Elon Musk CHANGES AGI Deadline..Googles Stunning New AI TOOL, Realistic Text To Video, and More

There were actually so many different AI stories that I nearly wasn't able to keep up. But I'm going to show you guys some of the most important ones because these ones were pretty pivotal. So coming in at number one, we have Google DeepMind actually giving us a very fascinating update. You can see that Google DeepMind said that we're sharing progress on our video to audio generative technology. It can add silent clips that match the acoustics of the scene, accompany on screen action and more. And here are four examples. Basically what they're doing is if you have a video that doesn't have any sound effects, you can use Google's new model in order to add sound effects and a theme tune behind it. It's really, really cool. So what I will do right now is I'll show you a couple of the examples and then we'll do a deep dive onto why this is probably happening. Sam, Pretty fascinating about these examples is that we can actually see the prompts that they used as well. So it takes a combination of the prompt and then of course the video. You can see right here, prompt for audio, wolf howling at the moon. Then right here, which, I mean, if you just saw the wolf howling at the moon, that one was pretty, pretty nice. It actually sounded like a really, really high quality scene. So I do think that, you know, Google have, you know, shown us this demo in a way that makes me believe that this is pretty much near ready because the limitations that they speak about, they don't seem to be that crazy. And after all, this is just an audio model. Then of course we had this one. A SLOW MELLOW HARMONICA PLAYS As THE SUN GOES down on THE PRAIRIE I thought this one was really fascinating too because the sound of this one actually sounded realistic. The sound for the jellyfish pulsating underwater marine life, ocean didn't sound as great as many other things could be. But I do think that it was pretty decent in with regards to what it was able to do. And then of course this one right here, this is the one that I want to talk about the most because the prompt for the audio was a drummer on stage at a concert, surrounded by flashing lights and a cheering crowd. So with this one, the reason I liked this one so much is because it actually synced up to the hits of the beat. So I think that what Google have shown us here is that considering the fact that these output generations, if you didn't know these, were also AI generated. So a lot of the stuff that you're looking at here, like all of These generations, they are generated by Google's in house system, vo. If you aren't familiar with veo, basically around two weeks ago Google sort of shown us, okay, some more of their capabilities of their video model. And right here you can see the video model in action, being able to generate a drummer hitting the drum. And like I said, the most impressive thing is, is that it didn't just take poor representations of what was going on in the video, it managed to actually sync up the hits with the actual audio that the system gave it. So this is something that I think is remarkably, remarkably important. So do have an additional webpage right here where they explain how generating audio for video works. So essentially they talk about how it uses video pixels and the text prompts in order to generate rich soundtracks. And, and I think that this is where Google actually comes over the edge because I think Google is, you know, using systems that are a little more advanced than what we initially have. Most systems that we currently have today, they only really allow us to do text prompts and then you have to kind of generate as much as possible and then hopefully they sync up to what it is that is on screen. Now usually if it is just like an acoustic scene, like for example a peaceful meadow or, or a relaxing park where you're just hearing like crickets or something like that, those things don't really require much on screen syncing. But in certain scenarios, Google's process here actually comes in very, very useful. So I'm wondering what kind of creative process Google are going to be having because you can see that they've got VEO and they've got this and it seems that they're building the entire infrastructure for their multimodal AIs. Now if we are continuing to talk about Google, we need to talk about Google's major shift. Google recently have made a major shift from research lab to AI product factory. Now if you remember, you've been watching the channel. This is something that I spoke about for quite some time and I spoke about how Google has everything they need in order to take on companies like ChatGPT. But the problem is that they have a situation on their hands where they're trying to focus on safety and not push out any products. And the craziest thing about this all is that Google DeepMind have been losing consistently researchers because people wouldn't push features out. Now I made a 20 minute video on this explaining the details of all the reasons and all the failings that Google had. But if you didn't believe me, this is someone who used to work at Google that actually went to join Luma. You know how Luma just recently released their new AI dream machine that people can actually use for text to video. He says, now you know why I left Google to join Luma. I was in the team that developed VEO early on, but I knew it would never be shipped to the masses for quite some time. The same for Sora, not until a company like Luma forces their hand, that is at least I hope, give me access. So basically saying that, you know, I knew that Google weren't going to ship out anything for such a long time, so I decided to leave and of course join Luma. And remember, this isn't the first Google employee to do this. Many of them have left to join OpenAI, many of them have left to start their own startups. So Google has had this kind of brain drain problem where they're losing their best talent to other companies and of course