# Huge AI NEWS : OpenAI Gives GPT-5 To US Government, Stunning New Humanoid Robot,  Metas New Model..

## Метаданные

- **Канал:** TheAIGRID
- **YouTube:** https://www.youtube.com/watch?v=PivqsoRDysA
- **Дата:** 03.08.2024
- **Длительность:** 16:33
- **Просмотры:** 82,096
- **Источник:** https://ekstraktznaniy.ru/video/14152

## Описание

Prepare for AGI with me - https://www.skool.com/postagiprepardness 
🐤 Follow Me on Twitter https://twitter.com/TheAiGrid
🌐 Checkout My website - https://theaigrid.com/

00:10 New Humanoid
06:06 Metas New Model
08:17 GPT-5 To US Government
11:07 Gpt4o Voice Mode
12:33 Gpt4o Long Output
13:28 Ultra Realistic Image Model 


Links From Todays Video:
https://x.com/skirano/status/1819444885943914688 
https://blackforestlabs.ai/ 
https://techcrunch.com/2024/07/31/openai-pledges-to-give-u-s-ai-safety-institute-early-access-to-its-next-model/?guccounter=1&guce_referrer=aHR0cHM6Ly93d3cuZ29vZ2xlLmNvbS8&guce_referrer_sig=AQAAAM0HTnrWjvNzyN1DevjXctPXyW1ncHHS8XUiZVSrCiT9AlWw6LPeF2BXaA58nf__Xb6lFg02qF6o1n6WN6lstXt4djHt9PcM-laDCa-r8QedfGw8j2cZXjdaV-TBGJ_e3dgOSfoPEq5GDt-gjHkKmjSnLrbq99v_BD9q4GhGd4CU 
https://openai.com/gpt-4o-long-output/ 
https://ai.meta.com/sam2/ 
https://x.com/adcock_brett/status/1819389188900610542 
https://x.com/AiBreakfast/status/1818630800113647948 


Welcome to my channel where

## Транскрипт

### New Humanoid [0:10]

coming in one of the first stories that I saw this was Brett Adcock this is the CEO of figure and he said only recently has time opened a window of opportunities to scale to billions of intelligent humanoid robots this was not feasible 5 years ago we are so fortunate to be in 2024 the first year in history when this is available now what he's referencing here is the figure product so I'll let them take it away and then I'm going to show you guys why this is crazy so there's a few things that I do actually want to talk about when talking about this robot because one of the main things that people haven't realized about this robot is that this robot is actually an electric upgrade and a recent industry Trend that we've seen is that humanoid robots are increasingly becoming you know and adopting Electric Power Systems over traditional hydraulic or pneumatic systems and this you know shift is essentially driven by several key factors that essentially make electric motors more advantageous for the development and deployment of humanoid robots now it was very hard to see but you can actually see that it does say electric humanoid right here so this you know marks this entrance into the you know electric humanoid space and this is rather fascinating because it was also Boston Dynamics that announced their recent Rob and in that video it did show us an electric humanoid so this is clearly an industrywide Trend and of course something interesting that we did see here was a much more upgraded hand that seemed to have this kind of you know increased grip area which kind of looked like the tactile sensing that we did get from Tesla's Optimus robot so this was one that I really did like here because although the fluidity of the last bot was nice this one also looks rather impressive too now Electric Systems tend to be you know less costly and easier to maintain more than hydraulic systems and you know the problem with hydraulic systems is that they're complex prone to leaks and require regular maintenance which can be expensive and timec consuming and in contrast we've got these you know electric humanoids which are cleaner simpler and generally have lower maintenance requirements and this makes them more practical for widespread use in various applications from industrial settings to home environments and one of the things you also have to think about really is the you know noise an environmental impact you know these robots you know if they're supposed to be in environment such as home environments and work environments the impact of the noise reduction is going to be rather important now for this robot this is going to be actually released in I think around 4 to 5 days depending on when I release this video should go out just after I'm recording this but you know seeing some of the comments from the founder Brett Adcock it does show us that this is very likely to be a remarkable engineering feat now the reason I think this is so incredible is because if we actually take a quick look back at Boston Dynamics robot this was the scene of their recent electric humanoid where we can see that it looks remarkably flexible in terms of how it's able to stand up and move around and navigate the environment so this is going to be a very interesting piece of future technology because I do wonder how it compares to Boston Dynamics recent Atlas robot that they're currently developing to which is you know the successor of their previous system which was you know loved by many and we saw that it was remarkable in terms of how human looking it was and it was able to do you know a remarkable level of things in terms of the flexibility and Agility of that robot now one of the key things that he said here is that this has been engineered for over a year figure 2 is the most advanced humanoid robot in on the planet and the thing about this company is that I got to be honest guys as someone who is watching the humanoid robot space I've not seen a single company other than figure make as much rapid progress as Brett adcook has in the short space of time that the company has been dedicated to its mission for I mean if we take a look at some of the other companies going on in the space it's honestly no shade to them but we have to really take a look at and really understand that this is remarkable speed in terms of managing to develop a Next Generation humanoid robot that is not only going to be ready for when AI system get remarkable upgrades but being engineered for over a year and already coming out with your second robotic demo that looks to be you know industry defining it truly is remarkable at what you can do when your company is in the hands of a determined entrepreneur you know it was only 4 months ago that we did get this robot where we saw that there was the speech to speech update which definitely you know shocked people because we realized truly at with our current systems with the current architecture with the current Hardware the kind of things that we can currently create and I think this is really good because you know people constantly say that you know this entire technology space is decelerating we're going through this trough of disillusionment yada y yada but I mean if you really do pay attention to exactly what's going on you can see that the technology here is increasingly getting more and more capable in terms of how good things are and I got to be honest guys looking at the entire humanoid space this robot was remarkably agile it was remarkably flexible being able to talk one of the first robots being able to do that this was all surprising so if they can do something that's even better than this I'm truly going to be impressed and you know surprised now meta are coming at us

