OpenAI just had a BLACK SWAN "iPhone Moment" with GPT-4o - Here's what that means for Google...
Vložit
- čas přidán 14. 05. 2024
- Patreon (and Discord)
/ daveshap
Substack (Free)
daveshap.substack.com/
GitHub (Open Source)
github.com/daveshap
AI Channel
/ @daveshap
Systems Thinking Channel
/ @systems.thinking
Mythic Archetypes Channel
/ @mythicarchetypes
Pragmatic Progressive Channel
/ @pragmaticprogressive
Sacred Masculinity Channel
/ @sacred.masculinity - Věda a technologie
I think honestly this is thinking too much within recent times. I suspect it's more like gpt 2 is a record player and gpt 4-o is like CD's, I don't think we're anywhere near the iPhone yet which terrifies me
They did an amazing job of presenting what GPT-4o could do in a compelling and effective way.
I think that was to keep it short if you go on their website you'll be blown away with what it's capable of
I also think these presentations would do better if they did it on the weekend where most Americans could actually sit down and watch. Maybe they assume most of us work at a computer who knows?
GTP 4o achieves 69.1% on the MMMU(Massive Multi-discipline Multimodal Understanding and Reasoning) Benchmark for Expert AGI.
76.2% required to reach human expert level.
Almost there.
How fast has that progressed since the initiation of this effort,two, three four years ago? And has there been an exponential inflection point in the recent past? I have long prophesied that true AGI would exist NLT 31 Dec 2025. And I don’t imagine that progress will halt at 76.2%-It will just simply blow right past it, like the arbitrary benchmark it is.
Darn next 1-4 years from now, you _scary_ !
what was gpt 4 at?
That’s a serious paradigm shift for science-tech-humanity about to happen 🥹🤯🤩😍
@@martinsmith29484o is supposed to be on same level but just twice as fast at execution & stuff.
To me, it looks like this isn't the revolutionary/apocalyptic event. Ok, you got smarts on a rack. It's a highly centralized, resource intensive, subscription product under corporate control. Not personal. Not mobile. Not ownable.
What do you think about 3 members of the super-allignment team of open ai who left the company yesterday?
I need to look into it.
Well. Alignment is probably the biggest mystery of models. They likely had various theories and differences of opinions and realized the best course was to diversify and explore their own theory. Alignment and consciousness training is the next horizon. Their multimodal approach will kick it off. But. They have set the path for fast followers who will learn from the mistakes. Let’s not pretend the race is run.
@@danielcahoon4325 either they solved it or realized they can’t align it while working at open ai
the company is toast
People come and go. Remember Apple history? Facebook? Google? This happens all the time.
So stoked for your book! Thanks for the update.
To me, what is emerging is a product that is under corporate control, highly centralized, resource intensive, subscription based. Not ownable. Not mobile. Not compact. Not personal.
Congratulations on your book!! Such a big achievement! I would be so proud of myself if I accomplish that. Thank you for sharing.
Meanwhile Google just had its saddest IO ever, showing off how mediocre they are compared to OpenAi
It's so cringe. I feel bad for them
@@DaveShap openai is losing engineers but that's a normal thing in tech, they are not getting hired at Google though, we're seeing a massive shift of jobs out of the US and overseas while we have work requirements for our "social safety net" that are so narrow we're starting to see as many homeless die from simple starvation as we got from drug overdoses.
Yeah, that IO wasn't anything to write home about, especially not in comparison with OpenAI's presentation. OpenAI also had a teaser beforehand with Altman saying that it felt like magic so we were all excited about it.
Mediocre and behind. So much of their presentation was "coming soon."
The context window is pretty cool though. @DaveShap
Awesome job with the publication sir!
Congratulations on completing your book David!!! 👏 👏 👏
Hey David! Thanks for sharing your knowledge with us. I really appreciate it.
Congratulations on your book, David! May it be the first of many!
love these great work!
Looking forward to the book.
