The best feature is that it can delegate questions out to GPT-5.5 in the background, so you're no longer restricted to a voice model that's several years behind the frontier.

I did report a fun bug with it though: it was interrupting me and laughing at my (not really intended as) jokes while I was still talking! They seem to have clamped that behavior down thankfully, it felt a bit rude and condescending.

I'm not Catholic, but this podcast presents a very interesting argument against talking to AI as if they were human: https://newpolity.com/podcasts-hub/debate-chatbots

It’s weird. The old Claude voice mode WAS able to use tools but when they revamped it, it lost that capability and is now pinned to Haiku :(

So, yay for finally a voice mode that’s powered by a frontier model and hopefully as good as Grok voice, but sad to still not see tool use while in voice mode.

(I haven’t tried it yet, only read the announcement)

The part that creeps me out is, we're living in an era where we're more disconnected from each other than ever before. Do we really need to be replacing conversations?! The demonstration video of old ladies sort of hints at something for me, which I think we already have a societal problem with the way we treat the elderly (and a massive elderly-loneliness issue) and there's kind of a sadness of imagining people becoming really close with this machine that doesn't really think. Definite ick factor.

People have tried "smart <thing> that helps blind people navigate" since the 80s, many, many, many times, and all such projects failed. The cycle of "wow, blind people could benefit from a navigation aid, why don't I make one, if there's none around, I must surely have been the first bright university student to think of this idea" is pretty well known in the community, and I'm personally quite tired of it. Nevertheless, I think this may be the one.

Circa 2020, I have said that people who are getting a guide dog now are probably getting their last one. I think we aren't far off from that prediction coming true.

GPT-Live-1 is the first version of a new generation of models, and we believe the full-duplex architecture + delegation enables entirely new ways of human-AI interaction.

Would love to hear your feedback!

For example I asked

“Why should LLM attention use dot product instead of cosine similarity, being that we often hear vector magnitude does not encode most of the useful information needed”?

The voice response was directionally right but lacked detail and was a little hand wavy.

The answer to the same question in a text chat was much higher quality.

The voice response replied “let me think about that…” so it appears to be invoking 5.5 as advertised, but it’s definitely weaker.

I had reasoning set the same for both.

I'm a little surprised by how much OAI is playing catch up here.

My dream would be open source full duplex with function calling or some kind of rudimentary text output. PersonaPlex is still interesting although it was looking like we would need to fine tune it to handle outgoing or avoid going off the rails easily.

Oh and there was also a small fail in the live translate demo: the grandma says "tell him that..." which the bot translates verbatim, whereas a real translator would of course understand that this is an aside not to be translated.

Well I guess at least I should be happy that they're transparent in their ads :)

"I'm going to stop you right there. Let's keep the conversation focused on the topic we were covering or a new relevant topic".

I tried to probe it for why it did that, what rules it was following, and it eventually told me...

"My role is to keep us focused..." and, "The behaviour you saw was my attempt to moderate tone".

I've heard of LLMs doing weird things like this, but it was the first time it happened to me. I hope they fix that. It was creepy.

For context, it heard my partner say, "I guess it's the same thing as you mom, because she's..." and then it cut us off.

I am wondering if this because Siri is so bad, people think of Siri now as a voice activation method rather than an AI assistant as it was intended.

The demo is so good, what was once a sci-fi / Iron Man Jarvis services is now real. I don't follow AI closely, but all previous iteration were at best ask and answer type of services. It wasn't real conversation. And whatever flaws it may have now, at the rate of improvement within a few iteration it will surely reach good enough stage for majority of people.

This is also scary. Not just for adults, but for kids. How they could become even more isolated.

I remember the PC era, the internet from Information Super Highway to Web 2.0 Then Smartphone. It may have been obvious to many but AI really is something much bigger than all the previous three, perhaps combined. And it is also the only one that I think is scary.

The full duplex is awesome, and the feedback that it is getting what you're saying is ok, but in some of the demos was a little overkill.

I'll agree that using the "Golden Girls" was at least more entertaining than the usual pitch.

Doesn't quite stop fast enough when you interrupt it. Can't find info quick enough so you have to change topic and then have it give you results later, etc.

