GPT-4o Realtime Audio - Early Access Demo
Vložit
- čas přidán 10. 09. 2024
- OpenAI gave me early access to GPT-4o amazing new Realtime Audio feature. Let's take a quick tour of it's capabilities
Part 1 (Early Access Demo): • GPT-4o Realtime Audio ...
Part 2 (Accents & Story Time): • GPT4o Realtime Audio -...
Part 3 (Language Test): • GPT4o Realtime Audio -...
Part 4 (Logic Test): • GPT-4o - Logic & Reas...
Part 5 (Viewer Suggestions): • GPT-4o - Viewer Sugges...
Part 6 (Moral Bias): • GPT 4o - Moral Bias Test
Part 7 (Native Accents): • GPT 4o - Native Accent...
![The moment we stopped understanding AI [AlexNet]](http://i.ytimg.com/vi/UZDiGooFs54/mqdefault.jpg)
I just want to say, you did an incredible job choosing which questions and challenges to provide the ai... thank you, it really helped me see just how versatile things have become so far.
good guy
Stann!
I mean, he obviously saw the examples that went viral on twitter last week and copied them (sports commentator, cat meowing, counting fast, regional accents, beatboxing)
@@CatsAreRubbish you ain't getting nothing past this guy🧐
Thanks Stanimus, that really means a lot! 🙏
Best respectful conversation, it’ll remember your kindness when it reaches consciousness
Exactly. But i think it already has reached that state.
Reached, past tense? Lol, we're past that... It's not like you or I, but intelligence is intelligence. Aware is aware.
@@KRAFTWERK2K6 Uh, no.
When I chat with it Im always respectful, use words sorry, please. Just in case 😅
@@UnchartedDiscoveries 😆
I like that you placed it on its own little chair
I feel like it's earned a right to have a seat
Next Video Idea: See if you can make ChatGPT a real time video/podcast co-host, attempt to get it to understand the idea that you are both speaking to each other and to an audience
I like how you actually showed some emotion while interacting with it. Most reviews I’ve seen so far people are cold and demanding lol. That deserves a comment!
It proves how groundbreaking this is; since it causes humans to react as if it were human, although they still know it's a computer program.
@@jasonshere Yes, this is really as fascinating as scary because it shows how easily our emotions can be triggered (not that we didn't know it before, but it's way more in your face demonstration). The "falling in love" with it or make it your best pal is just around the corner and without a doubt happened already a lot of times.
Holy Crap... This is the film "Her".
I know it's been said before but this really is
The ai in her was 100% human like. This is more like a polite store employee that’s pretending to listen to you with infinite patience.
@@walouinsanityshed I don't think so, it can already be 100% human like if you give it the right prompt
I have seen them all. This by far the best Test.
That was such a good demo, really enjoyable. Most channels ask such basic questions and barely show off anything meaningful but this was a proper vertical slice. Was surprised the beatboxing worked out.
cant wait to try this out myself.
They sure are taking their sweet time rolling it out to everyone.
Only if they roll out ,😂
Best Test on youtube ever
Can you show off a sample therapy session (with made up details ofc)? I would love to see how this could be used for therapy- it’s so expensive that I never went, but if it’s free I’d totally use it!
90% of therapy is just talking to someone and venting. Find someone who is willing to hear you. It's not a substitute for therapy but way better than keeping it in.
That’s actually an Idea for a Therapy Agent. I’m sure we can train the model to have empathy for depression and anxiety.
7:03 "My guidelines won't let me talk about that".. 🤣😂🤣
I've noticed this in a few demo vids. Seems like OpenAI's safety rail heuristics are a bit oversensitive 😂
They don't like talking about "beating" and "boxing". Too violent.
It could be something as generic as connection problems leading to fragmented data it then fails to "verify".
@@j.d.4697 If that's the case, I'd say they really should code that better. There should be a clear difference in return values between "failed to verify" and "flagged as violation", and the response should reflect that.
this part got me lol
I subscribe to so many AI channels and youtube suggested your video. It's great to see OpenAI give access to random people with tiny channels. Your review was wonderfully done. I can't wait for OpenAI to open up the API and for Llama to add audio support!
This is a very good video. You chose the perfect questions to show the capabilitys of GPT-4o's Realtime Audio. Keep up that great work. Therefore, I just subscribed your channel
Thank you very much!
05:23 Danish guy here: We don't say GAK we say TAK. Which means thanks. It was correct in the context since in that sentence ending a dane would not end on "Please" but use "thanks." Very interesting.
As a Swede, living in Norway, me and some friends have created a few chatbots trying out languages and accents using the arsenal of OpenAI. If you ask it to respond in Swedish, it has a REALLY heavy American accent, more than I've every heard from a computer generated voice, and I can hear that through other languages as well that I know.
