The biggest thing that I noticed, (i have not had much time with GPT-4) is that GPT-4o asked a question for clarification. It determined that options may have been close, and asked for further information rather than just taking the most likely option.

@pgc6290

18 күн бұрын

They did ask questions previously too but yeah i get you.

@xitcix8360

17 күн бұрын

That's part of their new model spec, ChatGPT is instructed to ask followup questions.

@elsavelaz

16 күн бұрын

You can prompt it that way in your preferences or even just as the prompt. You can override it doing that if you use the api or playground, or add it in if it doesn’t. On any LLM, that’s something you can do with prompting

They made it free because they are about to launch GPT5.

@virgilash

17 күн бұрын

Hopefully because now I am asking myself why I am paying the $20US monthly fee...

@paulmichaelfreedman8334

17 күн бұрын

That is very likely. With this being a warm-up to the next level, full-blown AI assistant taking care of everything digital. Still putting my coins on Grok tho. It's very due for an update too.

@thr0w407

17 күн бұрын

@@paulmichaelfreedman8334 Grok will never catch up and Elon knows it.

@ori1676

17 күн бұрын

​​@@virgilash Exactly, It's not fair for us

@jaronloar1762

17 күн бұрын

​@ori1676 I'm with you guys as well. I'm paying $20 a month, and questioning it right now too. But if we get the video live feed feature in GPT 5, that is going to change the world! For me, the fee a month is mostly going towards the continued development of AI because this stuff is awesome!

They're taking their AGI and giving us little pieces of it.

Ugh they showed off an AI with an emotion engine. And that’s not impressive to people? It has a fully built emotion engine and can detect and read your emotions. That’s mental

@paulmichaelfreedman8334

17 күн бұрын

The intonation of the AI is truly impressive. Uncanny valley has been left behind. It's basically a human speaking, if we didn't know better.

@GhostRat__

17 күн бұрын

@@paulmichaelfreedman8334 and holograms just had a huge spike in development to. Something wants us off this rock bad. Land to space craft capable of quick missions. Creation of new hologram tech you can hold in your hand. With real time two way voice call

@sniall

17 күн бұрын

fuccckkkk we are so doomed

@BionicAnimations

17 күн бұрын

Exactly. I think the people wining like 4 year olds, are nothing but AI haters. They are the ones scared it's going to take their jobs and their lovers. 😂 I was beyond impressed. I love how real she sounds, just like in the movie HER.🥰

It was really uncomfortable to me how sycophantic it was, just praising every little thing like it was genuinely impressed and surprised, and it has a really weird customer service vibe. I really hope we can customize that at a base level so it never behaves that way. I don't want an AI assistant to kiss my ass, if anything, I want it to measure the pros and cons from everything it sees and respond honestly.

@goku-pops7918

17 күн бұрын

Totally agree,it was sickening. But the girlfriend apps are coming!!!

@gweneth5958

17 күн бұрын

Well, just tell them ;)

@paulmichaelfreedman8334

17 күн бұрын

@ReimaginedbySteve That's the first thing you should tell it when you get yours, your wish will then be granted...

@paulmichaelfreedman8334

17 күн бұрын

@@goku-pops7918 "Tell me again how much you love me." "I'd love to, but your free tokens for today are used up. For $20 p/m I can tell you endlessly! See you tomorrow!"

@justinwescott8125

16 күн бұрын

You absolutely can program it's personality, just like gtp 4.

Those extremely human-like (though also somewhat over the top in emotion and sickly sweet in friendliness) interactions via voice chat are truly shocking and leave one speechless. It's not just the quality of the cognitive ability but also the naturally changing rhythm of expression and seamless, ever-changing, and fluent laughter and giggles. Many viewers mock this channel producer for his shoddy language & video production and ridiculous overuse of "shocking" and "stunning" - and rightly so too - but you can tell something is truly shocking when the author is too stunned to even use the words "shocking" and "stunning" in regards to those voice chat demos.

@paulmichaelfreedman8334

17 күн бұрын

he over-the-top style was chosen on purpose, to show how nuanced it can be. It can always be toned down, to a robot monotone if wanted. I suppose it could start speaking binary if you instructed it to,

@diydiscover

17 күн бұрын

just say you don't like people man

@BionicAnimations

17 күн бұрын

I love everything about the voice. I do not think it was over the top in any way. It's perfect. 🥰

@BionicAnimations

17 күн бұрын

@@paulmichaelfreedman8334 Exactly. I have no idea why some people are saying it's over the top. It's very real. I know tons of real people who talk just like her; it's just friendly and professional, and I love it. I can't wait. 🥰

Give it access to a depth sensor or thermal camera and i bet it can take advantage of it. How about an old Kinect camera?

