# GPT-5 Previews, Googles Next BIG A.I, Military A.I , AI Devices, and SLOWING Down?

## Метаданные

- **Канал:** TheAIGRID
- **YouTube:** https://www.youtube.com/watch?v=f8I70yzBjjI
- **Дата:** 14.06.2024
- **Длительность:** 30:21
- **Просмотры:** 34,326
- **Источник:** https://ekstraktznaniy.ru/video/14248

## Описание

Learn A.I With me - https://www.skool.com/postagiprepardness 
🐤 Follow Me on Twitter https://twitter.com/TheAiGrid
🌐 Checkout My website - https://theaigrid.com/


Timestamps: 
00:00 Military A.I
04:35 Googles A.I
08:18 Microsoft A.I
12:18 Elons Supercomputer
16:27 AI Slowing Down?
22:45 AI Glasses
25:47 GPT-5 Previews
29:22 A.I Community Launch

https://interestingengineering.com/military/robot-dogs-better-sharpshooters-study
https://x.com/sandersaar/status/1798953968598757474
https://www.bbc.co.uk/news/articles/cd11rje1mrro
https://www.eetimes.com/data-centers-doubling-power-demand-seen-stressing-energy-grids/
https://x.com/electrik_dreams/status/1799349009523290608
https://x.com/Techweek_/status/1799058830695026934

Welcome to my channel where i bring you the latest breakthroughs in AI. From deep learning to robotics, i cover it all. My videos offer valuable insights and perspectives that will expand your knowledge and understanding of this rapidly evolving field. Be sure to subscri

## Транскрипт

### Military A.I []

so with one of the craziest stories in Ai and honestly this is the kind of part of AI that most people don't really like to talk about because this is something that we kind of think just isn't going to happen but as you know the real world is essentially very dangerous and this is something that you all should you know be aware of now this is actually an old story but it was uh recently made aware to me via a few posts on the internet and essentially a Chinese team of research have found that with the right design robot dogs with machine guns can rival human Marksman so the reason that this thing isn't that crazy is because well I say isn't that crazy it's actually much crazier than you think is because the video demo of this is just I don't want to say shocking but it did kind of shock me because as someone who has you know uh looked at AI in the past and you know over the last couple of years um when you actually truly see the raw brutal nature of what you know um I guess you could say incredible autonomous systems you know on mass could really do it's quite scary you know especially if you have an imagination that can kind of run wild like mine it is very scary so essentially what this is this is you know huge implications for war so basically um they they demonstrate the feasibility of a legged strike platform I'm guessing that you know the translation here isn't that good but this is just you know a robot dog I'm actually guessing that what we actually we have here is a unitary dog so I'm going to show you guys uh this quick video here um but it's pretty crazy I'm not going to like Play the full thing because I don't want there to be uh any you know uh blockades on like shock it's not really shocking content I mean it's on YouTube but the video is pretty crazy because it showcases this robot dog and it's incredible like genuinely this is something that shocked me because I could see a world in a few years time where you know you have these you know humanoid robot dogs roaming certain sites what ever uh and carrying out certain actions that depending on which side of the you know Battlefield you're on it's either going to be a great thing or it's bad thing so this right here like it had remarkable accuracy now the accuracy I'm guessing you know right now is probably like 80 to 60% whatsoever they said it rivaled uh Marksman which is you know it's surprising because if we take something like this and then we look at the pace of AI Development Across the board and we can see that you know um across the board like not just generative Ai llms and all that stuff I'm talking about like across the entire field like when I was you know truly in the depths of AI like I'd be looking at you know like so many research papers every day like I would be eating my food and just going through the arive website and looking at autonomous drone storms you know uh just with a bunch of other people in Discord and just looking at the small incremental changes and this kind of thing here guys I can promise you that this is going to get a lot better in the future so one of the things that I'm thinking is that when is it going to be that we truly um you know change from I guess a Warfare that is one that is you know human to one that is autonomous and I think that you know increasingly in the future we're going to start to see more and more autonomous Warfare because it's going to be something that's far more effective I'm not sure how much a US solda cost um you know I personally don't think there's any value you know that you can truly place on a human life I think they are completely Priceless you shouldn't be like oh one person is worth you know $10,000 or whatever yearly salary at cost but I think that you know if you have you know a system like this that could simply you know run through a building and simply quickly annihilate and you know see Target I think it's going to have remarkable implications for Warfare so um this is going to be something that is uh pretty crazy we knew that this was going to be the case we also saw the implications of AI in dog fighting that was another area which was recently crazy but the reason I wanted to include this is because this kind of FS this Falls outside of the scope in which most people look at AI in right now because the spotlight is on of course open Ai and of course how you know AGI is coming and you know generative Ai and how models can reason and learn and those kind of things so I think this kind of thing uh whilst it's you know not the best kind of news it's still important to know that like you know the world is changing in a very interesting place and AI is really going to just dominate certain parts of the industries that we really didn't even think about so then we did actually get a update from Google this wasn't really like an update it was more so uh a pretty interesting interview with

