we have so many topics to talk about today. There have been a number of announcements and innovations just in the last few days, so let's just get right into it. First, Mark Zuckerberg, during an interview, talked about the future of Meta AI glasses. And if you watch my channel, you already know I love my Meta Ray-Ban glasses. So where are we headed? Full interactive holograms. Let's watch this clip. Basically going to be three different products. There's going to be kind of the display-less glasses that just do AI and can capture content. And you can, you know, listen to audio books and music and take phone calls and all that. Then I think that there's going to be another step up where it may not be... It's not going to be full holographic display in the sense that it's not going to be like your full field of view as a hologram. Um, but I think you'll get maybe like a little bit of a heads-up display. And I think that that's going to create a bunch of interesting use cases too.

you'll be able to get notifications. You'll be able to message with people. You'll be able to um message with with AI um answer any questions and not just have it speak to you, but also be able to kind of see, which is is higher bandwidth. So that's going to be exciting. And then I think there's going to be the most premium version, which is kind of the full field of you, which is you know we're having this conversation in the future and like I'm a hologram sitting on your Liv on your living room couch or you're here with me and and it's not just a video call, it's not just like there's a screen and you're there as a hologram. I mean we can we'd be able to interact, right? So like you want to play cards, it's like okay, here we we got a deck of cards, it's a hologram, we're interacting. I think that's going to be pretty wild and that that still is where I think it's all going and it seems like

meta is not alone in this thinking. Apple recently was reported to have given up on Apple Vision Pro 2, and instead is focusing on a much cheaper, much smaller version, and obviously less capable. I bought the Apple Vision Pro, and I got to be honest, I barely use it now, but maybe I'll do a follow-up review of the Apple Vision Pro soon and explain why. All right, and sticking on the topic of meta, apparently the biggest version of llama 3 might be right around the corner. In fact, what's at beta users on Android have access to it. This is a screenshot, and you can see right here llama 3 70b default and llama 3 45b preview better for more complex prompts. So apparently there's been a number of people posting this kind of screenshot showing that they've had at least limited access to llama 3 40 5b. You know, when it drops, I'm going to be testing it extensively, and I really think llama 3 40 5b is going to get open source really close to frontier models.

but with 405 billion parameters, it might not fit on any local machines, even with high quantization. But we shall see. Next, the text-to-video platform Gen 3 by Runway ML is now publicly available to anyone who wants to use it. It is a quality text video, except where is Sora, but now we don't have to wait anymore. We can use Gen 3. Check out some of these awesome demo videos. [Music] [Music] So really happy to see Runway ML release Gen 3 Alpha to the public. We're going to see some incredible creations, and it really is only a matter of time until we have entire episodes of television, entire movies completely created by AI. And what I'm most excited about, which I just talked about in a previous news video, AI video games. The next topic I want to talk about, and something that is a bit foreign to me, is AI role-playing. And apparently, there are a ton of people who are completely addicted to character. A. Now I saw a few people post

about it on Twitter, and I'm trying to find news articles, but simply typing in character. An addiction Reddit, there are a bunch of different threads going, and this one is all the way from a year ago. Character AI is tons of fun, but can be scarely addictive. This app is causing addictions to people. I'm addicted to character AI. I feel terrible. Why do so many people genuinely get addicted to character. AI role-playing is just not something that has piqued my interest in the past, and it's maybe just not for me. Maybe I'm just not in that demographic. It tends to be younger, maybe teenagers, who are getting into this, but it does seem to be a real problem. So, is this social media 2.0? Is this another technological innovation that is going to harm younger people's mental health? It sure seems to be heading in that direction. Plus, and this is something I've talked about in the past, we have declining birth rates and loneliness epidemics throughout the entire world, and AI as a solution to loneliness in

the short term probably makes sense, and will probably help a lot of people, but in the long run, if people continue to get disconnected from other real human beings, this is only going to exacerbate the issue. So I'm going to keep my eye on this trend. I want to know what you think. Have you used character AI? If so, what do you use it for? Do you find it to be addictive? What do you think about this in general? Drop your comments below. Next, Beth Jizos posted this clip where he says models training on each other's data is like a human centipede effect. And I think what he's saying is basically you get derivative data, and then the models don't really become higher quality based on training on other similar quality models' data. And it kind of makes sense. And I've heard Sam Alman talk about rather than using more and more data, these models should be able to do more and more with the existing data. And then Elon Musk hinted at Grock 2. So, sadly, quite true. It takes

a lot of work to purge llms from the internet training data grock 2 which comes out in August so that's a reveal will be a giant Improvement in this regard now obviously x. A's Gro has a ton of unique data that nobody else has through the xplatform Twitter but it takes a tremendous effort to curate all of the garbage that is on x to something that is actually usable and high quality enough for a training set for artificial intelligence but one thing is we know it's now coming in August and yeah Elon Musk hasn't always delivered on his timelines but I'm excited nonetheless all right next and potentially what is the creepiest video that I've seen on the internet in a long time Brett Adcock posted about the University of Tokyo researchers developed a new technique to bind living human skin to robotic surfaces the technique involves using perforation type anchors inspired by human skin ligaments another step closer to more adaptable humanoids this is gross it is absolutely gross to me

