# INCREDIBLE ChatGPT UPGRADE, NEW AI Robot's Walking, META'S New MULTIMODAL Ai [AINEWS#4] ## Метаданные - **Канал:** TheAIGRID - **YouTube:** https://www.youtube.com/watch?v=ZR9saRd1VII - **Дата:** 21.05.2023 - **Длительность:** 16:31 - **Просмотры:** 28,586 ## Описание AI Youtube Video Generator - https://vidiq.com/theaigrid/ Code Interpreter TikTok - https://twitter.com/JustinFineberg/status/1653748432434085888 Facebook image bind https://ai.facebook.com/blog/imagebind-six-modalities-binding-ai/ Tidybot - https://tidybot.cs.princeton.edu Chat GPT Stock Returns - https://twitter.com/dr_cintas/status/1660302839246123008 Claud 100k - https://twitter.com/dr_cintas/status/1660302839246123008 Sanctuary AI Robot - https://www.youtube.com/watch?v=k2GhgO7SnZQ Teslabot -https://www.youtube.com/watch?v=XiQkeWOFwmk Cyborg arms - https://www.youtube.com/watch?v=YErjSfsxwYo&pp=ygUFSml6YWk%3D MIT ROBOT - https://twitter.com/MIT_CSAIL/status/1657415488698187777 Welcome to our channel where we bring you the latest breakthroughs in AI. From deep learning to robotics, we cover it all. Our videos offer valuable insights and perspectives that will expand your knowledge and understanding of this rapidly evolving field. Be sure to subscribe and stay updated on our latest videos. Was there anything we missed? (For Business Enquiries) contact@theaigrid.com #LLM #Largelanguagemodel #chatgpt #AI #ArtificialIntelligence #MachineLearning #DeepLearning #NeuralNetworks #Robotics #DataScience #IntelligentSystems #Automation #TechInnovation ## Содержание ### [0:00](https://www.youtube.com/watch?v=ZR9saRd1VII) Intro with a Monumental weak and artificial intelligence let me quickly cover 15 of the most insane things you most likely missed so one of the first things that we saw was Chachi Beauty now trading so it says a chat TPT trading algorithm delivered 500 returns in the stock market the University of Florida study revealed a chat TBT achieved a staggering 500 returns in one advancing model this outpaces conventional sentiment analysis models used by hedge funds and apparently this is going to be revolutionary now here is the chart that you can see where you can see the cumulative returns of investing one dollars without the transaction costs in many different theories now essentially if you're wondering how did this work what did they do they essentially gave check GPT sentiment analysis which essentially just means that they gave chat TBT articles headlines news tweets and all that data they plugged into chat TPT to get a reading of how the market was feeling and then they trade based on that so the long story short is if everyone was sort of bearish and everyone was saying that oh no there price is going to go down maybe check TBT initiated to buy and when the stock was really high and everyone was saying oh my God this is euphoric that's when chat GPT could have initiated a cell that's just a basic overview of how that could have worked but I have one question to pose to you guys what happens when Chaturbate can predict the stock market with 99 accuracy does it break the financial industry well that's a question for another time we have ### [1:20](https://www.youtube.com/watch?v=ZR9saRd1VII&t=80s) Tidybot tidybot and honestly think that this one has a lot of applications in terms of being used in the real world and you're going to want to see this because I could actually imagine a world where these robots that this company has worked on this research is going to be literally in every household so let's take a look at the abstract then I'm going to show you the video of the robots that they talked about so this is actually really cool okay so they said for robot to personalize physical assistance effectively it must learn user preferences that can be generally applied to Future scenarios in this work we investigate personalization of household cleanup with the robot that can tidy up rooms by picking up objects and putting them away a key challenge is determining the proper place to put each object as people's preferences can vary greatly depending on Personal Taste or cultural background for instance one person may prefer storing shirts in the drawer while another person may prefer them on the Shelf so we aim to build systems that can learn such preferences from a handful of experiences via prior interactions with that particular persons so we showed that robots can combine language-based planning and perception with the few short summarization capabilities of large language models to infer generalized user preferences that are broadly applicable to Future interactions so this approach enables fast adaptation and 91. 