### Googles A.I [4:35]

Google's Deep Mind CEO and this was where he actually talk about with a 2 million context window people can upload books the size of War and Peace or one long hour videos and expect accurate intelligent interactions with it and he actually talk about so you know some of the future developments undergoing work on their new project Astra so they're working on ingesting live streaming videos that will work as memory aid for the elderly that's going to help in Practical everyday tasks and I think this uh you know snippet actually does show us a very interesting take on how Google intends to use AI because one of the things that we've seen recently is that you know people like Demis sais and companies like Google wi whilst they might not have the best technology and I say that because there is a small margin in terms of you know state-of-the-art versus the next best tech technology of course that technology is still incredible I think that what people can see is that Google's genuine goal and intention is of course to help the world with their AGI level systems whereas The Narrative from open AI has shifted to one where many people are questioning what they're going to do with that power so take a look at the future of their AI assistance because I think it's rather important is uh this million token long context actually extended it to 2 million tokens now and maybe we can go a lot further than that and that means uh you can now ingest you know all of War and Peace or you know even like an hour long video and then ask questions of the system about it and it can answer back intelligently and that opens up whole new possibilities for you know as an analysis tool or sort of research assistant um but also for uh then with Astra are streaming live video um obviously and then doing things like remembering I think the thing you're referring to is like remembering something that happened a long time ago ambiently in the video stream uh and it could be you know super useful I think for you know an elderly person to remember where their medicine is or you know where your car keys are or something like that so I just think there we've barely scratched the surface of what I can imagine these things will be used for and I completely agree with that we have honestly scratched the surface on how transformative this technology will be which you know begs the question how much is AI going to change our world and some of the people on these Frontier Labs I truly do wonder what goes through their mind sometimes because you know even in this interview he does talk about how like literally people have lit literally I know I say literally three times but the point is I'm trying to drive home the point that it's very hard to conceptualize to the average person who isn't keeping up with the pace of development on AI how crazy this technology will be because whilst yes some technologies aren't you know ready for certain applications I think the monument mental change is going to be when people you know it affects their day-to-day lives like when people saw the chat GPT demo you know demos of gpt2 and gpt3 had existed long before that so people who you know saw gpt2 and gpt3 they essentially weren't really surprised but with gbt 3. 5 you know just placing that in the public eye in the public view that was something that truly did surprise a lot of people so I think you know this entire interview I'm going to give you guys the link to that cuz I think it is pretty cool but it is something that you know goes to show that truly this isn't some kind of Hye kind of conference or anything but the true nature of AI is going to be felt worldwide very soon so one of the things in the AI industry that actually has a lot of people concerned about their security is Microsoft recall feature so if you don't