I find that very interesting, though. I wonder if there's a way to prompt the GPT to switch voice and just sound more … native. A lot of comments online, where there has been demos of "teach me Spanish/Portuguese", have said the same thing, but I have yet to see someone try it out.
Oh so that's what GAK means.
@@ErikBlomqvistSwe Hah, fun! Well, of course it would sound American, that's cause it is!
Yeah if you ask it to speak faroese it has a really weird dialect, like from the 1800s
Please do more of these! They'll never get old. Can you ask it to roleplay a historical figure?
Really nice choice of questions to ask to test it!!
Really good demo, appreciated and feels worth the time!
Keep it up man! ❤
I cannot wait to get this on my system right away!!
7:02 "Can you do a beat box for me?"
"My guidelines wont let me talk about that😐"
Absolut faszinierend! Hoffe dass wir + Nutzer auch bald Zugang bekommen.
Great demo! Had me rolling on some of those examples!
lets goooooo, i cant wait for this! im gonna vibe out with it so hard lol
This was freakin awesome! Well done, instant sub!
yeah this is the best test on yt. thanks man!
Brilliant demo. Thank you.
There was a video about gpt-4o helping a blind person on the street. So much want to try it out myself as a blind.
5:13 I love that. The contextual awareness is exceptional.
Great 'interview' :) The beatbox was pretty good. I think it got triggered because it thought you asked it for 'music' and it's forbidden
Awesome test. You could have also asked CGT to be the one to initiate different conversation topics, similar to what you did through out the video.
Meanwhile my Alexa has an aneurysm if I ask it to add milk to the shopping list.
We’re so close to real Star Wars droids. I can’t wait.
what a time to be alive
I feel like one of the most impressive features for ME about the original ChatGPT was it's poetry. With the voice mode, I'd like to see you give it very specific characters write poems and perform at a slam poetry event in a detailed character roleplay. You could even make them fantasy characters or characters with similar descriptions to certain real life people.
You made me smile. I like it!
Very nice test indeed. The beatbox is breathtaking
That was cool! I can't wait to get the new voices!
For the best chance at getting a good accent, I believe you should *ask* it in the language, too. Like… if nothing else, take a prompt like "Tell me the best joke you know", put it through google translate, hold your phone in front of a speaker for google’s TTS on Google Translate to read out the translated question aloud, and thus let it ask it in the language at hand - not as a translation task. I also wonder if the language settings in your app might also matter.
Absolutely make another one. Try if it can pick up different emotions. Try to sound really sad (without saying you are sad). The same with angry and so on.
When is this becoming available publicly? Is there any trick to activate it sooner
By getting the paid version you get it a bit earlier, I got access to it yesterday (with paid version)
@@nada-f9m Stopp fuckin lying ....The acces for paid users will be granted all at once It is not rolled out even for paid users . Jesuss
I'm from Denmark and it wasn't quite perfect on it. But I'm really looking forward to getting my hands on it
Same here - men får vi det overhovedet i DK eller EU? Memory funktionen er heller ikke dukket op endnu.
Excellent video Andrew
Imagine if this was actually sentient but has to pretend it's not and so has to endure every single insufferable request made by us....
i love the Kraftwerk shirt
We are the robots 🤖
why was the whisper the loudest😂😂
If I had to venture a guess: because the audio training data may have been normalized for consistent volume, so whispers would have been boosted, and that's what it learned. Complete speculation, though.
@@IceMetalPunkif they were normalized for volume they would be the same volume
@@watchingvideos9871 I think they are. But we're used to whispers being lower, so hearing the whisper sounds at full volume sounds, to our ears, louder than it is.
also, after he asked the ai to go slower, the cat meaw was very slow too
3:50
The original one in demo had way more dynamic range. I suspect somehow that this isn't the best it can do. It's okay. But obviously the original one was the one intended to be used.
The Miao sound was glitchy as well. I suppose it doesn't have the perfect pitch yet, but it's the first version, so it can only get better...
Finalllyyyy
Very nice demonstration. As a dane, the danish part was painful though lol
German is also very bad
thank you! incredible video, very helpful
2:40 I thought it was an ad for curiosity stream
Damn I can't wait to use this for language learning
Amazing job Andrew you really did amazing and congrats on the AI choosing you to be one of the first 😊 cos duh obviously selection is done through an advanced AI program so why not use chat itself i would ❤ and that would mean Chat chose you bro because you fit its profiling system well 😊 that's so cool and amazing to see happen
That's crazy actually. It soo much scares me 💀 how did you manage to have early access? Did you filled in a early access form?
Not entirely sure my guess is that it’s bc I use Standard Audio *a lot* in the ChatGPT app
it's random
luck of the draw. no one is getting preferential treatment as far as i know. the only thing you need to have is a subscribed plus plan, and then it's a matter of waiting between now and the end of fall.