2 accuracy on unseen objects in our Benchmark data set and we also demonstrate a real world approach where tidybot has successfully puts away 85 percent of objects in real world test scenarios now I know you might think this isn't that impressive but the reason something like this is going to be so impressive is because when you have unseen objects this is still able to get it right around 85 to 92 percent of the time which is honestly quite impressive and means that it's not going to have to be continually trained and will honestly know what to do so I think stuff like this has real world applications I mean I can imagine something like this going around after a stadium event just literally cleaning up or you know perhaps maybe a big party or something maybe in you know a rich person's house or something but um yeah it's definitely going to be interesting to see if you know this is going to be cleaning up a warehouse or something but um yeah definitely check out the paper because it is really cool and there's a lot more videos on this that I just don't have time to cover at this moment we had something remarkable and I was so happy when this was announced because this is a final Benchmark that can now be broken okay so anthropic a ### [3:35](https://www.youtube.com/watch?v=ZR9saRd1VII&t=215s) anthropic company that is funded by Google essentially managed to get their large language model to essentially have 100 000 tokens in a context window so if you're wondering how different is that to chat gbt chat tpts is 32 000 Max which essentially means you can now put in a 82 page PDF and have it analyze every single piece of information from that document this isn't even available in chat TPT just yet and Claude already has this and honestly I was meant to make a video about this where you can literally use their API tool and actually do have access to it and I haven't actually tried it out yet but it is something that is going to literally change the game which means that eventually we're going to get AIS that can literally write one thousand two thousand three thousand four thousand even ten thousand long word pieces of text just immediately from one single prompt so that's going to be interesting and of course we're now going to have less instances of that conversation where the AI actually hallucinates and forgets what it was talking about so this honestly might not seem like a big step but honestly like I said I've been going on about this paper for so long the scaling with 1 million tokens paper and many people didn't understand what I mean but essentially you just now have more memory in that and you can essentially extract more data from that specific tool so with that being said that piece of information right there is pretty insane and the reason this is also so good is it means now that chat gbt is going to be forced to roll out their 32 000 token window much sooner because they know that the competition is ramping up then of course you can see ### [5:08](https://www.youtube.com/watch?v=ZR9saRd1VII&t=308s) Epic Games that they actually managed to feed this large language model The Great Gatsby which is a very long book and they essentially changed one line ask it which line was changed and I was able to get the information in 22 seconds that is going to have game-changing implications in the next couple of years then we had epic games released their machine learning the former sample for Unreal Engine 5. 2 which demonstrates how the engine's machine learning technology can help create a character with deformations driven by full muscle flesh and cloth simulation so essentially without the technical jargon basically we're getting much more realistic visual basically without the technical jargon we're going to be getting much more realistic video games with a lot less processing power so this is going to be super duper interesting because AI is affecting everything now what was also ### [5:54](https://www.youtube.com/watch?v=ZR9saRd1VII&t=354s) Nvidia Research cool was we had Nvidia release a ton of research papers around 20 different research papers that all were involving Ai and this one right here that you're seeing right now is essentially where they got data simply just from a video and we're able to map that onto a 3D character and essentially what you're seeing now is a tennis character and they're able to get that data just from getting video files and having AI essentially analyze that and then map that onto a physical character now what you're seeing on screen now is mocap which is traditionally quite expensive and something that many companies can't even afford to use so if Nvidia is able to streamline this technology very quickly and apply it to many different 3D softwares they're going to have a huge part of the AI Pi in the future not that they already have the large AI Pi with all the everyone with everyone using their gpus and stuff but it's definitely going to be really interesting to see how this technology progresses because they're literally able to get this from real-time World data and I think that's also going to affect games and other interactive software then of course we had AI being able to essentially compress textures that are usually quite high in file size so essentially what we're having here is of course as you know in order to get a video game to run you need graphics and those Graphics can sometimes be quite demanding so essentially what they did was they used AI to compress these graphics and have it at the same file size but four times the quality which means you're going to get a higher quality game running on the same exact system so essentially they're just using AI to essentially make things much more efficient and it was proved to be very effective then essentially what we ### [7:28](https://www.youtube.com/watch?v=ZR9saRd1VII&t=448s) RealTime Rendering had from a video was them actually getting data from a single image so essentially they had one single image but they were able to get accurate depth details and able to essentially have a 3D model of someone's face from one simple image and this is something that was done in live as well there was no you know long term render this was real time rendering which is