### Microsoft A.I [8:18]

know basically Microsoft had this feature where as you can see right now I'm recording my screen and I'm doing whatever if I wanted to I guess you could say remember what I was doing essentially I could just go into the recall tab bar and I could search exactly what I was doing and essentially you have an on device thing that basically records your entire computer screen but using generative AI you can find a way to basically search through the entire recording to find out exactly what you did in order to recall information so essentially Microsoft unfortunately had to backtrack this feature I guess you could say in a sense but I'm going to make sense of some of the details so the article says that Microsoft is making changes to a controversial feature announced for a new range of PCS powered by artificial intelligence as it was flagged as a potential privacy nightmare the company buildt the recall feature for co-pilot plus as a way to make users lives Easier by capturing and storing screenshots of their desktop activity but of course you know people claimed that hackers might be able to misuse the tool and it saved screenshots making the feature opt in instead so previously uh the feature was on by default which is pretty insane if you think about it because it's like you're forcing people to do something which is not really against their will but you're not really making it obvious to them exactly what's going on like I can imagine a bunch of people you know having this installed on the device and they not essentially not really taking advantage of it because a lot of the times they just have so many pointless features so putting this on their device especially when there have been so many privacy issues in the past is pretty pretty difficult so essentially what they're just doing right now is they're basically just stating that they're making it easier for people to choose to enable recoil on their co-pilot PC and improve privacy and security safeguards because I'm not going to lie to you guys right now this is something that's in its early phases and it would be awful if this kind of thing proves that it could be hacked and someone could use this to you know find out whatever banking information you have or find out what kind of things you get up to online either way it doesn't matter what you're doing on a computer your right to your privacy is something that you should have and I think this does kind of threaten it a little bit if they were just going to implement this natively on the system without making you know initially be able to opt in instead of opting out so I think this is a good thing because you know you don't really want this trend of people just you know embedding generative AI into everything because right now that is of course what it does feel like but it's important to remember that you know you do have to remember that security is a big thing because a lot of the times there were even like leaks from chat gbt of people's personal sensitive data by simply putting in a basic prompt and it would literally output someone's name their address their email their phone number it was pretty crazy chat gbt open AI they managed to patch it pretty quickly but it's important to you know roll out these features slowly because if you don't test them uh things can go pretty wrong and I do think that one of the features that recoil had was it's going to be able to see exactly what's going on in your computer and to be able to tell you know exactly what you should do because there was even a Minecraft demo which was remarkably fascinating on how it could navigate and understand what was going on in the world and provide real time advice to the player not sure if it's going to be used by a lot of people but I'm sure in some scenarios it's going to be very applicable so now in a bit of Elon Musk news Elon Musk is actually planning the world's largest supercomputer in Memphis now this is a piece of news that isn't as big but it's we knew because we knew he was building a super computer but we didn't know he was building the world's largest supercomputer and so the reason that this story is so crazy is because I think the world is entering a new phase where we're starting to see

### Elons Supercomputer [12:18]

these huge supercomputers start to you know become developed and start to go online of course at least earliest next year when we're going to see these Next Level Foundation models actually get built and we're going to be able to see a lot more generative Ai and a lot more different AI systems you know trained quicker because if you think about it right now one of the biggest constraints that we do have is the limited compute think about it like this when you are training a model and you're trying to train an even bigger model you basically don't even have compute for a lot of other things we initially saw this with open a eyes GPT 4 the problem was is that they didn't even have enough compute for inference people were only allowed to send 25 messages to a powerful AI system is GPT 5 going to be the same open a ey they're trying to build an 100 billion computer called project Stargate and Elon Musk is building the world's largest super computer in Memphis now what's crazy about this is that he actually said that he wants to get this done by 2025 and I'm guessing that he's going to be using a lot of the recent Investments whether it be from Tesla from x. to be buying tons and tons of Nvidia gpus in order to ensure that this data center is going to be powering the next level of whatever AI infrastructure he needs now of course some of it's going to be for his exto AI company which I do think is probably going to make some significant strides in the coming months and years in regards to AI because they seemingly have been doing very well in whatever Vector they've been chasing and I think that as well with the fact that they've already got the huge infrastructure with Tesla I think they're going to be able to combine those forces to come up with some very fascinating things regarding embodied AGI so this kind of project to which Elon Musk says should be online by 2025 of course Elon Musk and his dates do have you know not the best trap record in terms of being on time but I do think that kind of factory is going to need an enormous amount of power to which the question is you know how are we going to do this and I've seen you know a lot of politicians start to even talk about nuclear energy being one of these main fuel sources and you can even see here that data centers doubling power demand seen stressing energy grids so essentially electricity demand from data centers Ai and cryptocurrency miners will surge by 2026 the Paris based International Energy agency has said in a January report after consuming 400 60 ter hours worldwide in 2022 it's going to be more than double in 2026 and basically what they're stating is that look these kind of data centers that are going to be building the next level of foundation models and these Le Frontier AI systems they are going to be truly demanding a huge amount of energy and that is you know I guess it's kind of a concern because you not only need power for the data center you also need cooling so there's all of these crazy things going on that just lead to not inefficiencies but just more demands overall in terms of the infrastructure requirements to run these kinds of giant level systems so we're probably going to see some large level construction things be taking place in the next couple of years as these supercomputers are being built and as these massive data centers to power all of this AI is going to be built and I think you know whichever company manages to build out the infrastructure the fastest and of course the most effective compute are going to be the leaders in generative AI because as we know one of the main bottlenecks to AI is of course energy SL the data sensor problem so it's going to be interesting to see how the actual economy responds to this if they're going to be welcoming this if the energy demands can even be met for these data centers and how the entire problem is navigated now you can see here this is a tweet by Yan lean and I think this is a tweet in response not I think I know