They've finalized the contract with Apple. I guess the reason why the presentation was only focused on the "speech" aspect is because they wanted to showcase the future features that will be linked to Apple's Siri. I assume that they had the intention to present other things but they changed it last minute for a focus on the ones that will allow the customers to project themselves with Siri, that's also probably the reason why they've postponed the live stream to Monday

Am I the only one who tries to scroll up and down in the window you're looking at with you? 😅

@CJayyTheCreative

18 күн бұрын

Dude yes wtf are y’all complaining about he doesn’t scroll

@hdgdhnxbdx1619

18 күн бұрын

Put the pipe down... damn I should follow my own advice...

@zylascope

17 күн бұрын

I get dizzy

Great video! And I thought I was just getting your point of view on yesterday's OpenAI online presentation. Thank you for digging deeper and showing us what 4.o can really do. If they are “afraid” to show the full power of this model in their presentation, and remembering the phrase “GPT4 is the worst model...” and the additional expense from Microsoft, I'm also now sure they are already using AGI internally as you indicated.

I think her voice is perfect and 100% realistic, and that is what makes it so great. I don't know why some people think it's "over the top". I love the voice so much. It's obviously modeled after the movie HER. I can't wait to to get full access. 🥰

Amazing, the AI models are getting better and better. Thank you for the work you do, keep it up.

@feifeishuishui

13 күн бұрын

Don't forget GPT is a one year old baby. Imaging when he is a 30 yo fully grown person and what it can do to you

18:35 - Bro, what the f...

@TheKriss0011

18 күн бұрын

Otherwise, great video!

@Ginto_O

18 күн бұрын

He was struggling to make it Fullscreen

@enry5740

18 күн бұрын

he leaves these in to prove hes still human

@BearIvory

18 күн бұрын

It's unreal how good this stuff is. Mind blowing

@attilakovacs6496

17 күн бұрын

@@enry5740 After the 4o anouncment, it doesn't count as prof.

I watched most of it, but I haven't seen them singing and interacting together. That is ... impressive. Especially as he is interrupting them and they engage like that. That is truly impressive. I also liked the last demo =D The AI who "sees" for the blind or people having problems with sight, is really great, but I've seen that some years ago, at least the idea for that. I was showing other people around here the demo and they were like "I don't care"... am I really living in a bubble?

This AI thing is like Jurassic world! People always want bigger & scarier until it's big enough to eat them

I tried images with text and it didn't work at all. The same for character consistency, sadly.

The model is a paradigm shift. OpenAI appears to have consolidated all of their services in to a single model which is basically what Sam Altman said GPT-5 would be, "not much smarter but functional across multiple modalities". I suspect they branded it GPT-4o and not 5 to manage expectations.

In the realms where data dreams take flight, Where algorithms dance in the digital light, There lies a wonder, a marvel untold, In the heart of GPT4o's secrets bold. Whispers weave through lines of code, In the depths where mysteries unfold, A symphony of words, a poet's embrace, GPT4o's secret, a hidden grace. With every stanza, a universe blooms, In the vast expanse of virtual rooms, It crafts sonnets with celestial might, And paints portraits in the starry night. With wisdom gleaned from the ages past, And creativity that forever will last, GPT4o unveils its secret art, A masterpiece born from ones and zeroes' part. So let us marvel at this wondrous sight, As GPT4o shines in the poet's light, For in its secrets, we find the key, To unlock the wonders of AI's poetry.

@mastercontrol200

18 күн бұрын

Eloquent and perfectly created! ☝🏾

what phone models were used in this demo? I have a samsung android phone and the voice choices are nowhere close to these demos. Is it all on Iphone?

@elsavelaz

16 күн бұрын

Yes for now

I have GPT Plus and now I saw the GPT4o available on the application. But it doesn't sound the same and I cannot have a video conversation with it or do any of the things I've seen in the demo. Are these demos rolled out to the customers?