### Metas New Model [6:06]

again with another open-source model release which is rather incredible they're introducing meta segment anything model 2 which is Sam now this is a model that is rather effective at segmenting any precise selection in any video or image this is really crazy because it has many different use cases I mean you could use this on a variety of different things but if you remember Sam 1 and the kind of things that people were able to do with that software I think this is going to be truly gamechanging and you can see right here that you know using Sam 2 you can select one or multiple objects in a video frame right here you can see the boot being selected and different objects and of course you can see here Sam 2 is capable of strong zero shot performance for objects images and videos not previously seen during model training enabling use in a wide range of real world applications now one of the most insane things about Sam was the fact that it has realtime interactivity and results it says Sam 2 is designed for efficient video processing with streaming inference to enable realtime interactive applications so imagine being able to track things in real time I mean this is going to be something that opens I mean I can't imagine the kinds of possibilities with this stuff because you're going to be able to you know track things in real time see where certain objects are I mean you could use this in you know augmented reality just a variety of many different things and of course you can try the demo yourself for example what you can do here it says you can you know create effects right here so if you click this try it out button click accept you know it manages to load up the demo and you can see right here you can pretty much track anything in this video so for example let's say I wanted to track the ball I could just click the ball right here and you can see that for the rest of the video if I just click track objects you can see right here that for the rest of the video it is going to go ahead and track exactly where that object is and it's pretty crazy I mean it's just you know completely remarkable at how AI systems are able to do this like literally being able to track someone's foot here is uh it's weird I mean it's definitely weird but I think you know it's interesting you know I think this kind of technology is going to be used by some creative people to do some really genius stuff

### GPT-5 To US Government [8:17]

now an interesting thing right here is that openi has pledged to give the United States AI safety Institute Early Access to GPT 5 its next model it says opening I CEO Sam mman says that open is working with the US AI safety Institute a federal government body that aims to assess and address the risks in AI platforms on agreement to provide Early Access to its next major generative AI model for safety test now if you're unaware of why this is such big news is because number one the next Frontier Model is supposed to be a model that is much more capable in terms of its reasoning abilities and abilities to produce factual responses now this is something that many people including people like samman have said that they're always going to need to test these models thoroughly before public release and it seems like this is probably going to be the first time that they're going to collaborate with an external government body to see exactly how well these AI systems truly are in terms of their safety and reliability now this is an interesting move because open AI has been under so much criticism recently because of all of the people in their departments just basically and I do want to say it's of course not everyone just a named few but there's some pretty prominent figures that have left open AI citing safety concerns and basically stating that look we're not getting enough compute to do AI research here and if I want to do meaningful research I'm going to have to go somewhere else like anthropic if I want to get it done and you can see right here in May open aai effectively disbanded a unit working on the problem of developing controls to prevent superintelligent AI systems from going rogue now I do wonder and this is just a Wonder pure speculation that are openi going to truly give them the most capable model because you have to understand that for openai they are a product company meaning that what openi do want to do here is they're in the business of making money and a lot of money now of course what they don't want to do is put themselves in a position where you know they're giving their models to the US government and then the US government's like hey uh maybe you can't release this it means that openi is not going to be able to you know release a product or whatever I mean you do have companies like anthropic Google and meta that are hot on the heels and right now currently I would argue that opening eyes GPT 40 is not the model leader when we've got models like Claude Sonic 3. 5 that are just so effective and I would say I do wonder if openi are actually going to stick to this if other companies manage to get ahead of them in the AI race as you know Google is really working hard on the next model and some rumors are basically stating that it's going to be better than GPT 5 so we will see if that is going to be true now there have been some incredible demos from the GPT 40 advanced mode and this one actually blew