going to show the complete applications of this in many different softwares so honestly this was something that is going to make video calls and stuff like that much more interesting and much more lifelike because I also saw I guess there was in Google's IO event there was a little bit of application in there if you know I would have the clip but it is a very small clip that uh there's no ### [8:11](https://www.youtube.com/watch?v=ZR9saRd1VII&t=491s) Image Generation point showing it but um you can see the results right here from this um and this just goes to show you there's literally one image and then you get the output image of a accurate 3D model so this is definitely something that is gonna have a lot more implications on how we process data and what kind of data can be extracted from an image but I did release an entire video on this if you want to go see it is in the link in the description then of course we had them being able to generate images from a text but this was personalized image generation so all you needed to do is submit one image then you can simply change how that image absolutely looks you can add snow on it you can add a hat on a dog there were many different things on this and I have tons and tons of examples in a video Linked In the description and video honestly once they release this tour I think this could I wouldn't say just throw in mid Journey but I'm pretty sure mid-journey is going to now have to work on something like this because this tool was far better than anything I've ever seen when it comes to image generation in terms of the personalization so this was something that was really cool ### [9:11](https://www.youtube.com/watch?v=ZR9saRd1VII&t=551s) AI Content Generator then of course we had this insane AI content generator for YouTube to be released and honestly I think this is insane because this is actually what I used to help me in my video creation process which has allowed me to grow the channel very quickly so essentially let's say for example you know that some AI content was released in Google's recent event I can put okay I want to make a video about Google's IO event in 2023 so I'm just going to click go and boom I get a ton of information instantly you can see it gives me the title keywords that I can put in my description it gives me the actual description it's then going to give me the entire video script which I want and of course I can go into chat activity I can literally just click expand and it's going to give me more data on that and of course it's also going to give me the voiceover it's going to quickly get an AI voice over that I could literally just import into my software and then of course it could give me some of these thumbnails of course we're not going to use but I know how to design thumbnails so that is pretty cool now this is going to be something that is going to change the game because I'm sure that in the future they're going to add maybe some video creation stuff and if you do want to try this out this is actually not free but it is only a dollar so I mean yeah usually the pricing of this AI stuff is actually around 39 a month or 79 a month the free tier is like 7. 50 and free ones aren't actually that great these are the ones that you want to use but of course as you know it's pretty expensive so because I'm an AI Channel they actually managed to give me an exclusive deal where you guys can get actually access to this for only a dollar which is actually really cool so I'd actually recommend you guys checking this out and try that out because uh even if you don't want to try out growing a channel it's definitely good for Content creation and literally just exploring at how you can literally grow with this so definitely try this out because it definitely helped me a ton and my channel has been absolutely growing faster than I ever imagined that literally saves hours of research you can see right here as well they do have an AI coach which is still in beta feature which actually does a lot of the work for you so it's definitely something that you should try out and ### [10:59](https://www.youtube.com/watch?v=ZR9saRd1VII&t=659s) Metas New Multimodal AI then of course we had meta introducing image by now I've got to be honest with you guys meta is severely underrated everyone's talking about chat GPT what Microsoft's doing and then everyone's talking about Google and Bard and meta are slowly cooking up an insane storm and I can guarantee you never heard about this and I'm gonna be releasing another video about it but essentially what we have here is image bind which is holistic AI learning across six modalities and the six modalities that you can see on screen right now are depth heat map text image to video audio and I don't know what that is I am you I'm not gonna like I do know what that is but I don't but what I can do is explain to you some of the parts of this because it actually really is intriguing so one thing that they actually did talk about was they actually talked about how when you actually have crossed mode retrieval essentially what you can get is you can get different sources from sources that don't exist so for example you can have audio to image generation where you have the penguin calls and then you have a generated image based on what that sounds like which is really interesting as you can see right here as well on the left we have image and audio and then you get a retrieved image so