people that want to start their own thing because they don't like how Google's handling this AI craze. Now essentially you can see here, it says over one week in mid May, two companies introduced AI products being built using one of Google's major breakthroughs. OpenAI, of course announced the new model that underpins ChatGPT. And of course Google announced AI overviews. But remember, the overviews didn't go well, which were pretty, pretty embarrassing. You can see here. This article continues to state the discontentment about pushing too hard on commercialization is a mirror image of the internal critique from the last two years when Google was struggling with bringing generative AI to consumers. Researchers who wanted to ship products departed for startups because they thought the company was moving too slowly, according to people familiar with the lab. Brain researchers have also mourned the loss of their brand and some even welcome the prospect of stronger leadership. Basically some of them thought that, look, the Google brand isn't what it used to be and maybe we need new leadership if someone can't navigate this AI industry. So overall, this article details a few things about how the company has combined its two AI labs to develop commercial services and a move that could undermine its long running strength in foundational research. So the problem is, is that Google is one of those labs that like, they are so good at doing research and making breakthroughs. Like I've already said, they made the breakthrough that powers ChatGPT, but they're thinking, do we now spend more of our time focusing on actually commercializing some of our breakthroughs rather than just experimenting in the research labs? That could potentially lead to more interesting things. Overall, it's a very, very hard thing. It's, you know, very, very hard to prioritize whatever it is at a company that, you know, could decide the fate of your business for the next 10 to 20 years. On one side, if you don't prioritize product, people get frustrated. They're like, okay, we're just gonna, you know, all flock to ChatGPT or whatever state of the art system there's going to be. But of course if you don't make incredible breakthroughs you can, you're never going to get to that next level of AI where you truly have outstanding products that stand the test of time and have an incredible moat. It's a very hard position for Google to be in, especially since they're a bigger company. And the problem is they don't have the newness to be able to make mistakes and for people to not critique them. If OpenAI kind of does something wrong, it's like, ah, they're kind of like a newer kid on the block. But Google's a behemoth, a company that's been around for years that has a pretty, pretty impressive reputation to uphold. So it will be interesting to see how Demis Asabis and Sundar Pichai managed to work together to get Google off the ground. I do hope that they manage to ship good products and keep a steady hold on their research initiatives because someone once said, and I don't remember who said it, but they said that if there's any company that should achieve AGI and, and that you probably trust with AGI, it's most certainly going to be Google. In a bit of strange News, we have TikTok actually introducing Symphony, their new creative AI suite. Now I don't usually cover AI social media tools, sometimes I do, but I wanted to cover a broad range of things that are going on in the AI industry. So basically Symphony is designed to elevate your content creation journey every step of the way. It basically blends human imagination with AI powered efficiency. Now this tool is basically an evolution of their Creative Assistant. It's an AI powered virtual assistant that basically just helps you make better videos overall. So you know how sometimes people look on social media, they'll try and find different trends, they'll be like, okay, what is going on here? What are the best practices? What kind of ideas should I be making? Well, this is a platform that actually uses generative AI to be able to analyze all of those things and then come up with something very effective so you can See right here, this is the thing in action. You can see you have your product, you have the description, then of course you have the media assets. Then of course you can see you're able to actually create TikToks that are AI driven in just seconds. You can see that once you manage to import your content, it actually allows you create AI generated content pretty quickly. Now this is relatively different because it's not creating AI generated content, it's synthesizing your content with AI in order to produce that content at mass scale. So I think this is probably the right approach because I think if you can get content out faster with the help of AI, that helps everyone who wouldn't want to save time. But of course people do not want AI slop on their timelines, which is why stuff like this, I guess it still works. And something that they do have as well is that they have translate for global reach, Tailor your message for audiences around the world by translating the script and dubbing the voiceover in multiple languages with just a few clicks. This is something that's going to bring down the global barrier. And of course they have this, a pre built AI avatar selection that allows you to use it for commercial use. So like stock photos, they provide quick, accessible, cost effective narration for brands bringing your product to life. Next, we had Meta release a huge amount of open models. It was so many that it's hard to even fathom what's going on at Meta. Now these aren't game changing models but, but they will change the community a lot of times. A lot of the community innovations that we see are built off the back of open source models. And this entire ecosystem of open source development is continuing to thrive thanks to Meta. Take a listen to what they had to say and then we'll dive into some of the more intricate details of what they released.