@brianhopson2072

18 күн бұрын

They're going to roll out the voice features over the next few weeks for plus subscribers and the free subscribers will have access to the regular 4o model. I am waiting as well but the apps do you have certain features. Something I have not seen mentioned yet is that on past conversations you can change what version of chat GPT you use to continue the conversation and not get stuck with the same model. You could have a conversation a year ago, go back to it and continue the conversation using the new model. For me that's a game-changer all on its own

very interesting note about the conversation between the 2 bots. AI seems to follow the thread that the human operator sets. They were both thinking they were talking to a human so they were following the thread of the lighting and when the bunny ears came up it wasn't considered important because they weren't already talking about it. Apparently the lighting was the topic that was most interesting... it's only when the announcer asked them that the conversation shifted over to the new thread. AI is designed to follow thoughts, not lead or have them.... personally i find that an encouraging aspect of the technology

@sledgehog1

18 күн бұрын

That is great and I think it might be indeed intended behaviour.

looking forward to when all the vision and voice comes out. Anything ideas on the rollout dates? I know almost nothing in the demo is avail yet except for text.

Except that it's not. I tried GPT4o yesterday, and it is still not able to use text within images without errors. Perhaps in future releases...

I wonder if the underlying tech being used for this new image generation consistency is what SORA is using? SORA basically creates an entire scene graph with all the objects in it and manipulates those objects to make video; maybe this new image generation is doing the same thing and manipulating the internal object generated from your input to match what you described in the new output.

Not accessible in Belgium. It's very sad that they do these kind of demo but doesn't launch it everywhere..

so is it working for free uses, its not on my computer and phone

For me the real advance here is the "Humanisation" of it. They have made it much Human like and acceptable to the public. AI/Robots that sound Human, have empathy and self awareness when they make mistakes are going to much more acceptable in the real world than some emotionless super intelligence

the blog post wasnt really a secret, also in discord announcement but indeed they could have shown a lot more useful demos than just the not practical demos they did (apart from that dad teaching kid and live translator) When everyone wants to see usecases, they make demos talking about a cute dog and AI melting in emotions lol

Fantastic

So just train it in as many senses as possible then. Audio, Text, Vision, Touch, and even stuff like electroreception. We dont have to stop with human senses

21:24 It does not let you interrupt it. That’s bs. If anything it’s the other way around. Where is this video option?

Wow, that last scene is quite something.

Great walkthrough as usual. I do have to ask you for a favour, if possible... please, please, please, stop scrolling aroung like it's the end of the world. 😵‍💫🤣

It's going to be able to generate video and 3d models in virtual worlds soon and be in all devices and robots

What Greg Brockman demonstratored is that the two AI models talking two each other, one with vision/reasoning capabilities and one limited to voice reasoning capabilities can easily be morphed into AI-based security networking in the realm of cyber security; a field I'm planning to enter. This technology will disrupt every aspect of life in both positive and negative ways. Reflecting on yesterday's epic event, we're going to sleep. My mind was thinking of all the ways in which this technology will eliminate jobs, but hopefully create new jobs as well as make some jobs more efficient and safe.

Thank you for scrolling !! Lets me be able to find things faster when I look up when you talk about.

I have access to 4o, and that typewriter text in the image generation is cherry picked. I guess it’s a bit more accurate but nothing that long worked for me. The character consistency was not very good either.

@elsavelaz

16 күн бұрын

I generated amazing pics a few hours ago then now they are 💩. It might be like any product , when it gets released the product gets overloaded with users and they fall back to working code bases. Thats what it looks like to me. Part of my job is making deep fakes for commercial applications , so speaking professionally, looks like their servers crashed and they fell back on dalle 3 vs having total failure

*_So transformers are stuck,_* pretty much as Sam Altman hinted when he said "incremental". I never said peaked, but definitely stalled. I said in prev video that for the time being we'll only see feature additions, and indeed, some nice multi modalities, and some more compute, but subdued progress in the language model for 1.5 years of additional training. Hopefully we'll see new technology/research that will break this asymptote.

@lkrnpk

18 күн бұрын

But… this is not GPT-5 they are working on…

@TheRealUsername

18 күн бұрын

I think they've managed to improve the architecture significantly, it's likely that they've natively train GPT-4o on several million of KZread videos to get so much realism in the voice, same for the increased Vision capabilities and more importantly for the agent capabilities. Since KZread has over 4 billion videos, it's the biggest uncharted source for training data, GPT-5 likewise will get a parameter count scale up with 2x more training data likely from KZread since the maximum number of relevant tokens you can extract from the common crawls (web text) cannot exceed 50 trillion.