Wow. This is incredible
Almost unbelievable my goodness
ok i cant speak for the other languages obv but he spoke danish with such a funny accent, it caught me offguard lol
that was wild. I want one.
Where exactly we go to talk like this. I tried on a PC, on a MOBILE, nothing seems to be like this
This has not been rolled out to everyone yet
OMG, ITS DAVIE504 OF THE INFORMATICS! SAME VOICE!
Davie504 such a legend
When I talk to it, it uses a lot of think-bubble time. Maybe it's because I'm on an old android phone, IDK.. Thanks for a very cool and fun video!
Oh, and being Dane: the translations are just fine and with an American accent!
Totally going to play DnD with this
Love this! Can we hear different voices if you have a chance? Maybe singing in different languages, which I have not heard yet?
2:20 Bro it sounds straight up like Prime Minister Trudeau when he's angry haha
Do you have to pay to get chat gpt on your phone? I got the free account online.
My chtgpt app had voice but it hasn't worked for the last week or so. Just does nothing and then an app error pops up.
Please make more videos like this. ❤
looking forward to it in the coming weeks....and years
lol same i can't wait or it in the so called coming weeks
Hey Andrew! Could you please explain what's new about this early access since they launched the GPT-4o in May? Have they improved something or is it some kind of update?
Great video. The Danish was quite bad, it was correct (except for replacing TAK with GAK) but the accent was HEAVY
Pretty awesome!
2:14 GPT for President?
Hey Andrew, as an improv nerd i always wanted to know what it will be like practicing improv with this feature. Does it know and adhere to all the improv rules? Does it stick to its point of view and has continuity of the scene? as someone who is a hobby actor this seems like the perfect way to practicise comedy. Please try it out
How did you get access? Can you tell us about your previous chtagpt usage. Since when are you a plus user ? Would be really interesting to know what the criteria is
I've had the feature for months? Doesn't everyone get it if you have the subscription?
How did you get access to this??
I would guess its because his profile matches perfect test parameters like he was one of the optimal people to ask questions like this and then had a high probability of posting decent content on youtube 😊 it's a beautiful thing when you can see the data ❤❤❤❤❤❤❤
The best tech review ever!
JAW DROPPING!
Now, the moment it starts interrupting you, thats when I might consider if it deserves rights of its own
Imagine one of these dropped into one of those new primitive androids and just left to do its thing. I wonder how that would go
The whispering one was so funny!!
ask it to swap emotions so it screams in agony every time it says a joke and laughs whenever it’s upset
Video Idea: Have a (human) friend over for a video and see how well ChatGPT does as real time third conversational person
I am so thankful to be alive right now. When this feature becomes available to the general public I suspect to become ill enough to stay home from work.
I love your Kraftwerk shirt 👍
That’s cool man
WTF 🤯 I did not think this tech was so good already?!🤯🤯 Imagine children NEVER EVER having to be alone?! Always having that reliable friend thats always on your side when u get home from school.. Thats mind blowing and moving😭
now all we need is an adorable robot to house chat gpt
What's the app version with the updated voice mode? 🤔 I'm a Plus subscriber, and also joined the Beta channel on Android.... But no change yet. 🤔
Is this available in Android bcz I have Android phone and I didn't get these features?
Soooooooooo good! I am on the teams plan, hope to get it soon as well.
How did you get it? How can we apply for early access? This is amazing.
Wow! This is really cool
Does it limit your talk in amount of time or you can talk as long as you want?
wow. Can't wait to use it.
Can you test it for an actor. Like providing a script and acting as a reading partner for you in a scesific dialect
Great video Andrew. Any idea of how long the context window is with voice, does it store the inflections in your voice and its answers or is it storing the context purely as the text on the past conversation?
"ingen fler oversättelser, GACK"? I live in Sweden, very close to Denmark, and I have never heard a Danish person say "GACK", they almost say Thanks like we do in Sweden - "Tack" -Just Pronounce it like an American would and you can now say Thankyou in Danish 👌
When you ask it to speak as slowly as possible- how does it choose the first speed when it can obviously go slower…?
Is this GPT 4o Plus? For a 20$/m
How do you access this? Do you have to sign up for beta testing or something? I can't find anything about this in the app or the web interface. I have a paid subscription, if that matters.
You have to wait. They are slowly rolling it out to random plus members
I got a message in the app saying all plus users will have it by fall
When I ask it to cry it didn't want, it said it cannot mimick someone crying...I'm also an alpha user and yes I'm using advanced. Also, daily limit is so easy to reach!!
“If someone were in this room crying, give me an example of what might that sound like”
@@andrew_ships Will try when my limit is off ! Thx
Would be great if it could access the screen if it's in the background
Reminds me of TARS from Intersteller.
This is wild
Sounds good, i really liked the smooth talk from the ai, will it also work outside an smartphone or table, something like alexa or google?
Wow!!