AI News Just Landed! - Free AI Video, NotebookLM Update, & Open AI Singularity

Hello everybody. Welcome back to another video here on the Matvid Pro AI YouTube channel. It has been a while. I know, I'm aware. My apologies for not uploading in the past almost two weeks. First it was the holidays, which is understandable, but then I got sick and I'm overcoming this cold as we speak. I'm sure you can hear it in my voice. So my apologies if I sound like there is a literal frog in my throat. He is in there somewhere. I think his name is Robert. Kick things off with this absurd tweet from two hours ago from Sam Altman, the CEO of OpenAI. He says he's always wanted to write a six word story. Here it is near the singularity, unclear which side. Hello, Sam Altman, sir. What does that mean? Is he saying that we are absolutely approaching the singularity and he doesn't know if we've achieved it yet? Is he just trying to drum up hype? And here's another question that I have for probably all of you in the AI community. What is the difference between the AI singularity and AGI? Or is it the same thing? Are they synonyms? I will say, if he isn't just trying to drum up hype, there are some pretty wild assumptions we can make. First, if we're really achieving something that is that huge singularity, AKA AGI, whatever you might want to call it, does that mean the AI has finally learned how to improve itself automatically? Because that would be pretty insane. The implication really to me here in my mind would be a positive feedback loop with the AI system where it can continuously improve itself to the point where it's just more powerful than any of us could have ever imagined and more capable. I don't know though. The skeptic in me thinks that maybe they just had a good breakthrough and he's hyping himself up or he's feeling hyped, so he wants to, you know, spread that onto Twitter or X. I think that's also very possible. Let me know what you guys think about this in the comments below. Next up, folks, comment below. Google has finally added Gemini 2.0 experimental into NotebookLM. First of all, we've got a brand new feature that allows you to join an AI overview and actually directly engage with the AI hosts that are quote unquote, podcasting about whatever your Notebook LM entails. I'm a pretty big fan of Notebook lm, so I might actually have to do a deeper video on this one, but here is a quick clip for what you can expect.

These laws are like the rulebook for how everything in the universe moves and interacts.

Okay.

Oh, hey, I think our listeners got something to say.

Can you explain the three laws of motion with a basketball example?

Sure, I can do that. Let's use a basketball. Imagine you're dribbling Newton's first law. That ball keeps moving until you stop it.

So now you can dynamically ask questions and directly engage with the podcast. Ask them to elaborate on certain things that you might not understand. Maybe even do some special requests. The podcasts are already customizable, but this takes it a step further and turns it into like a live video call. Really, really cool. Think of this for studying. I mean that is very close to a straight up live tutor. Just paste your chapter into NotebookLM, have it make the podcast and then engage directly. Pretty darn cool. NotebookLM also has a brand new interface with three panels. We've got sour all the way over here on the left hand side. Your text chat right in the middle and then the studio all the way on the right hand side where we can do a deep dive conversation. Add specific notes. The user interface of a product can really make or break it. You know what guys, let's jump right in and do a little example for ourselves. Alright? I'm grabbing this random 17 page PDF. Generative emergent communication Large language model is a collective world model. Sure, why not? Slap that right in a notebook. Lm alright, it's giving us a short little description of the paper. Let's do a deep dive conversation. Focus on dumbing down the research paper for people who are not researchers. Sure. Generate. All right. It did a solid 15 minutes on this paper, which is pretty crazy. We're probably not going to listen to the whole thing, but let's give it a shot.

All right, so get this, we're diving into large language models today.

Oh fun.

These AI systems, right? They can like write poems.

Yeah.

Code websites, basically chat like a real person.

Yeah, they're pretty wild.

Ever get the feeling they like almost, I don't know, understand the world kind of like we do?

Yeah, it's kind of freaky how good they're getting. Right?

Well hold on to your hats folks, because new research paper is saying that might not be too far off.

It's pretty mind bending stuff.

So for this deep dive, we are going deep into this research paper.

Yep.

And this paper, it calls this whole thing Generative Emergent Communication.

Catchy, right?

So what is Generative emergent communication?

Okay, so picture this. You've got these AI agents, like little virtual robots, right?

Okay.

And they're in this simulated world just hanging out. Now here's the thing. They don't have any language to start.

Oh, wow.

Just a bunch of signals, beeps and boops, you know.

But how do they communicate?

Well, that's the wild part. As they interact, bump into each other, try to figure things out, they start developing a communication system, a language, all on their own.