### AI Slowing Down? [16:27]

this is a tweet and response to this but this is something that I do want to talk about because I would argue that this is one of the biggest debated things in AI right now because this is one of the biggest Paradigm shifts that we could potentially have before artificial super intelligence and basically the biggest debate that's going on right now in AI is whether or not we are at this inflection point that is going to lead to you know logistic growth or exponential growth so essentially this guy says right here and he's on the open-endedness team at Google deep mind he says like everyone else I am extremely excited and optimistic about AI progress but since there seems to be the misconception that it is really easy to make predictions about the future for exponential Trends let's just draw lines on a log scale this is a reminder that while observing an exponential trend for example compute or neuron Network parameter increases we can never be fully certain what we are observing is in fact following a logistic Trend before the inflection point both Trends look the same so basically what he's stating here is that look before we get to this inflection point and I should Mark this with this right here you can see that if you just didn't have this like this was completely white and we're only looking underneath right here we can see that both of these follow the same pattern they both decide to go up just like this so to the naked eye to anyone you know if you're watching a graph go up like that you're thinking ah you know the growth is probably going to be exponential however there are usually s curv graphs which do happen and this is where the things get mature and things start to taper off which is this logistic kind of growth right here however of course there are some cases where the exponential growth just keeps on increasing and it just goes up in a straight line but the problem is that currently many individuals including myself find it quite hard to predict where on Earth this inflection point is going to be and of course where that growth might start to taper off now I personally do think that this inflection point is going to be driven by one or two factors one of them being the things that we previously discussed number one being of course compute capacity of course as you know powering these data centers is no easy feat you powering these dat caters and a lot of compute to do so that is going to be a proper problem so essentially the energy SL compute problem is basically one of the biggest problems that we're going to have and of course I think the safety issu slash the kind of issues upon regulation to where these models start getting bigger and better and then of course there might be more and more regulation around what things can kind of happen so if we start to run out of energy SL compute in order to train bigger models I do think that you know of course there are literal physical implications and physical inflection points that are going to drive the logistic curve but of course there is the you know slow not slow but like hidden theory of you know once we get to the point of automated AI research that's when we're going to get you know exponential growth grow because if the AI can do a type of research to where it can make itself much more efficient and much more effective at solving problems then growth is likely to be exponential because you could duplicate a th a researchers leave them running 24/7 in a loop where they just publish AI research and just continue to test it and iterate and improve upon that we could have you know a decade of AI research comprised into just 12 months so I think that what is likely to happen is that before we get you know these factories these you know huge huge data centers online I think the growth might taper off around gp5 or GPT 6 but once that compute is built I think then it's going to be basically like this where we have a small inflection point things taper off and then we go to another one again so I think overall that is likely what we're likely to have overall because I think that whilst yes this big Trend does happen in a lot of Technologies you know you've got the garden hype cycle a variety of many different things I think that what we're basically seeing now is something like this to where you know you have like gbt 3. 5 then of course you have gbt 4 with a lot of the other Foundation models there and then GPT 5 is apparently going to be the same kind of jump anyways so we're then going to have that slow uh you know growth and then of course that taper off until these models get online I do think that once we do have these massive supercomputers online I do think that you know all it takes is automated AI research to be solved and then the graph just goes up maybe not indefinitely due to physical constraints of the actual world but the growth does start to look more exponential in that area but the problem is that this is very hard to judge without open ai's you know input because right now all AI research is essentially private and this is what yanon was basically saying he said you know I've made this point so many times the beginning of a sigmoid looks like an exponential not only can we never be fully certain to what extent we are observing isn't in fact following a logistic Trend before the inflection point we can always be certain that every single exponential Trend eventually passes an inflection point and saturates into a sigmoid so continuing an exponential Trend beyond that inflection point requires a paradigm shift basically you require a certain breakthrough if you want whatever curve it is that you're doing in order to go up to the next kind of sigmoid curve so that's what you kind of need because no physical process can grow indefinitely there's always you know friction in terms of the Dynamics equation that eventually become dominant energy consumption heat dissipation Quantum effects thermal fluctuations communication bandwidth mass energy density basically you're going to reach the physical limits of Earth even if you're trying to you know grow something indefinitely so even processes that appear exponential on a long time scale are actually a succession of sigmoids so what I just spoke about the sigmoids where you have one that is there and then you then it's kind of like followed by another and another that's basically built on the back of every other paradigm shift and it says a good example is Moos law which is get getting saturated right now but the exponential progress of the last seven decades is due to a succession of technological Paradigm shifts that weren't pre-ordained basically you know you didn't even think about them so each Paradigm behaved like a sigmoid each sigmoid overtook the previous one and then of course it seems exponential now

