builders of GPTs.
- GPT-4o is in the API now, 2x faster, 50% cheaper, and 5x higher rate limits compared to GPT-4 Turbo
Demos:
GPT-4o is real-time responsive. The model picks up on emotion and can express emotion. Conversational apps are now possible with GPT-4o.
The coding demo had a number of examples:
- Summarize code
- What would the plot look like if you have this function or not
- Overview of a plot – what does this plot mean?
- What months do you see the hottest temperatures?
- What are the axis ranges?
Demos Suggested by X Users
- Real-time translation
- Tell feelings by looking at a face
New Model Test From MATLAB
Sending a request to the new GPT-4o from MATLAB is about 2x faster than sending a request to GPT-4 Turbo. It takes .5 seconds to say this is a test on GPT-4o versus 1 second on GPT-4 Turbo.