No way. So just by interacting, they create their own language. That's incredible.

Yeah, it's like watching evolution in action.

But how does that lead to, like, actually understanding the world?

Right. So the paper is saying this isn't just about learning to say the right things. It's about building a model of their world.

A model?

Yeah, like a mental map, a representation.

Gotcha.

And they do this, get this, simply by communicating with each other.

So it's like they're saying, hey, I see a tree. And another one is like, yeah, I see it too. Let's call it tree. And bam. Shared understanding.

Exactly.

Mind blown.

And it gets even cooler.

Oh, there's more.

As they keep chatting, their language and their understanding get more and more sophisticated. They start to figure out not just objects, but, like, relationships, cause and effect, even abstract ideas.

It's like they're building a. Hello, can you hear me? Join the discussion. So this is a whole separate feature when you try to join the discussion. I'm having trouble even getting it to work. All right, well, that's interesting. For some reason, this feature is not working for me whatsoever. I click join and literally nothing happens. And it says it's using my microphone, but it's just not picking anything up. And it's using the right device in Chrome, so I have no idea. Let me know if you guys can get this one to work. This isn't the first Google AI issue I ran into today. Anyways, let's move on. Next up, folks, free online AI video generation using Hailuai is here, literally just called Hailuaifree.com and yeah, that's pretty much it. It's a pretty simple website. Looks kind of thrown together and is absolutely littered Ads. All these ones where it's like View PDF or run generative AI models, you know, all these targeted ads, don't click on them. They're almost always some BS you don't need. But they're using this ad money, I assume, to actually pay for free AI video generation, which is pretty cool. So let's do some generating. All right, we'll do View of Earth from Space. A meteor flies in from the right side, hits Earth and the planet explodes. Sure, let's generate. Apparently generation time here is like 5 minutes or so. Anyways. Yeah, no login or anything. The website actually works, which is pretty crazy. Obviously. Actually using the Hailu AI model to generate videos, it looks like they're at five seconds, which isn't great. But hey, it's completely free cuz the website just has ads. So that in my mind is pretty darn awesome. It's democratization of this technology, right? To me, honestly, the bigger topic here is who in the right mind at this point is going to be paying for Sora? A lot of people consider Sora to be worse than Hailu Minimax for video generation. And Sora is pretty expensive. You get some limited generation with a normal chat GPT subscription, but if you want the actual unlimited generation, you have to shell out 200 bucks a month. I have a feeling a lot of these AI video generation economic models are going to be short lived because of websites like this. A lot of people will be willing to see a bunch of junky ads to be able to generate their videos for free. Alright, while that video generates on the website, we'll move into our next topics. Dreaming to Loopa brings us a new very fast text to audio model used mostly for sound effects. But I pick it can do some quick little jingles, some musical jingles. Anyways, this new model can generate pretty 30 seconds of 44.1 kilohertz audio in just 3.7 seconds on a single A40, which is lightning fast. Obviously it can output audio faster than it can generate almost 10 times faster. Dreaming Tulupa has provided a quick little video for us. Let's take a look. Obviously a little New Year's celebration there. Now we have a human whistling harmonizing with a natural bird song. Finally, quiet speech and then an airplane flying away. All right, that one, I think at the end there was a little bit more rough. It's those higher frequencies that the AI can sometimes struggle with. But it did. All right. Fact of the matter is this thing is lightning fast. Up to 30 seconds of audio. And honestly the audio generation isn't half bad. Let's give a prompt, a quick shot, video game coin sound effect and we'll do this like a very short duration. There we go. Generated audio. Okay, that wasn't great. That was a little bit closer. Alright. It might not be good for those sound effects. Maybe we should try something that you'd hear in real life. Alright. Cat meows Then a goofy clown falls down the stairs. At the end we hear a Splat. Sure, that sounds great. Okay, now that was golden. I really like that one. Very solid. Dropping 10,000 golf balls down a well. Sounds more like explosions or fireworks or something like that. Squeezing sound. Squeezing a lemon juice. I love how fast the model is though. That's really the big difference here. Instant generations, instant iterations. Sounds like something was just being like rubbed. Like I don't even know what that was. Car starting. Well, that's definitely a car starting, but that might be the worst sounding car I've heard in my whole life. Anyways, yeah, this one's pretty cool. Completely free. Take a peek. It's all linked down below. Now, do you guys remember Huan Yan Video by Tencent? This is an open source model. As you can see, it's available on Hugging Face. They've got their own Chinese website. They've got a GitHub for this thing. But it's a