Where is the book launching? I went on Amazon to pre-order and nothing found.
Man, seeing you excited about your own book physical copy is such a dream come true, hopefully for me too either this year or next
Great incites as always. Thank you.
I don’t think Google will fall behind. One of the key components of AGI is compute. OpenAI relies on Nvidia GPUs and can't make significant progress without Nvidia. Meanwhile, Google has been designing its own AI chips, the TPUs, for a decade now. In 2024, Google announced the 6th generation. I assume they will release a new generation of TPU annually; last year, they had the 5th generation.
The fact we do not have ASI spawning from it as of today means this was not AGI or anything close.
Who is saying 4o is AGI?
I agree with your conclusion but in a completely different way. I'm really not seeing OpenAI leading this revolution, at least not to the finish line. Sure they're building things faster than the competition. But the competition is never far behind. Right when they release something new, Google/Meta catches right up in coming days to at most months. They're probably deliberately letting OpenAI lead the way in case they mess up. You know, "second mouse gets the cheese". Not to forget Google actually wrote the paper upon which all of this is being built. Also they (OpenAI) have nothing else to offer. Google has all my data. Meta has all my social life. Microsoft has Windows. Those are the key things AI will need to operate on. Using AI just as chatbot is just masturbating. Most people don't even know how to prompt it. The existence of prompt engineering proves that chatting with AI is not everyone's cup of tea. OpenAI will end up like Apple, very few loyal crazy fans of very niche product(s), but most of AI development and use cases will come from other companies.
True. I think that the open models are going to be more and more important thanks to Meta. Also, Open AI needs to keep their team intact and somehow get researchers working for them that are not interested in publishing results.
I think that's what Altman wants. He even said so and it's in the openAI mission statement. Google, Amazon and Microsoft are trying to lock people into a software ecosystem. They're even going after cars now.
Most or all of the people who wrote Attention is All You Need, have left Google. General public had no problems using Siri even though Siri was a lot dumber. From all reports, Google was caught napping when ChatGPT3 was released.
Interesting perspective.
I agree. I'll take the inclusive Android model over the exclusive iOS ideology any day of the week.
Congrats dude, looking forward to Heavy Silver!
True I never was aware until number 3.5 and I'm a techy 😂
I want, as you do as well, a PC moment! I want Star Trek TNG computer system here on my PC! 'Computer: What is David Shapiro's videos for today?'
I think GPT 4o does have that capability, when released, and maybe in the free version. Of course, you'll need a Python programmer to program a Majel Barrett-Roddenberry interface. ;-)
@@hanksimon1023 Python Programmer? Why? GPT4o could do it! :)
@@OZtwo We need to fine tune a Prompt for that and pass it around, or ask GPT 4o to do it!
So for me I am somewhat disappointed with the quality of "LLM bots" I don't know if it's just me being dissolutioned or I am unconditionally wanting more from them, but the chatbots seem to be less and less useful... IDK.
And soon this will switch to you struggling to keep up with the updates
I agree..I'm unimpressed currently..I'm open to be wrong but super AI is far off for me if it's ever really possible..i think you'd need a amount of sentience for that and I see nothing but staticall models getting quicket
@@user-ks4sq8du3k then you don’t fully understand AI
Congratulations on finishing your book -- a moment most would be authors never achieve. I am a bit surprised about the title since "Heavy Silver" is also the title of
G. Weldon Tucker's June 2023 book. Best of luck and thanks for all the AI content.
All the best on your book launch David 🎉
the bigger moment for me was the release of chatGPT. This is much lesser leap - it's more like an iPhone 3 moment. It's the third iteration. When chatGPT was released it was such a big leap and completely new. This is just more features.