don't know why now if this is what humanoid robots look like in the future. I say no, thank you. But this is maybe a baby step towards humanoid robots actually having human skin, kind of like Terminator. So with eyeballs and it being forced to smile, this is the thing of nightmares. Thank you to on-demand for sponsoring this portion of the video. On-demand is the FastTrack to deploying real-world AI applications. Let me tell you about it. First, they have bring your own model. So with on-demand, you can easily deploy your own custom models directly from Hugging Face, like Llama 3. This allows you to integrate the models that you know and love easily into your applications. They also offer bring your own inference. So if you have models that are hosted on an external service like Amazon SageMaker, you can easily integrate those into your applications as well, allowing you to use your preferred inference service. They also have a plug-in marketplace where you can explore a vast array of AI plugins to either enhance your projects or to share.

your own with the community. You can create, distribute, and monetize your own plugins through on-demand, fostering collaboration and innovation. Next, I want to talk about Playground. This is your testing ground where you can configure, debug, and see live outputs from your plugins and models. It is an invaluable tool for fine-tuning your AI applications before deploying them to production. You can export your generative AI applications into any programming language. And finally, Automations combine AI agents and plugins to manage complex workflows effortlessly, which means you can automate repetitive tasks so you can focus on what matters most: innovating and developing AI applications. So check out On-Demand. They have a suite of powerful features for AI development. If you're a developer, a researcher, an AI enthusiast like myself, On-Demand has something for you. Thanks again to On-Demand. Now back to the video. Next, Brett Adcock also gives an update about Figure. So news figure plus BMW Group's Spartanburg plant fully autonomous AI-driven vision model neural networks for all grass. More details below.

going to play this video. I'll play it sped up, and I'll cut some of it out, but it really shows some pretty cool progress for figures robot. Next, Andre Karpathy, who is one of the leading minds in artificial intelligence, posted this over the weekend: 100% fully software 2.0 computer. Just a single neural net and no classical software at all. Device inputs audio, video, touch, etc. directly feed into a neural net. The outputs of it directly display as audio, video on speaker screen. That's it. This is something that I have been talking about for a while now, and I've been talking about it as input to large language model direct to compute. And I really do think that is the future of the computing model. There is no need for software at that point. There is no need for applications at that point. And what I've also been saying is there's likely not going to be a need for developers at that point because if everything is done with the large language models, the only thing we really need is engineers who.

can control and kind of guide the large language models, but at a certain point, if you read situational awareness, we're going to have AGI that can actually do the research and guide the AI itself. So, I really do believe that this is the future of software development and deployment, and apparently Andre thinks so as well. So, someone commented, It would be even nicer if the neural net can self-improve learning, and then Andre responds with, In context learning is learning, then you bunch up things, and the next time your computer goes to sleep, it fine-tunes on it. So cool. Somebody asked, So it can't run doom? and Andre says, It can probably very close to simulate Doom if you ask nicely. And that's the point. You can ask it to do anything, and dynamically, it's going to create the quote-unquote software or at least a simulation of the software in that moment, and that does seem like the inevitable conclusion of software in general. All right, next lm.org relies route llm, which seems really cool and akin to agents with an orchestration.

layer on Top. This is an approach that I really like. Not all prompts require the highest level Frontier Model to answer. In fact, that's kind of what we saw with mixture of agents, and a lot of prompts can actually be handled by much smaller, much more efficient, and much less expensive models. So let's read what this says. Not all questions need GPT-4. We introduced route LLM, a routing framework based on human preference data that directs simple queries to a cheaper model. This is fantastic. With data augmentation techniques, route LLM achieves cost reductions of over 85% on MTY bench and 45% on MLU while maintaining 95% of GPT-4 performance. Fantastic. So a theme that I continue to hear from the likes of the biggest chip makers in the world is they really want to push the majority of AI compute down to Ed devices, so your phone or your computer. And of course, I love that. I love local models, and I love having the privacy, the security, the low latency.

that you can achieve by running everything locally, and now with things like route LLM, you can actually do even more of the compute on device cheaply, super efficiently, while still achieving GPU performance. And only when you absolutely need a large Frontier cloud-based model, do you just send it off to them. Comparing against commercial offerings, Martian and UniFi AI on Mt. Bench, and achieve the same performance while being over 40% cheaper. Our model data sets and code for serving and evaluating LLM routers are all open source. We are excited to see what the community will build on top. They released the code, they released a bunch of other information about it, and I plan on making a full video on this topic specifically. So that's it for today. If you enjoyed this video, please consider giving a like and subscribe, and I'll see you in the next one.

Robots with HUMAN Skin, LLaMA 3 405b, Grok 2, Gen3 Video, Figure Robot, Meta AI Glasses

Artificial Intelligence (AI), Open Source, Generative Art, AI Art, Futurism, ChatGPT, Large Language Models (LLM), Machine Learning, Technology, Coding, Tutorials, AI News, and more

** Exclusive Pinecone Partner **

Robots with HUMAN Skin, LLaMA 3 405b, Grok 2, Gen3 Video, Figure Robot, Meta AI Glasses

Scripts