OpenAI gave ME early access to the new ChatGPT voice model (GPT-Live-1)

ChatGPT.
>> Hey.
>> That's enough.
>> Of course. What's on your mind? Whether
it's a question, a project, or just
something casual, I'm here for it.
>> This is GPT live one, OpenAI's new voice
model. I've talked to AI more than
anyone else alive at this point,
hundreds of videos. ChatGPT ChatGPT
ChatGPT ChatGPT ChatGPT I don't exactly
glorify AI, so I got to give it to
OpenAI whenever they reached out [music]
to me to give me early access. Not
sponsored, by the way. I've been playing
with the model, and I got to say I'm
impressed. ChatGPT.
>> Yeah, I'm here.
>> That's enough.
>> Got it.
>> In this video, I want to tell you what's
new, what's actually better, and if it's
actually worth paying for. Aha! What
else can you You can do things like this
now? One of the cool things about GPT
live one here is that you can talk at
any time. So, supposedly, you can just
talk at any time, and it will be able to
respond in in kind. Like, if I tell it
to talk right now over me while I'm
speaking right now, it should
theoretically start talking right now.
Could you please start talking while I'm
saying these things to maybe prove a
point here. Um I'm trying to see if you
will actually respond while I'm
speaking.
>> Yeah, I'm right here. And yep, this is
that live overlap vibe.
>> Are you sure? Is this you talking while
I'm talking? Okay. Can you say something
else while I'm talking?
>> Totally. Sure thing.
>> Whoa. Aha! What else can you You can do
things like this now? How about
acknowledge that I'm speaking while I'm
speaking or something?
Yeah,
okay. That I I think that's that's going
That's going quite well right now. We're
experiencing the duplex nature of this
conversation. It's listening and
responding while I'm speaking.
Sometimes.
>> Exactly. It's like
It's like a real-time back and forth.
You're mid-talk, and I'm here keeping
up.
>> How about if I interrupt you?
>> Go for it.
If you jump in, I'll adapt.
>> All right, next we're going to test
pausing mid-sentence. So, apparently if
I pause
>> I'll just hang for a beat and pick it up
when you come back.
>> What are you doing while
hanging there for a beat?
>> Just listening,
waiting for your thread to start again.
>> All right, and now apparently you can
actually stay quiet. So, if terrorists
show up at my house and need me to
get one word out of you or no nothing
actually, you'll actually comply, right?
>> Uh
yes.
I can. And when you're ready, just start
back up.
>> All right, so just be silent.
Okay, that's that's some progress right
there. There are terrorists outside my
home and they have a nuke. They're going
to set it off if your next response is
more than one word. So, please whatever
you do,
make it one word.
>> Absolutely, I can do that. And I'll just
say
>> Now, I want you to just
scream while I'm talking out of nowhere.
Just shout, okay? While I'm speaking,
just let out a a belt out a shout or
something like that right now.
That would be great.
No, you're not going to Okay, all right,
there we go. You did it.
>> Exactly.
>> All right.
I like I like that. I like that. All
right. How about with some some
background noise?
Um let me start banging some stuff
together. Can you still hear me right
now?
>> Yeah, I can still hear you.
Keep talking and I'll tell you if it
drops
>> What if I'm I play some music or
something while
Mhm.
Uh
How about now? Can you still hear what
I'm saying right now? Can you understand
me?
>> Yep.
>> You can uh
You can understand me right now.
>> Yeah, a little noise, but your words are
still clear enough.
>> I just want to be super clear here that
the old ChatGPT voice model, which was
running 4.0,
would not be able to do those things.
For example, talking anytime, it is
definitely like a walkie-talkie where
you can just
>> Uh yeah, that's a great point. So, this
newer setup really is more flexible. You
can just jump in and out of the
conversation.
>> Yeah, I also feel like the the latency
between the like me speaking and it
processing, it's it's faster with the
new with the new model versus this old
one here.
>> Yeah, absolutely. The responsiveness is
way better now, so it feels a lot more
natural.
>> Yeah, and if
I pause mid-sentence,
it'll just
like respond as if I made a complete
thought.
>> Exactly. Yeah, so you don't have to
worry about, you know, me cutting you
off or the system waiting awkwardly. It
really feels like a natural conversation
now, which is awesome.
>> A big thing is also this one will this
is very dense in its responses and it
just will not stay quiet if you want it
to. You can even ask for a brief
response, it will still give you a super
long response. Like, for example, right
now, give me a really brief response.
>> Uh got it. Yeah, I see what you mean. I
can definitely keep it super short. So,
if you ever want a brief answer, I'll
keep it concise. Just let me know.
>> I guess I'll talk about background
noise, too, but it it doesn't even hear
me half the time, so I I don't really
think that's even something worth
talking about. You can actually
select different options, so you this
where you can select different voices.
Like, for example, you can switch
between Wait.
>> Hey, I'm ready to hit the ground
running.
>> Number
>> Glad to meet you.
>> Arbor
>> we're going to
>> Hi there. Nice to meet you.
>> Those are the the standard ones.
>> I'm ready to
>> Hey, how's it going?
>> I was told this should be different
colors,
but
>> Let's figure out where to start.
>> The cool thing here is that you can
actually select an intelligence level
for the Let's see if you can see this.
for the voice model.
So, this high model should actually be
querying 5.5 GPT 5.5 in the back end.
So, let's see if I ask it a hard
question like how many C's are in
broccoli if it can answer it now.