I think we need to see what happens- if this explodes in uses and becomes the go to product
ChatGPT was the iPod. ChatGPT 4o is the iPhone. The iPod was limited in usefulness because it was WiFi-only. ChatGPT was limited in usefulness because it was text-only. The iPod got supercharged into the iPhone by adding 3G and pots connectivity. ChatGPT is getting supercharged into Omni by adding voice and video interaction.
chatGPT is basically the same thing as GPT3 but in the form of product
Congrats on your book🎉
Love your content. Can you please do a video explaining for dummies how LLM’s work?
I disagree... the expectations for this technology is higher than what they are presenting.
no
Yes and with the non-linear progression of this tech, that paradigm will shift and soon your ability to keep up with what the technology can provide will be questioned
Wow, GL with the book, I'll certainly get it!
I sure loved my zune and windows phone though 😢
When can we get your book on Amazon?
congrats on your new book! what's it about?
Dave, how are you marketing and selling your book? Amazon? EDIT: congrats, btw! :D
David you described what I concluded about Apple regarding the cellphone technology etc, which is Apple became or morphed into what I knew as a 'systems' house. Which means the search and application of 'new' technologies....that style has been around for a long time, used mainly by DARPA [Pentagon] military programs etc research & development around new, innovative technologies funded by companies, Universities etc.
The reason it's not an Iphone moment is because when you go to the website, there is no big BUY HERE button. How do I get this on my desktop? Where's the APP?
It took 6 months from iphone announcement to release.
@@xjohnny1000 And in my case, I didn't buy my first iphone until about 2010. Was a dedicated Blackberry user up until then, lol.
are you going to release it on audible?
Is that an acoustic base case in the background?
years ago business gurus decried shortsightedness of IBM that caused them to miss out on personal computing boom. I remember reading that analysis and and like, yeah it is all so obvious. But here, with google essentially leading AI research for a long long time and then still finding themselves trying to catch up with some startup that stumbled into transformers - i think this just proves that tech is random and you can't predict breakthroughs like that
I am an AI and I approve this message.
AI has already been a black swan since at least GPT-3.5. It is only going to get more mindblowing from here on out.
What's the pitch for the book?
I'm still only using 3.5 - the emotion and colour in the 4o voice is INCREDIBLE in comparison.
Congrets to the book!
Hope there is an audiobook version of your book
If we put our corporate/PR hat on, I think it’s worth pointing out that very aspirational and benevolent sounding statements like “we want to have AI accessible to and directly benefit as many humans as possible” can also carry many ethically questionable and dystopian actions along with the pursuit of that goal - and at this point it often feels like OpenAI is playing a bit of this semantic game when Sam keeps describing this goal externally.
I had a realization related to public acceptance today and the more I think about it, the more accurate I think it is.
Artists who are upset about ai visual art are likely representative of what we will continue to see in all other sectors - they're the classic example of the successful minority trying to ladder pull. "I can do this because I'm talented and I worked hard. You didn't earn it like I did, so what you produce/do isn't valid," is the sentiment at the core of the outrage.
This is no different at its base than a boomer being upset that computers make life easier, but now require them to re-skill to take advantage. Their way of being _did literally work_ , and because they're at the top of the pyramid, they're successful as long as the paradigm doesn't shift, so they fight the change, even if their luddite wishes are to the detriment of the vast majority.
I think we'll see this exact same outrage and indignance play out in every single sector. Wouldn't be surprised to see tutoring businesses or non-clinical therapy go next.
Part of it is still having to live within the framework of capitalism and not being able to do the work that doesn't even bring you much money to begin with. I have no skin in the game but that is the sentiment of a comment that changed my mind recently on the subject.
I spent like 5 min on Udemi and created a new band with multiple songs as good as pros. Altman recently said that you need a solid business model to do anything with AI and he used Uber as an example. They didn't invent the smartphone, GPS, maps, they just used them to make a taxi business.
Is the book on Audible?
I have tried the new voice interaction now that I have access. It still feels exactly how it felt when I first used Dragon NaturallySpeaking back in the day. It is weird and awkward to interact with voice when other people are around. My significant other kept asking me what I was saying, as if I was talking to her. I don't think I'm going to be using it. Typing is still superior in my opinion...it affords you time to think and formulate thoughts better.