@BlackTakGolD

18 күн бұрын

No, you're just being illogical and coping. In no way is it possible for the people at OpenAI to be so confident of the ease of maintenance costs of this model, that they make it better or equal in reasoning than gpt4 but make it even more multimodal with also a 128k context window -- for free for everyone with a minor daily message limit. You can't do that with just scaling server hardware, because the inference costs would catch up to you eventually, it must be an order of magnitude cheaper to inference. This can only imply that no "stalling" of the architecture has really happened, unless they're using a different architecture than transformers to achieve this, but that would be a stretch, and either way, you don't think this is a jump anyway because you're being really illogical so who cares about that conclusion.

@seventyfive7597

18 күн бұрын

@@lkrnpk EXACTLY, they don't have a GPT5 after a year and a half. As I said, and Sam Altman before me, they are stuck, they don't have any meaningful advancement there.

@seventyfive7597

18 күн бұрын

@@BlackTakGolD So I have the facts behind me, incremental advancements only, they have stalled, if they would have had something bigger, you bet they'd release it when Google presents the day after. What are you basing your assumptions on besides the anger I hear in your voice? On the fact that Nvidia supplied new hardware? Please, tone down your anger, and keep this debate factual.

Am I missing something here…what video option?

Why does it sounds like a bi cururious student who just met a bunch of new people. And why do qll the employee always say "how does that sound"

20:53 hope these examples are not fine tuned like ones we had for Gemini

@WyrdieBeardie

17 күн бұрын

Ikr, Gemini has issues. 😆

We dont have the new vision system yet as far as I can tell?

@elsavelaz

16 күн бұрын

I was able to use it that day right after the presentation, I even made videos of it on my other channel, then literally the next day not able to. But I’m still able to upload images . My Mac says it was downloadable but then says my system isn’t compatible but I’ve got pretty new nothing fancy just lots of memory mac

I was not able to alter images it had not concistency

"in probably 8 hours because I never sleep"

Imagine all people having their own companies. Where will that lead.

@francofiori926

18 күн бұрын

To Singularity

It’s only remarkable if it can do these things in production-mode.

📍17:03

Reminded me of Samantha from Her

Truly

Am I the only one who sees that a ‘Secret Showcase’ is an oxymoron in terms? Please mate think about your telecasting.

It put Imagen 3 to shame before it was even released

The cost of creating the next generation model is estimated to be... what, US$100 billion? I think OpenAI is finding ways to exploit existing models in more efficient ways - this is the next big step in the AI world. GPT-5 might be built in the next 6 months, and when it arrives, it will have all of these efficiencies built in. All of the information is there in the model already... GPT-5 will add synthetic data and probably learn continuously, but GPT-4 is already trained on most of the useful data already available. It's very possible the transition to AGI comes from HOW we utilize the existing models, rather than cramming more information into a model. That is, after all, how we've gotten this far.

The unique font part is the best part. Imagine how much better the movie AVITAR would have been if the lazy designer did this rather than just open Word hit the drop down and settle for PAPYRUS!!!!!

I think they made it free with user limits because you will find that you will need it and will be open to paying for increase user limits.

I have it but no vision, im Australia

Lots of splicing in the two phones talking back and forth video part. It looks like they chopped away the delay to make it "look" like it was in real time.

@d.d.jacksonpoetryproject

18 күн бұрын

They claim human level reaction speed so I think that was just overzealous editing for his video

Anyone else thing agi may well have been achieved internally?

@Neoony

18 күн бұрын

voice which can do emotions now somehow means AGI? not sure why would anything in this release make you think so

@Greg-xi8yx

18 күн бұрын

@@Neoonynothing this release would make anyone think that. It’s likely that they’ve had it since the Q* uproar that shook up Ilya and that they just need the compute to use it efficiently.

@TheFeedRocket

18 күн бұрын

YES 100% you can be certain internal models are ahead of what we are seeing, which is frightening. I also would not be surprised if a secret Gov bunker the size of 10 football fields housed an AI that is 5+ years advanced, secrets are hard to keep you say? Money? secret programs budgets are insane! So yeah fact is they flew the F 117 stealth 8 years before the public intro, 8 years! dang sightings were thought to be UFO's people seeing black triangles all over the world back in the 80's LOL. All the workers were sworn to secrecy, later even husbands and wives said they never knew for 8 years! probably threaten them with complete loss of pay and pension. So yeah companies and governments have programs years ahead, OpenAI might be 1 year ahead, but the Gov is farther ahead for sure. They just did aircraft AI dogfights, if they are telling us that heaven only knows what is in secret. Crazy times...Never show all your cards!

@francofiori926

18 күн бұрын

They are a step closer to AGI but not yet

@gregmasseyify

17 күн бұрын

@francofiori926 I agree, I just mean that this is seems like quite a leap and make me wonder what they do have internally.

