OpenAI Spring Update [video]
youtube.com
youtube.com
They clearly want to steal the thunder from Google's announcements. But they could only do that if they had inside information into what Google is launching. Otherwise it would be smarter to wait and react. Remember they already said this is not a search engine or gpt-5. So while this is probably cool it's not a silver bullet and probably won't overshadow what Google launches.
Unless it's almost exactly the same main announcement. In which case a day would make a difference.
Chatter seem to suggest they're announcing a voice to voice assistant, apple partnership and Google drive integration. All things Google is also rumoured to announce.
It definitely feels like the big guys are panicking to find their moat.
Voice to voice chatGPT. You can interrupt it any time (don’t have to wait for it to finish), it’s real time (no 2-3 second lag), and it picks up on emotion (!).
It's GPT-4o, the o stands for "omni", which I'm guessing means multimodal input and output. Rumors of an end-to-end audio-to-audio real time voice demo, that's what I'd really like to see.
I also wonder if they got a Scarlett Johansson impersonator to do the voice. It sounds eerily like the AI in Her, except for the still present problem with voices dropping out or glitching.
Nonetheless the real-time interaction is extremely impressive. So is the desktop app. Their optimization work must have been incredible to both speed things up and free up enough GPU capacity to make it freely available.
Source: I've listened to a lot of AI voice samples including many inspired by Scarlett's voice (we're adding an voice-controlled AI assistant to our software), and I can tell you that this sounded nothing like her :-)
I'm dreaming of the possibilities of an audio-to-audio model for any sort of sound. I want to be able to do things like: "What's the tune I'm humming?" "Why is my car making this noise?" "Can you separate the speech from the instruments in this clip?" "Can you make the sound of a steel drum in a hailstorm?"
Edit: I'm asking because of the obvious data security implications of having your desktop app read from the clipboard _in the live demo_... That would definitely put a damper to my fanboyish enthusiasm about that desktop app.
I need to know if my spidey senses are on point.