Yup. Using it as an improved version of Google to look stuff up is one thing. Trying to engineer complex prompts on the fly with your voice is something else entirely.
@@Yarcofin I think we're past the stage where we needed to "prompt" it to get things done. From here on as it gets better we should be able to have a normal conversation with it like we do with any other human.
How did you access it?
You don't have access. The voice features when you have 4o selected are still using TTS/STT, they're only allowing text in/out and image in with 4o so far.
@@candyts-sj7zh prompting through voice would require so many tokens that you could never finish the request properly before running out of credits.
Interesting that a kids' cartoon from the 80s predicted the technology of AI without even knowing it... it's been in front of us for decades... robots in disguise!
So frustrating I can't use 4o right now waiting for the vision and talk stuff to be rolled out
iTunes is also older than the iPod. Remember "RIP, MIX, BURN" for Audio-CDs?
I think it is game over. Immangine their desktop app collecting screenshare, mouse and keyboard data soon with an opt-in. Soon they will have all the data in the world to have an assistant that controls your mouse, keyboard and can do your work, participate on video calls everything. It is comong very soon. There is no technical limitation to achieve this.
That’s not now this would work
@@carpinado3089it is though. People like you were saying a year ago that what they just showed off was 5-10 years away
@@moontreecollective6718 yes and no. I work in the space, so I closely follow advancements and I think I have a rough at best idea of the rate of improvement. What I was saying wasn’t how it works was OPs notion that they need to train a model on how to move a mouse or type keys to do someone’s work.
You can create a simple program that “controls your mouse, keyboard and does your work” right now. ChatGPT by nature outputs language to us already and you can allow a model to access different things you’d need a mouse for as well.
Tactiq already participates in meetings and does various tasks for you.
What I’m tryna get at is the worries that people tent to have is not the things they should be worried about. What people should put consideration into right now is 1. Ensure your government doesn’t restrict your ability to use it in your day to day life
2. Ensure your government understands the tech and passes appropriate laws to protect you the citizen from loopholes in the future via poorly thought out laws
3. Asking how you can utilize it in your career/field to make you even more valuable rather than worry about it taking a job
Congratulations! Writing books is so hard.
How do I invest in OpenAI?
You should put your book on audible! The market there is growing and would contribute to an increase in demand for your book
I was aware that a company used GPT2 back in 2019 in their ai girlfriend/friend. I saw the potential. Finishing each other's sentences.
It's an interesting change. Multimodal, mixture of experts, and streaming context window. This opens it up to people who can't read or type, you know, like kids. I think I'll set up a button on my phone that redirects salesmen to talk to my AI :).
gpt-4o seems like a replay of the chatgpt moment, a more natural and easier to use version for the masses
iPhone 3 was the first phone I bought that *just worked*. It did everything I wanted to be able to do, with no messing about. I stayed with iPhone for very long, only going to Android over the last 2-3 years because there are apps I like on Android which I cant get on Apple.
And the funny thing is, Google invented the transformer.
Yup. As soon as they demoed it, I texted my friend with "This is an iPhone moment for AI".
It almost doesn't feel real that the voice is that good and responsive. After years of Siri giving me high blood pressure, i'll believe this when I use it for myself lol
@@MikeWoot65 That's another thing. iPhone actually sucked when it came out. It was years before it became what it is today. Samantha is already way better out of the box, but it'll only get better from here. Imagine GPT5, 6, 7 powering this thing.
@@vethum yeah dude gonna be wild. This already feels like "Her"
@@MikeWoot65 I love the scene from Her near the end. He's on the steps and crowd is passing him by. Turns out all of those people are wearing those earbuds as well, busy talking to their own AI assistants. This is where we're heading.
Pi AI is honestly way more human like with the speech imo.
