Justin Uberti

Head of Realtime AI @OpenAI. Created WebRTC. Past: CTO @ultravox_dot_ai, Distinguished Engineer @google (Stadia, Meet/Duo), AIM. Amateur mathematician/musician.

[Justin Uberti @juberti]

New post on the OpenAI eng blog from two engineers on our Realtime AI team here in Seattle, outlining how we designed our v2 realtime infra and how we've optimized it for easy scalability and low latency. Check it out!
https://t.co/MuGbCumHAY

[Farza 🇵🇰🇺🇸 @FarzaTV]

Watch me control my computer with just my voice. This is the future of operating systems.

No hands.

GPT-Realtime 2.0 is very, very underrated.

Demo:

[Justin Uberti @juberti]

If you’re using gpt-realtime-whisper as the ASR model in a cascade pipeline, you can try this as an end-of-utterance signal.

[Justin Uberti @juberti]

@xanderberkein there's nothing exactly like that right now, although punctuation + timeout on deltas would probably work quite well

[Justin Uberti @juberti]

@xanderberkein there's nothing exactly like that right now, although punctuation + timeout on deltas would probably work quite well

[Justin Uberti @juberti]

Guess who's back, back again. Whisper, but now with realtime streaming. Check out the new gpt-realtime-whisper transcription model in my https://t.co/b2UTuSxhOI demo.

[Justin Uberti @juberti]

@xanderberkein The internal VAD should be much more robust than the previous server VAD. Or am I misunderstanding your question?

[Richard Sutton @RichardSSutton]

The bitter lesson in 26 words:

Don’t be distracted by human knowledge, as AI has been historically.
Instead focus on methods for creating knowledge that scale with computation, like search and learning.

[Justin Uberti @juberti]

@xanderberkein This model has internal VAD, no need to set turn_detection.

[Justin Uberti @juberti]

@rown congrats on the research preview! Looks like a promising direction 😁

[Justin Uberti @juberti]

@syllogistic LOL, the future is hard to predict. Nice demo here, would love to learn more about how it works.

[Newcomer @NewcomerMedia]

.@sama posted this week that he's "pretty excited for voice models to get great."

But when will that happen?

What @OpenAI's @juberti told @jameswilsterman at Cerebral Valley Voice:

[Justin Uberti @juberti]

@yi_ding there was an issue if you specified `null` for `turn_detection` which led to a bad state. I removed `turn_detection` altogether and the problem went away.

[Justin Uberti @juberti]

@yi_ding this is a good question, we are debugging on the server side now

[Justin Uberti @juberti]

cool realtime UX demo

[Levin Stanley @levinstanley]

Using @OpenAI gpt-realtime-2 to get a glimpse of future voice-first experiences.

A market dashboard you don’t click through.

You direct it.

Say, “Focus on Apple,” and the whole interface changes.
Ask, “How did it do over the last 30 days?” and the chart updates.
Say, “Go back,” and the market view returns.

No menus.
No filters.
No hunting around.

Just intent.

What makes this model especially interesting is the interaction loop: you can interrupt it, add more context, change direction, and it keeps reasoning in real time while updating the experience around you.

The interface doesn’t ask you to navigate.

It just takes you there.

[Justin Uberti @juberti]

@levinstanley @FonsoWealthy @OpenAI try speed=1.2

[Justin Uberti @juberti]

We're looking for a creative iOS engineer to join our realtime AI team here at OpenAI Seattle to help build the future of human-AI interaction. If you know WebRTC, AVFoundation, and/or Core Audio and like open-ended challenges, apply at https://t.co/Yjvc5kuUPR or just DM!

[Justin Uberti @juberti]

Incredible. No notes.

[Daniel Green @dgrreen]

The Sam Altman and @miramurati texts from the day he got fired from @OpenAI in 2023 just became evidence in the @elonmusk v. @sama trial.

It felt like a meaningful moment in AI history, so I turned it into a musical.

The lyrics are the texts.

[Justin Uberti @juberti]

@tsahil and @ArinSime discuss our recent blog post on delivering low-latency voice AI at scale. Quote: "I don't recall up until this post [...] any company or vendor that took the time [...] to say technically why they made the decisions that they made."
https://t.co/SceG8bjTUu

[Justin Uberti @juberti]

@yi_ding there's no playout audio so there shouldn't be any echo cancellation. Your fix is forcing input audio to be committed, but that should already be happening, so I'm confused about what's going on here.

[Justin Uberti @juberti]

@yi_ding I've seen this too but haven't had a chance to debug it. Does increasing gain help? Or do you have to manually send input buffer messages?

[Justin Uberti @juberti]

@atiorh It’s a new architecture

[Justin Uberti @juberti]

@urban_coder @ScaleAILabs @OpenAI Tau-voice doesn’t have gpt-realtime-2 yet