Today at Meta Fair we are sharing some of our latest research models and data sets. We've shared some of the papers for this work, but by sharing more of the artifacts we should enable the community to innovate faster and also develop new research. Now this is part of our decades long commitment to open science and sharing our work publicly. What's included in the release today? There's a multi token prediction model that can actually reason about multiple outputs at a time, enabling faster inference. There's Meta Chameleon, a model that reasons about images and text using an early fusion architecture. There's Meta Audio Seal, a new technique for watermarking audio segments. There's Meta Jazco, a technique for music Generation that allows better conditioning on chords and tempo. And there's Prism, a data set that enables better diversity from geographic and cultural features. There's a few more things as well. I really am excited to continue our work towards open research. I look forward to seeing what the community will build with these latest artifacts and of course look forward to sharing more with you over time.

So here you can see Meta sharing new research models and data sets from Meta Fair. This is pretty remarkable stuff. Meta seems to be one of the only companies backing the open source area for from a position of strength. Whilst yes, companies like Google have released things like Gemma 7 billion parameters and those smaller models, Meta are taking a clear stand. I think Meta's initiative is that they want to be the leaders in the open source community and it makes sense because what they've released are things that are truly, truly impressive. You can see right here they've released Meta Chameleon. It's a family of models that can combine text as images as input and and output, any combination of text and images with a single unified architecture for both encoding and decoding. So this is very, very fascinating. Take a look at this video that they had on their page because I think Meta is going to become one of the major players that sneak up on everyone, especially since they're supposed to release Llama 3.

Meta Chameleon is a unified multimodal model with joint modeling of text and images in one transformer. It's able to perform any combination of interleaved text and images in its input and output without the need for modality specific modules. Most current late fusion models use diffusion based learning for image tasks and tokenization for language tasks. Built upon an early fusion architecture, Meta Chameleon uses tokenization for text and images, making for a more unified approach. We believe this approach can scale better than late fusion or modular models while being easier to design and maintain.

So yeah, Meta have released some pretty interesting stuff. Audio Seal being the first audio watermarking technique specifically designed for the localization of detection of AI generated speech, making it actually possible to pinpoint AI generated segments within a longer audio snippet. So all of these things that they've released, like even the Prism data set, which actually increases the diversity of certain things. I think Meta is taking a different stance. You know, people have realized that you look didn't mean to do that. People have realized that look, OpenAI has the LLM area down. Meta. They're doing things that are truly innovative because of course they have v jepa. If you don't know that is a unique architecture that apparently might actually lead us to systems that truly understand what's going on. And of course, they have the open source area where people are going to be building upon it. And then of course, we're going to get innovations on top of that. Now, if you were living under a rock, you may have actually missed one of the most important announcements in Void Video generation, Runway introduced Gen 3 Alpha. So Gen 3 Alpha is the first of an upcoming series of models trained by Runway on a new infrastructure built for large scale multimodal training. Long story short, their text to video model is absolutely insane. Now, I did cover this earlier this week. It was truly, truly impressive. What we saw was something that was so amazing. And I think one of the key features is about Runway that you probably should be paying attention to is the fact that they have photorealistic humans. Everything about their system is really, really good. But I have to be honest with you, the photorealistic humans genuinely looks better than OpenAI Sora. And that isn't some sort of clickbait, some sort of, you know, exaggeration. I've looked at both videos now, and every time I look at the photorealistic humans that comes from Runway, I struggle to see any true issues with the quality of the content in terms of just the, you know, nature of how it looks like. It does look extremely realistic. And I still struggle to believe that what I'm looking at is truly text to video. So, yeah, it's pretty strange that right now we have text to video that's so high quality and there aren't any, any imperfections that would make you think otherwise. But this thing is truly, truly impressive with as to what it's able to do. Because, I mean, like I said before, I just can't see enough mistakes in these clips for me to realize that it's AI generated, whereas with other models, you can actually see that. So I would argue that this AI system with the photorealistic humans, which is a special feature that they decided to train on, I'm guessing that whatever kind of training mechanism they used for this one, they had a lot of high quality data sets. And as we know, data definitely affects the output. So with this right here, what we can see is that Runway is going to be the leader in terms of these photorealistic humans. And I think it's also going to be pretty fascinating because this marks a new stage where not even I can tell if something is AI generated. And for sure, I'm going to be questioning nearly everything I see online Especially when there's a HD video of a human, because I'm not going to be able to know if that is real or if it is AI generated. Now, unfortunately, with Runway, we don't actually have access to this yet. It does look very, very interesting and very, very effective as it's able to cover a wide range of styles and is able to essentially merge things that haven't been merged before in a very effective way. So it will be interesting to see who releases first. Will it be Runway? Will it be Sora? I mean, we've got a very fascinating, you know, few months for the rest of the year in terms of what the releases are going to be like and of course in terms of AI development. So it will be interesting to see what kinds of models do come out of this area. Now, in AI Video News, we got introducing the research of Our Foundation Model Character 1, available today at hedgerow.com Basically, this is something that can actually do reliable headshot generation. Tell stories that if you didn't know before, this is something that AI particularly struggles with. And we've seen that AI systems have been able to do this with Microsoft's Vassal one, But Hedra have gone ahead and of course released this for the public to use for free. It's something that's going to allow next level content creation with emotionally reactive characters. Now, there is one example that I want to show you that shows you how crazy this is because it's pretty, pretty incredible with. As to what you can do with this kind of technology. Let me show you. You can see here. Uncanny. Harry. AI says, meet my mate Dave. He's had a belated Father's Day message for everyone. Sound on, please. I made Dave on Father's Day with a midjourney image, an elevenlabs voice to voice, and a new tool called Hedra Labs. It's the closest thing to acting that I've seen from an AI generated video. And out of all the examples, even the one on Hedra's page, this by far seems like the most interesting example that I've seen so far.