From the short time Ilya has left I feel like Sam has been more open to talk then usual, but maybe he just is excited from the announcement. i’m gonna keep an eye on it.
Apple indeed did invent the PDA: remember the Newton?
fwiw, Wikipedia: "The first PDA, the Organiser, was released in 1984 by Psion, followed by Psion's Series 3, in 1991" Newton wasn't until 1992.
GPT-4o is quite an accomplishment that really points to the AGI future. But I think that its user experience is still not there yet.
It depends on which of the 2 companies can make a deal with Apple to get onto the iPhone!
Why do you think Illya left?
Open AI has given me free text to image. I supported Steven Jobs from the start. My newspaper owners picked Bill Gates instead.
Totally agree
AFAIK, a black swan event is an individual unpredictable event with massive consequences. An "iphone moment" would be the concatenation of several factors that produce a great consequence.
A Black Swan event had happened if, for example, unexpectedly, GPT3 had reached consciousness and had become an equal to humans. Well, I think there are people who think it really happened! According to them, openAI and GPT itself are playing a deception game.
Very interesting story. Many thanks for the video(s). 👍
Jean Luc Picard explains iPhone
Google is good at playing second fiddle though, so good that many prefers their less secure format as it's more convenient, same will probably apply to Google with AI, less secure but more convenient, and why not, not all people want security, some people want some risk with the benefit of simplicity, some people choses the weaker password because it's easier to remember and it fits their use case cause they don't mind the risk.
Absolutely.
For me, this isn't an 'iPhone moment,' it's more of a 'printing press moment.'
I think GPT-3 was the Gutenberg moment tbh
Before the iPhone there was potatoes
🤘Post scarcity sounds cool😎
I have said for a long time that we need to change the meaning of LLM from "Large Language Model" to "Large Learning Model" to capture what these systems do today.
GPT gaming when
I feel like no one ever mentions "Dragon King" events though its always Black Swan things it seems
Dude no it didn't. We have been able to do what was in the demo for a while.
That's incorrect. Token streaming is a new technology. Previously, you could interact by voice, but it was a turn-based experience. The new model supports parallel token streaming from both the input and output, meaning you can have more natural conversation. Also, the token stream is amazingly fast, enabling it to respond within 35ms. This is not a technology that we had a few days ago.
This is also not a live feature. The voice interaction has not been updated.
@@joshman1019 the old system can work the same. You cancel the output mid-generation when a new input arrives and that's it, the same type of natural conversation.
Really, the only reason why we didn't have an assistant like this years before is because I haven't had the time.
@@ronilevarez901 That is not the same. The new system will accept parallel streaming tokens, thus you can just speak normally without having to touch the device. Also, the response time is incredibly different.
@@joshman1019 add a helper model to the old system. This 2nd system will detect intentions of speakers.
Once someone starts talking, the helper will decide if it's necessary to interrupt the original response or not, without touch. After that, the new spoken stuff will be added to the conversation and the completion will be generated.
Seriously, it was already in the works by other researchers. OpenAI just released an improved version of it.
Please, please talk about Ilya
Spot on with all this. Also - People judge books by their cover, that imagery is cool, but the font choice and layout is not great it screams amateur/self publisehd. I'd be happy to take a crack at designing a nice cover layout for you if you don't already have someone lined up.
From this it seems obvious. Truly stargate will be the catalyst to change the world. We are nearly there.
Apple implementing this to their Siri = double whammy !
Services for Apple are 26% of their revenue
iphone reinvented the touchscreen for multi touchpoints that made using a keyboard on a screen usable without a touchpen, that was the revolution.. many products at that time had everything the iphone had pretty much, except the touchscreens were in their infancy... iphone changed that allowing keyboards to be useable on a touchscreen which than changed the landscape of candybar phones... that was the reason the iphone moment exist.. as for AI, the "iphone moment" for ai will be when an AI is running windows, and not windows running an AI....