### AI Glasses [22:45]

whilst many people are dogging on rabbit or you know hating on Humane or whatever it is you know whichever kind of product is in the spotlight for being absolutely quote unquote awful I want to highlight this demo by Sanders saw on the brilliant AI glasses by brilliant Labs AR this is basically open- Source AI technology that is truly incredible I saw this on Twitter and honestly I'm surprised I didn't retweet this before cuz it's a remarkable demo that actually shows the usefulness of generative AI when combined with Hardware that actually makes sense to use I do think that this is where meta smart glasses are going in the future but I want you all to take a look at this because this is leveraging your phone connectivity and audio and actually you know some of the kind of generative AI apps that we use today and basically this is you know it's incredible I want to show you guys this because once you realize that you know augmented reality is going to be a thing you're going to start to realize that and I know that many people have seen this before that this form factor truly does work the future of AI enabled glasses they are the opposite of Vision Pro they're lightweight they look like normal glasses and they cost $349 they're also quite different from the AI pin which is trying to be a standalone device while the glass is leverage your phone for connectivity both of them have cameras and microphones and while the glasses have an AR heads up display this has a weird laser projection and more importantly the glasses are open-source community-driven platform hey Noah I'm making a video of you can you introduce yourself hey folks I'm Noah your AR smart glasses wizard I know stuff translate anything and use the coolest AI models like GPT and more to make your life a breeze where could I find a hat that goes well with this outfit close by you're right in Venice swing by gor and Bros on Abbott Kenny B viid trust me they'll have a hat that screams I belong in a movie what can I make for lunch you've got tomatoes onions maybe some apples like look at this okay like imagine just having glasses on your head 24/7 that you know I guess is basically now like an iPhone okay but you know if you don't have any problems it's literally just glass you know it doesn't like strain your eyes or anything but you could literally say hey what am I looking at on my screen hey what's in this fridge what can I make here hey is this out of date do I need to change this tire I mean I can truly see a future where this kind of application actually becomes really effective like especially since the text right here it doesn't look that invasive at all and it kind of you know seamlessly integrates now I think one of the main things as well is that with these glasses I think over time along with Hardware Innovations and software Innovations and along with AI systems that you know we can effectively run on device and you know essentially make the inference a lot faster I think these kind of products are going to be trly the future of AI so I think this demo you know it shows like exactly what we can do especially with project Astra and all those kinds of things I mean next 5 years can you just imagine the kinds of you know technology that we're going to have especially combined with AI I truly believe there's going to be you know an entire age of abundance and