### Gpt4o Voice Mode [11:07]

my mind essentially chat gbt advanced voice mode he asked it to count as fast as it can to 10 and then to 50 hey I want you to count from 1 to 10 really fast as you can 1 2 3 4 5 6 7 8 9 10 okay now even faster 1 279 okay now louder and faster and count up to 50 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 and I think this is a little bit uncanny as to how it managed to take a breath in the middle to pause I mean when you think about it it's a robot it doesn't need to take a breath it's not a human it's just mimicking what humans do which is I mean strange I mean this is so strange I mean I don't even you know have the words to describe the kind of technology that we're seeing and if this is as bad as things are I can only imagine how crazy AI is going to make certain Technologies and certain interactions with our current Tech I mean honestly this kind of thing just makes me wonder what the world is going to look like in terms of just the technology world for how we interact with AI systems that are lifelike and you know really truly humanlike because if you know someone was on the phone to me and I told them to count to 50 and then they did that I personally would not be able to tell

### Gpt4o Long Output [12:33]

that it's an AI now in some more open Aus they actually did release well not release unfortunately from open AI but they've spoke about GPT 40 long output so it says openi is offering an experimental version of GPT 40 with a maximum of 64,000 output tokens per request and we hope this experiment helps you explore new use cases that are unlocked by longer completions and Alpha participants can access GPT 4 oong output by using the GPT 4 o 64k output Alpha model name you can see the pricing right here and this is quite a small announcement so most people did miss this but it is rather fascinating to see how this works because one of the things that they did talk about on GPT 40 was its ability to just generate you know pure completions of tech so I'm wondering if this is just going to be able to generate entire essays from scratch entire books of course it's you know a lot more than you know 64 tokens but I do Wonder on how that coherence is going to compare to some of the other models that exist in the space now most

### Ultra Realistic Image Model [13:28]

people did miss this and I nearly did because the announcement didn't get that much coverage but I saw this and I was like I have to cover this because it's absolutely incredible so this is flux point1 a new era of creation and basically this is a new open- source image generation software that is unbelievably realistic now when you see images like this of course looks you know really realistic until you look at the badge and you think okay then maybe it was AI generated but the thing is that whil yes you know few years ago this would have blown any person's mind I still think that you know the kind of realism that we're looking at is here now so I'm going to show you guys a picture that it recently generated and you're going to say wow okay this is incredible so this was an image that was on their page and this is an AI generated image now I'm going to show you a few more in a moment but when we start to look at images like this that are iPhone style that look like they were captured from someone who literally just took them on their phone I think these are the kind of images that are the ones that are probably going to cause the most problems in terms of the trust levels for individuals actually believing you know what they see online because if I were to see this image and I was browsing on Twitter there would be no immediate red flag that this is AI generated at all I would look at this image and I would think that okay this looks you know immediately realistic but one of the things that you're about to see is that this model can actually generate completely new photo realistic images that are just iPhone only so you can see right here these are AI phot photos made with the flux. one pro boring Snapchat photo Circa 2015 I mean if I were to see this image on someone's like old Instagram I wouldn't think it was AI generated there's no obvious tell here like genuinely there is no obvious tell I mean I guess maybe in the back the car you know kind of merges with the house so I guess if you're you know analyzing every image with a microscope maybe it's something that you could see but even then you know most people on social media look at an image for you know maybe just 2 seconds and in those two seconds you have to kind of tell yourself exactly what's in the image you're like okay it's just a girl standing you know near some houses you're not thinking wait is it say I generated yada yada so this kind of image and this kind of software is probably going to bring more Credence to the dead internet Theory I mean if we take a look at this image right here this image you know just doesn't look AI generated at all of course if you read the text you'll see that you know it is AI generated of course and of course if we look at you know this image right here I mean this is probably the most shocking one of all because if you know I saw this picture like I said before there's no obvious tell that this is AI generated I mean maybe over time we're going to find some things like you know t-shirt patterns or whatever but you know you could always prompt the model to have a plain T-shirt with like just an apple on it that it would still get right so maybe if someone's t-shirt doesn't have text on it maybe then you could say it's AI generated I don't know it's a very strange area that we're moving into but let me know what you think about flux 1. 1 do you think this is good bad for the misinformation but I do think that this is the kind of you know photo realism that is you know really dangerous