Has anyone worked on integrating a GPT chatbot in a videoconference room (like Zoom or Chime?)
I would like the GPT bot to participate in every chat like any other videoconference attendee, so it can interact in real time. In the simplest case, it would listen and take meeting notes and write (and later, possibly translate in near-real time).
Thanks.
Hey there and welcome to the community!
These are great ideas!
They are also insanely difficult right now.
This has not been achieved yet, or at least, it has not been developed in the public eye quite literally until a few days ago when there was a tech showcase in the World Summit. Even then, it was demonstrating capabilities from better chips, so don’t exect this to happen on normal hardware easily, and not in zoom.
You also take for granted multi-turn speaking. ChatGPT (and pretty much all language models) are trained on a single, perfect, 1-1 turn-taking ratio. Multiple speaker participants significantly complicate this task, even for real time. When should it answer? When should it be quiet? How would it know when it should speak?
I love the enthusiasm, but these problems are giving all kinds of corporations a run for their money. It’s not easy, and it would take a significant amount of work and resources for this, with no guarantee of either success or by getting blown out from your own competition.
That said, this:
Is achievable.
This:
is not. Yet.
Unless you’re comfortable looking at meta’s research paper on the AI they developed for this, and have the technical aptitude to implement it into a product. I don’t even know if they released that model publicly yet tbh.
@test.mgr we’re seeing people actually build these real-time meeting chat bots with the following stack:
- Use Recall.ai to send a bot into a meeting (like Zoom) and get back real-time transcripts
- Take the real-time transcript and pipe it into OpenAI to translate
Hi Macha,
quick question: do you have any idea or resources on how to integrate a chatgpt into a group CHAT setting, though? I understand that it is impossible to use the real time voice-to-voice (yet) but maybe it is possible in a written context?
Maybe you’ve experienced something like this before. Sounds like you know what you’re talking about, haha