you get the image of a pigeon then you get of course the audio over merge revving and then you get someone driving by some pigeons which is really cool because this is I guess you could say a different kind of sense because you're combining the different modalities and then getting out a different output which actually is insane like I didn't really think about that because of course as you know if you heard a pigeon and then heard um you know a motor Revan probably instinctively have that image in your head but to get that retrieved image just based on um you know two different outputs is really interesting so I think this is going to have a range of different applications and you have to remember that meta are constantly releasing new stuff and the only reason it doesn't really get covered is because that stuff isn't enough to make a whole video about but when you combine those small parts um it's really going to be really crazy so uh yeah meta's doing some crazy stuff when it comes to AI then essentially we had MIT researchers that taught a robot dog to perceive a 3D World using a neural volumetric memory so this technique allows the bot to climb stairs step over the gaps and run autonomously with a single neural network the great thing about this approach is that it actually makes the robot dog smarter and more efficient so it can make decisions faster because it combines the perception of the world its memory and its ability to control movements all into one system so this was something that caught everyone by surprise and this was and I will make a video about this the first human cyborg AI that is essentially a complete robot so as you know many companies are actually working on theirs as you know Tesla bot is currently working on theirs open AI is actually working on theirs but this company seems to have done it and this is the essentially first fully functioning humanoid robot that is powered by essentially their own AI system this is called Phoenix I'm not actually sure what the AI system is if it's a large language model if it's something multi-modal there wasn't that much discussion on this but of course there will be an end of the video talked about how this as you can see right here powered by carbon the AI system which has human-like intelligence now what will be interesting is to see if there are going to be some inreal person tests where you actually compare the mobility the dexterity all these kinds of features against the robots that are going to be coming out like Teslas and like open AIS in the summer so it's definitely going to be very interesting to see how they compare once against one another because right now we don't actually have any others and this is the first so very cool by Sanctuary then of course we had a major update to Tesla which actually has their Tesla Bots currently walking now of course this is currently quite slow but as you do know from the evolution of Boston Dynamics you do know that robots will actually move much faster as time goes on so you can see right here they're talking about how these robots are working on more torque control and how you're able to actually have such balance and precise movements where you know you're not even going to be able to crack an egg now many people are like okay this isn't you know that crazy but understand that there is environment Discovery and memorization what you're about to see is how it actually memorizes the entire environment which means it's going to be a lot better when it comes to navigating certain places so that is something that is going to be key in its ability to actually work in an environment and actually know what's there and know what's not there and then of course respond to changes what I also found interesting is the dexterity in terms of their arms you can see that they're actually able to perform well on certain tasks pick up certain things with definitely hand and human-like movement which is definitely something that a lot of these AIS do struggle with so seeing something like this is actually quite impressive and Tesla proves us as well now this was something that was uh caught me by surprise I mean this was something that was very incredible I mean a company a Japanese company essentially created cyborgs and we can see here that this is a essentially an addition to the human body and this works in tandem with how you want it to work and essentially you control four other hands that are on the back of your back now essentially this was created at the University of Tokyo a bunch of researchers decided to do this and essentially it does make human cyborgs now I'm not sure if they're planning on releasing this as a full-fledged product but it is an exploration of what could be to come now what's also cool here is that open air actually released some nice features for chat GPT such as browsing and such as plugins and this is essentially game changing because as you know one of the main limitations of chat GPT was of the fact that it only has data available until 2021 and with these plugins you can actually use certain applications and get maybe a booking you can get certain applications run honestly it's really effective so I would recommend you try this out just go into your settings go on beta features and then check to see if you do have access to web brows using a plugin because it's slowly being rolled out this week --- *Источник: https://ekstraktznaniy.ru/video/14844*