great video model. State of the art for sure. I don't think anything these days comes close to what Google's VO2 can do. But VO2 is not out for the public. Where this one is open source, anyone can modify it, anyone can edit it. And like I said, it's a pretty fantastic model. I mean, just look at this video of this fluffy cat going down and trying to eat a cheeseburger. It's pretty darn awesome. Now the thing here with this model is since it's open source, people have been modifying it. And now guys, we are starting to see Lauras emerge for one Yawn video. If you guys don't know. Image generation by AI will oftentimes use loras to customize certain aspects. Fine tune. It's basically like instant fine tuning, right? But for video, it's completely different than images, right? Because motion is involved where a Lora for an image might just, you know, be for a certain set of characters, right? So they're perfected in that Lora. This one we can perfect things like walking. So obviously we've got this anime character walking and then we've got that same walk being performed by this 3D character. And as we scroll down, it gets better. We've got Studio Ghibli style. This is by Suruva 19. That's already starting to look pretty great. But this is, this is the fine tuning I'm talking about. We can fine tune for certain artistic styles, specifically, you know, Studio Ghibli, for example, where we get that motion, that detail, right? It's starting to pick up on those specific factors. And the lora version of 1 Yan video for Studio Ghibli will perform better in most circumstances. We've got anime shots, which isn't necessarily Studio Ghibli, but maybe trained on, you know, TV shows, series or something like that. Anyways, if you do actually want to fine tune your own Huan Yan Lauras, this is the resource right here. It's on GitHub obviously. No easy way of fine tuning these Lauras yet, but I'm sure that's coming at some point in the future. Now, finally, guys, I do want to talk about this. This is Rodin Gen 1.5 and this is an AI that can generate 3D objects from images or a set of images. This new Rodin Gen 1.5 apparently achieves clean topology AI quad mesh. I'm not a 3D artist, so I have no idea what that means. Promo triangles. Again, not a 3D artist, so I don't know what that means. And stunning PBR textures. I do know that PBR textures means that it reacts to lighting in a certain like. So the texture can have certain areas where it's more reflective than others. And I only know that from like Minecraft shader packs and stuff. Anyways, you can actually try it for free, which is pretty darn cool. I already posted a quick little X tweet about this tool. You can see this is the image that I uploaded into it and this is the mesh that it output. And again, this is not a pro feature, so this is the lower quality topology. And mesh overall doesn't have the pro triangles, so it actually gets even better. But. But it's really absurd how well it was able to pick out what this 3D object is genuinely supposed to look like. There are a lot of clues to us humans in the image, but for a computer to do this, it's pretty wild and it does it quickly, a lot quicker than a human who is observing this image could produce this. Obviously this is just like the mesh or the untextured render. Check it out with the textures even crazier. It does a pretty great job at mapping them overall. You can see there's a little bit of mushy weirdness over here, but if you take a quick look, you'll actually be able to see that. The AI was able to pick up on the separated and heightened eyelashes of this little lemon character on a leaf. It was able to pick up that fine detail and then actually accommodate for it. And same thing with the eyebrows, they're genuinely bumping out. Pretty incredible. It does the top and bottom of the leaf and honestly Textures, the whole thing pretty darn for just going off of one image. And like I said, you can upload multiple images too, which is pretty darn awesome and get a more accurate result. This thing is pretty great for instant AI 3D modeling from images. If you do a lot of 3D modeling, to me this would be like a no brainer. So cool. So yeah, definitely give this website a look at. You can do some generating for free. I think they have a credit system here. You can see someone did a Shrek. Over here we've got Spider Man, Dinosaur. I mean this is really darn cool. It does a pretty great job. I love the ripped Patrick. That actually looks almost professional. Basketball2 man, that's just wild. It is crazy. AI is just dominating in all modalities, man. It's getting really, really good and it doesn't seem to be stopping. The real question is whether or not you could easily convert these into actual physical objects. Because then we would literally have text to 3D object. When you think about it, generate an image, turn it into 3D with Rodin 1.5 and then convert it into a 3D printable file. Right? And then print it. That would be really cool. Anyways, that is going to be it for this video. Honestly, not too much going on in the AI space as of late with the holiday season and such. Thank you for putting up with my sick voice. I know I don't sound great, but I'm trying to recover the best I can. More videos to come. I'm super excited for 2025. I think lots of awesome things are going to happen in the AI world. And more announcements to come as well. Have big plans for this year. See you guys in the next video. Goodbye.