This is a move in the right direction, but there is lots of engineering still to be done!

The concern is though as these get better will people struggle to distinguish these with real human connections?

I really hope at some point we can 'option click' or whatever, and choose multiple threads to archive. It takes FOREVER to archive the cluttery chats one by one. That's my one wish feature, please make batch cleanup of my client reasonably easy. Imagine having to delete one file at a time in a folder of 50 plus files. Barf.

Disgraceful.

On the technological side, it's a marvel!

IIUC the literature, there is serious loss of functionality associated with lack of verbal interaction. People can say "they should just talk to more people" or "more people should make time for them" but the fact of the matter is that it doesn't happen, and if this helps terrific.

It was awful, it kept overhearing us talking and thought we were talking to it, interjecting with nonsense because it couldn’t really understand what we were talking about. And when I would say “ok render a score card” it couldn’t drop to the text interface or anything, it had to stay as voice, so it fumbled around trying to read the scores back to us.

It’s a very stupid use case but I viewed it as a stress test to see how well the voice mode and multi-modality could work. It failed miserably, but I’ll be interested to see if this new version does any better (not that I actually need this use case, writing on a card with a pencil is just fine.)

Awesome that they've improved that aspect of voice chat, though.

Whether that's due to my slight autism or massive nerdery, I don't want more realistic voice. I already switched to non-advanced voice in gpt, and I cannot imagine wanting the mmmhms, the yesses, the laughs, in my ai interaction. I want to ask a structured question and get a structured meaningful response. Informative and structured are really the KPIs. The umms and ahms of existing gpt advanced voice are annoying enough, the recent increased usage of first person almost a deal breaker (when asking for bike technique on lose surfaces yesterday, it literally gave me "back when I was learning bikes riding" story - eww).

Fascinating to see the architectural advances though, even when they deliver something I personally don't need :)

One thing I noticed is that we lost vision feature for some reason on the live chat?

This was an extremely useful feature. Not sure if it’s a regional thing or that they just removed that from the current live chat.

I imagine it will be even more useful with this new version.

Not super impressed by the model constantly interrupting the user in the other demos though.

I used it to help me figure out how to turn off a feature in the rental car I was in (adaptive cruise control, I love it but snow blocked the sensor and I wanted just normal cruise control but couldn't figure it out while driving).

This kind of voice chat is awesome and I'll be even more excited when open models have this functionality. I'd love something like this paired with Home Assistant (assuming we ever get decent hardware).

Almost felt a bit *uncanny valley* for what "natural" conversation is supposed to be like. If the "uh huh" isn't timed correctly, it'll feel like a zoom call with lag.

The model is not interested in the conversation, it is just "serving" a conversation.

Which is very different from the engagement of SOTA text models.

But I just want to say that talking with AI casually is critically lame. I cringe every time I have to ask my Google Home to turn on the lights and people are having full-on conversations with it? And, imo, dangerous considering how sycophantic AI is. The stupidest, most gullible, most insecure person right now is looking at this thinking they are about to make a new friend.

https://imgur.com/a/ABGWRTO

Do engineers that work on this for long periods of time stop seeing the forest for the trees and think this could be mistaken as human? I'm saying this as someone that assumes all narration/most VO work will be fully AI fairly soon (for better or worse.)...

Except that you can't actually delegate since connectors and tools are not supported.

Seems like a shift from previous voice models where it sequentially processes voice to text then feeds it to LLM and then back which cant escape the clunky lag

not sure how pipecat stands now, gpt live seems like it takes audio tokens and does inference on it directly

This is an abuse of user trust and violates people's privacy.

Seems a bit disappointing, but the 3 overlapping questions example was impressive.

https://si.inc/posts/hertz-dev/

Meanwhile, Siri struggles to send basic texts to my kids.

Nooooooo!

Basically have an older lady (not their target audience) blatantly reading a teleprompter.

Why are they going after this audience? Retired people have no use for delegated tasks or information. They also are the least likely to use it and not get frustrated.

This time is the most natural version that exists and it is a natural as a conversation.

To Downvoters: Why aren't you feeling the AGI?

sounds cynical in my ears. energy demand of these toys will cause many problems, people elsewhere starving being one of them.