### GPT-5 Previews [25:47]

this incredible period of time now the clip you're about to see is probably the most important clip in the video and I don't know why but I actually lost this video I'm glad I found it again and this is a clip of the Microsoft CTO talking about the next level in AI now pay attention because I'm going to break down exactly what he says and why this is so important and it's important because I'm just going to tell you why just watch all of the things fall in place to have uh like really durable memories uh with the systems uh like reasoning I think gets a lot better so you know some of the fragility in the current models are can't solve very complicated math problems and has to you know sort of bail out to other systems to you know do very complicated things I mean the way that maybe I would phrase it is if you think of gpte 4 and like that whole generation of models is things that can uh perform as well as a high school student on things like the AP exams um uh some of the early things that I'm seeing right now with the new models is like you know maybe this could be the thing that could pass your qualifying exams when you're a PhD student like that is incredible imagine an AI system that could literally do your PhD I mean the thing that makes that absolutely incredible is because that means that if the Microsoft CEO who not CEO CTO who actually has you know likely been in some GPT 5 early previews is seeing that these models are able to reason at a level which is much more advanced than prior systems it likely means that we are nearing potentially an automated AI research and if he's stating that you know you're basically you know doing your exams as you're a PhD student that means that we're not far off from further iteration Cycles where we do get to that level where AI automated research is you know solved and the reason I talk about automated AI research so much is because that is the only thing that you need to automate in order to have you know pretty much exponential growth in AI because once AI can do you know the AI research them um we can pretty much exponentially scale that and then of course compound those breakthroughs on top of another which over time just gets you this truly compounding growth period so I think this is pretty crazy because like I said gbt 4 which many people use on a day-to-day basis is quote unquote a smart high schooler so what do you think the vibe shift is going to be in the general public consensus when they realize that holy crap we have access to a PhD level researcher in our pockets that we can use on a day-to-day basis to reason about many different things uh and it's just you know it's kind of mind-blowing okay I really don't even have much words to say about this but you know this is going to catch a lot of people off God but I will be ready because you know this is something that we've seen coming so if gb5 is incredible in terms of its reasoning abilities and able to you know truly understand and you know the multifaceted things it's able to do don't be surprised because clearly here you're stating that this is going to be on the level of a PhD student who's able to complete that exam all right let's Crank It Up you're about to Rock New York Tech week picture that crowd buzzing with excitement feeding off your energy you've got this bring that passion let them feel your Vibe and show them what you've got all right let's Crank It Up you're about to Rock New York Tech week them feel your Vibe and show them what you've got and what you just saw was a GPT 40 demo where open a eye showcase their new flagship model in New York Tech week it was pretty cool uh and it just goes to show that gbg4 is still pretty crazy in terms of what it can do so for 200,000 subscribers what I wanted

### A.I Community Launch [29:22]

to do was to relaunch my community and essentially this is post AI preparedness where you get exclusive access to one of my post AGI Frameworks the framework that I'm using to navigate the world post AGI and so far it's been really successful my personal strategy for making money with AI and exclusive tutorials on how to actually use Ai and of course invest in Ai and generate insane return so all of those things basically taking advantage of the AI Revolution this is currently there's not that many slots left for free members so if you do try and join now you can get it for free but of course it's going to be actually priced in at around $19 a month so if you're someone that's been supporting the channel for a while now this is your chance to you know I guess you could say learn AI with me because I'm going to be doing a lot of stuff here I'm going to be posting daily about pretty much everything that's going to be beneficial in terms of AI and I just wanted to make this quick announcement because I know some people might have missed this