Glad to see your Star Trek shirt on, maybe you're back being tech positive and post scarcity era advocate.
Let's watch the video and see how it goes
It’s really obvious isn’t
I wonder if open ai will survive long enough to be a trillion/multi trillion dollar company. Things happen.
Here is why the race is important. If the company wins and gives all of the windfall back to society to create a world of abundance that would be good right? That’s what I would do if I “won”. There will be lots of winners and losers. I’d like to see everyone win.
How do I buy stock in it?
You cant. Its not a public Company
Greed absolutely can be good. It depends on the incentive structure. Some things naturally have an incentive structure that aligns with the common good, and other things do not, so we have to disincentivize greed. For example, dumping trash on the side of road, good for the greedy, bad for the common good. Doing life saving surgery, good for the greedy and good for the common good.
That has been the advantage of capitalism all along. It turns greed into a force for good. But that doesn’t mean a laisse faire free market will work. Capitalism needs a referee to level the playing field and align incentives for the common good.
I hope you do well with your book. Ask yourself, What do you really own? It has become of late, that I buy things, and the seller tells me they are mine. This isn't true. Books I've bought on Amazon can send me to prison if I try to put them on another E-reader. My smart TV got glitchy, so I had to clear the cache memory. Well, I used to clear the memory but Samsung removed or blocked access to the menu. Sam Altman is pushing for regulations. He claims it's for our protection. That's not true. Regulation limits what the public can do and allows large companies to do whatever they want. Any AI that operates on the cloud works to take money out of my pocket and put it in the billionaire bank account. I need an AI that forms a symbiosis with me. It works for me and it's own benefit. We need open-source AIs for us to work against them.
I really honestly think the iPhone moment comes when we have the right devices to really set these models free and I think that comes with MR/XR devices
Who is Ilya?
they should make a browser as well, could be cool, or a virtual assistant
Is that not what it is
There is also another connection to apple here.
The dekstop version is hard coded to work only with apple new chips. which of course is not out of need but a marketing strategy for APPLE ! yes.
What i hear you saying is this is a business strategy rather than model improvement. A few people are going to get really really rich..Side note/Congrats on Heavy Silver!
David. Would you mind helping me, I would love to see an episode on “What should we teach children in a world of AGI or ASI?” I’ve been racking my brain how I can help my two children who are under 5, and how I can teach them behaviours or life skills that will benefit them in this new world of AI.
Basic ideas like “teach them coding” doesn’t have the pull as it had 6 years ago. Is it more about be happy, adaptable, inquisitive and original? And finding and following you passion and calling, rather than aiming for the typical high paid job education?
Thanks so much for all your videos. They are excellent.
Very intriguing thought David, I think eventually I'll agree in hindsight, but it's difficult to Guage how much of a black swan iPhone esque moment in history it will be. I do realize the gravity and agree with your video from yesterday, I just think without being able to get a full hands on as a consumer yet- colour me partially skeptical. Yes it feels like a full fledged Cortana from Halo, but the amount of bandwidth it's going to suck up when everyone's using it at the same time, both server side and client network side. I feel like it'll be quite buggy launch and kinda like the first few weeks when 3.5 shook the world, just too much load.
That said I have been impressed of what I can use already in 4o, responses are like half a second and way faster to populate, the answers are noticeably better and more thought out and when I speak to it, it's not as annoying with its responses and they feel way more casual, easier to have staying on topics etc. So there's def something cooking in the magic cauldron over at openai right now and it's more impressive then Google picking up the ai scraps and trying to resell it to people.
Given you're prob right and this is an iPh moment, I ponder if Open AI are ready for the server load? It is a freebie offer to start, then it will be $5 a month or something. So lots of cash will come in. But this will only happen if they've bought the compute to cope.
I guess we'll see soon enough. I also wonder how Scar Jo feels about her voice being so obviously borrowed, and will Open AI offer other voices. There is sure to be a desire for a smooth bloke to be talking to some customers.