My old dad, he was old school, you know, British, stiff up a little. He loved me and I loved him, but we didn't talk about it much, didn't have that sort of relationship. So when he got the cancer and my mum was looking after him, I'd see him. But I always thought I'd had time. My day. I got a call and I rushed to the hospital. I was going to tell him everything, but it was too late. He'd Already passed. I never got a chance to tell him, thank you for being a good dad. Thank you for teaching me how to be a good dad. Thank you for teaching me how to be a man. Thought dad's still around. Go and tell him now that you love him before it's too late.

So, yeah, that is pretty, pretty incredible. I mean, if we look at how the face is moving, how the skin is folding, it just looks remarkably impressive. And it doesn't just look like they've isolated the face. You can also see that the, the rest of the area is moving as well. So this is a truly, truly pivotal moment because I think maybe they're going to have cartoon versions be a lot more realistic. Because, of course, if you could just put in any realistic human's face and it be able to animate it, it's going to present a lot of issues and maybe even there could be some liability for Hedra themselves. So this is the kind of technology that starts to blur the lines between fantasy and reality. And this also brings the question, what kind of things do you think people are going to be in the future when we are actually looking at these kinds of content? If we can imagine a stage where AI gets to the level where the latency drops down to, let's say, a few milliseconds, are we going to be interacting with humans or these things over the Internet that are basically digital twins? Something to think about. And for the last clip, we have Elon Musk talking about Tesla's new announcements and of course his AGI revised date Tesla.

So you'll be able to access Grok through your Tesla and you'll also be able to ask it to do whatever you want to. You could ask your car to go pick up a friend or anything you can think of it, the car will be able to do it. Yeah, you'll be able to ask a Tesla to go pick up groceries, pretty much anything. Optimus is really going to be next level. You'll be able to skin Optimus in a wide, you know, pretty much with anything. So people on the Internet have asked me to make cat girls real. And actually you can make cat girls real if you, you have a robot. You have a robot cat girl.

Yeah.

Optimus will be able to pick up the kids from school and Optimus will be able to B school if you want. Be able to teach kids anything. Yeah, it'll support any language too. Do I think AGI will be next year? Probably it's not next year. It's say 2026 at the latest for AGI at the latest. I hope it's nice to us. So one defines. AGI is smarter than any human. I think it's. We're less than 24 months away from that. Yeah. Please be nice to us. AGI humanoid robots will usher in a level of abundance that was hard to imagine. There will be no shortage of goods and services.