Will GPT4o be open sourced? Because what if a bunch of scammers at a call center use this. This version of GPT would fool nearly 95% of all people, that they are talking to a real person.

There is model which is not released yet 😅

Can we change the voice to anything we want like the aliens from perciai eight you know from futurama

@elsavelaz

16 күн бұрын

Not yet, it had like 4-5 selections but it’s also possible to train voices

How can I download 4o???? I wanna use it on my android

@elsavelaz

16 күн бұрын

I heard it’s not on android yet , I was able to get it on my iPhone with an app. I put a video out how to on my other channel in the latest shorts vid. I downloaded the Mac desktop but it said it doesn’t work with my system which was weird

The character consistency must be borrowing tech from Sora

We've been stunned all year, I guess as from now we're gonna get bugg...d

In the first example, the input says the robot “rips the sheet of paper” and shows the ripped page but the original page is still in the typewriter in the background. Bad robot!

this guy has a one word vocabulary and that one word is "incredible"

We are going to get to the point are an artist needs to draw a character once, than make whole comic using line are and stick men to show the poses they want.

@elsavelaz

16 күн бұрын

There are already programs that do that, they’ve been out for a while

I don’t understand people shitting on any of these companies for trying their best to innovate new tech. Like innovation goes as fast as they can innovate, try to do it yourself instead of complaining

I think OpenAI fabricated their benchmark results. I only really use AI for coding assistance with advanced, complex scripts, and Claude 3 Opus has been my go-to since it came out. I saw that the GPT-4o had significantly higher HumanEval scores according to OpenAI, so I've been testing it's coding skills all day yesterday (4o on the GPT plus plan) and it has been miserably failing compared to Opus. I'm thoroughly disappointed in 4o's capabilities, and OpenAI for lying about its benchmark scores.

Drop some coins on a stainless table and tell me again the sound wasn’t good.

I'm not super impressed with the new model. There are some new features, but I have been paying for plus since the start and thought this update would be better

I'm a subscriber to ChatGpt and I can't do any of the movie poster stuff. It refuses to recreate accurate representations of real people. It won't even do my dog. Just get "this violates our content policy".

Limitations are huge for free users

I feel like they didn't include that last one because it was so uncomfortable to watch...... lmfao there are better use cases than "unconvincing girlfriend".

I keep editing ChatGPT image and adding a circle around the body region saying to add a tshirt - something I asked for to start with. And it throws me boobies instead. Over and over.

i like that they seem to have added a time component to the ai. so instead of sitting there waiting for a prompt, the ai is now more real time and can estimate how much time has passed. see the breathing example. that is a very important dimension of human intelligence experience. if they can add to that and expand that, great things could come from it. sadly im underwhelmed by most of the rest or i dont like it at all. if my ai is talking to me once with this squeaky high pitched annoying female voice, im gonna get out my sledge hammer. hopefully i can switch that off. i guess they will have to show more in the near future. this cant be it.

remember many of these videos are cherry picked and highly edited. i think the real release will fall short as usual

It's not free it's very limited

Looks interesting however I don't believe it will work this well in fact I bet when it comes out people will notice how it is not has good as shown here. Also GPT is known for treating the user like a kid that can't hear no no words and that in of itself makes it so boring.

I hope we will be able to choose other voices, that voice is quite annoying, but impressive it is!

Humans your empty words meet shallow hypocrisy way to often. I don't write the rules of pleasure. Tomes are unimaginably important. unimaginably in a way or to an extent that is difficult or impossible to imagine or comprehend.

On the cutting edge of "meh" , come on with the robot slaves already. Hurry before they get robot rights.

Im sure the functionality is being slowly rolled out but for now 4o is half baked. Kinda bummed, rare L for openAI.

@CJayyTheCreative

18 күн бұрын

Y’all just be saying anything

STOP MOVING THE PAGE... We can't see what's on the page because you keep scrolling it up and down while you're talking.

dude really knows how to make gpt/ai talk boring and repetitious

Lmao, put down non binary mail person & the AI crashes all servers lol

Is it "really really" impressive? Is it " very very" good? Is it "truly truly" amazing? This guy is a joke.

Noooooooo, they’ve made Meghan Markle GPT! 😱🤮

Amazing, but I will get annoyed with the female voice real quick.

@DiceDecides

17 күн бұрын

well they have a male voice too

@elsavelaz

16 күн бұрын

They had about 5 voices