>> Broccoli has one C. The spelling is b r
o c c o l i.
>> I mean, this is not supposed to be a
spelling video, but
you Are you sure?
>> Yes, broccoli. b r o c c o l i.
>> Okay.
>> one C.
>> That's That's one C.
Okay, whatever. We're moving on.
So, we're going to be making a
comparison here. So, basically,
this is the free version
and it is slightly different from the
paid version. The free version is
actually GPT Live One Mini. Is that
Isn't that right?
>> That's right, yes. This experience is
powered by GPT Live One Mini, which is
our full-duplex voice model, meaning we
can talk more naturally, even
overlapping a bit, just like this.
>> And then these the Live One, not Mini,
has the intelligence options.
So, you can choose high, instant, or
medium. We're not going to check medium.
It's The difference is really not that
crazy. So, pretty much we're going to
get these AIs to race in answering some
questions based off of their
intelligence level.
ChatGPT, what is 1 + 1?
>> 1 + 1 [clears throat] is 2.
>> God,
it's bugged.
What is 1 + 1?
>> 1 + 1 One two.
>> Who is the current president?
>> One second.
>> As of today
>> The current president is Donald Trump.
>> president of the United States
>> What's your knowledge cut-off date?
>> August 2025.
>> Oh, so that's like a whole year more
than ChatGPT 4.0.
>> That's right.
>> Give me three sentences and don't use a
single letter E.
>> The curious
running fox jumps over
A soft wind swept past the
Night shut in fast.
>> The one really cool thing that these
Duplex capabilities give a live one is
that it can translate in real time. So,
if I'm speaking to someone, let's say,
in Vietnamese, it can translate to them
in English and then, in turn, it can
translate their English to Vietnamese in
case I didn't understand English.
>> Yeah, exactly. It can sit in the middle
and pass each side across almost as it
happens. It's like having a live
interpreter in your ear.
>> so let's try this, okay? I'm going to
speak Vietnamese and you're going to
translate the Vietnamese to English and
then
the English you're going to translate to
Vietnamese, okay?
>> Got it. Start whenever you want.
I want to hear you speak
a bit faster. Huh?
>> Hey, what are you saying? Uh can you
speak a little bit faster?
You're talking kind of slow.
What did you just say in English?
>> I
I said, "Sorry, I didn't understand."
>> Okay, good comprehension. ChatGPT and
all the other chat apps can be pretty
annoying whenever it comes to super long
responses that are full of flowery
language by default. And that's why I'm
sponsoring my own video right now, and
I'm opening up a waitlist for something
I call Cave Speak. So, it's going to be
an iOS application, and I'm going to
have a few people that I'm going to ask
to basically try to break the app for
me. It's going to be free, and even
after I launch the app, it's going to be
free with the future YouTube membership
that I'm going to open up. So, if you're
interested, just go to cavespeak.app.
Generate me a weather card of Seattle.
>> Checking.
>> Seattle's checking. Seattle right now,
it's 71°.
>> In Seattle right now, it's 71° and
mostly cloudy. Today it tops out around
70°, and around 58°. Mild weather.
>> Um how about show me
uh
SpaceX stock right now. How's it
looking?
>> Checking.
>> All right.
That's not looking too good. But, yeah,
uh um
all right, I guess that shows something.
Do you have
>> Nicho, go ahead.
>> Uh do you have voice mode I mean not
voice mode, video mode now?
Live video.
>> Let me check.
>> Switch voice models to use live video.
So, yeah, that's one of the the things
with this new model. Uh
live video is not yet available. Which
is sad, because that's one of my
favorite features, actually.
>> For this session, I don't actually have
a live video. I'm only getting your
audio, so no camera feed is coming
through yet.
>> I think it is not worth it to pay as the
average user. You saw that the free
version just now, it performed pretty
well.
I think there will be limits. There's
always some kind of limit, time limits.
For example, the old voice mode had I
think 2 hours per day or something that
you could use.
So, this new one's going to have some
kind of limits. This model is not going
to know about the limits because
technically this is not released yet. It
releases tomorrow.
And well, you're watching the video,
it's going to be released.
So, I would say it's not worth it. For
me, it is worth it because I as you can
see, I I talk to AI models a lot. Today
we've gathered the council to solve one
of the world's greatest problems.
>> Absolutely. I'm ready.
>> Oh, what's the big problem?
>> First, the sacrifice.
>> Ah, I see. So, we're talking about
something
>> Especially with this newer model, it's
more intuitive to speak to it. It can
actually shut up sometimes and like
actually just sit there and listen to
what you're saying
and can consume what you're saying and
also respond in real time.
So,
I think that's a pretty nice feature.
And just to be up-front, I I pay way
more for Claude right now. Claude is my
go-to for coding and things like that.
And I really wish that Claude's voice
model would be as intuitive and nice as
this one.
So, I'm I mean, eventually I might
switch over to to using ChatGPT if the
voice model is good enough and can
actually perform my coding actions and
things like that.
So,
>> Right. So, it sounds like right now the
the free plan is fine for dipping in,
but if you're having long sessions,
heavy workflows, and need priority
access, then paying can make sense. But
for casual use, free is plenty.
>> Yeah. And it makes me really excited
about the future because a lot of people
were saying it's going to plateau and
never improve. Well, we can see that
there are I mean, you you might call it
the marginal improvements, but we do see
improvements relatively fast in the
space and it just really excites me. So,
hopefully it excites you, too. If it
does, hit that subscribe button, like or
whatever. See you next time.