SORA Demo FAKED? Elon Musk’s 18 billion. New AI Characters, New Humanoid Robot

So with another rather fascinating day in artificial intelligence, let's take a look at some of the most interesting stories that you probably want to know. So one of the stories that I actually picked up was one that was about Sora. So if you don't know, Sora is OpenAI's video generation tool. And it's actually pretty interesting what people found. Now, I don't think this is a big deal in terms of like how bad it is, but some people think that this is basically a crazy deal. So. So essentially they basically did the behind the scenes of the Sora clip. So if you haven't watched it, basically there was a Sora clip which was called Lighthead, okay, Or Balloon head or whatever it was called. And in that video, it looked really, really good. But the main issue that people are having is the fact that the video was edited a little bit more and the clip wasn't entirely AI generated. There was some rotoscoping and there was some manual editing. And for some reason people are having a go at this and they're basically saying that the AI just isn't good enough. Now I'm gonna include the video. So you know what the video is. You can see in the trailer here that it's about a guy who has a head which is just entirely a balloon. And it's an interesting concept that we wouldn't really be able to film if we didn't have AI generated tools where we could literally generate stuff like this on a whim. Now, the article in question essentially just breaks down the behind the scenes. And the. The funny thing is, is that the actual article isn't like a dig at OpenAI or Sora. It's literally just a behind the scenes thing where you can understand how it works. It says that the user interface allows artists to input a text prompt. OpenAI's ChatGPT then converts it into this longer string which triggers the clip generation. And at this moment there is no other input. So you can't even put images or references, stuff like that. And they actually explain here that explaining wardrobe for characters, as well as the type of balloon, which was our way around consistency, because shot to shot generation, there isn't a feature yet for full character consistency. Now this is where like many people and millions of people were stating that, oh, Sora is so bad, it's so awful is because initially when you generated some of the clips, there would be this face which you can all see does have some resemblance to, I guess you could say a human or some weird plastic creature that of course wouldn't be good to have in the final, final clip, which is why a lot of people are basically stating that, you know, this is generative AI and this kind of technology is awful. Now you can also see here that this was a clip that was also generated. And with this one, what we also had was you can see that the color has changed a little bit. And of course there was a head here that was had. That basically had to be rotoscoped out and just brushed out of the scene. So you can see right here that it says Roto. While all imagery was generated in Sora, the the balloon still required a lot of post work. In addition to isolating the balloon so that it could be recolored, it would sometimes have a face on Sonny, as if his face was drawn on with a marker. Then this would be removed in After Effects. And of course, there were some other artifacts. Now, I don't think that this is the most interesting thing. A lot of people are taking this and stating that, look, oh my God, Sora is awful because it generates inconsistencies. There were, you know, these mistakes in there. But I don't understand why people are stating that when literally OpenAI themselves stated that Sora isn't perfect. And Sora does have its own issues. The thing that I think most people should be paying attention to is the details around Sora, because this is a highly exclusive technology that we haven't even had access to and stuff that we've never even been able to look at yet in terms of a screenshot for the user interface. So the thing that I wanted to note down here was of course, the render times. You can see here that it says clicks can be rendered in a varying segments of time, such as 3 seconds, 5 seconds, 10 seconds, 20 seconds, up to a minute. And render times vary depending on the day. And demand for cloud usage says generally you're looking at about 10 to 20 minutes per render. And from my experience, that duration I choose to render has a small effect on the render time. If it's 3 to 20 seconds, the render time tends not to vary too much from between a 10 to 20 minute range. And we would generally do that because if you get the full 20 seconds, you hope you have more opportunities to, to slice and to edit stuff and to increase your chances of something that looks good. So this is actually pretty interesting because this actually gives us some more details on how long it takes to generate a clip with sora. I think 10 to 20 minutes is. I wouldn't say it's Insane. But something that we now know is that this is pretty clearly far away from, you know, I guess you could scale, you could say wide scale usage. Because of course I'm guessing that the inference cost for this must be truly incredible because I mean, 10 to 20 minutes for a render is pretty crazy. But something that I will say, okay, as someone that you know has done work in that industry before, I will say that whilst yes, 10 to 20 minutes does seem crazy because when we enter text right now, we get text straight away and it's pretty intent. And for an image, we wait like literally 10 seconds and the diffusion models just literally generate the images within seconds. When people are thinking about a video, it's like, whoa, why is the video 10 to 20 minutes long? I going to be honest guys, if you have anything and you know a lot about vfx, you'll know that visual effects is so timec consuming. It's something that you have to get right. And rendering stuff, if we're able, like with, with Sora 5 or maybe Sora 4 or whatever version we're able to get stuff that's really accurate with character consistency, it's going to change the game, especially if the render times are lower because render times are absolutely insane and companies shell out big bucks to render farms where they have stuff rendering all the time in order to their products done within the time. And a lot of times some movies don't even get done because the VFX artists are literally breaking their necks to try and get everything done on time. So I think this right here could be a complete game changer if there's more efficient architecture or if, like I said before, they scale up their compute to have that. And that was of course something that they did actually talk about. So I wouldn't be surprised if they did that. But now here we have Elon Musk close to raising $6 million from Sequoia and others. Now, I saw some tweets where people were surprised about this and I don't think this is surprising at all because like I said already on Twitter, this is where you are raising money based on future valuations. Of course, the company right now is, I guess, you know, we can argue semantics here, but the point is that yes, the company right now is raising $6 billion at a valuation of $18 billion, but I think most people who are investing in that are betting on the next 10 years what Elon Musk is going to do, because he does have a track record of actually completing the stuff that he does. And of Course, once again, that can be argued, but I think we all know that AI is going to be the most transformative technology. And if Elon Musk sets his mind to something, it's pretty rare that he doesn't succeed. Of course, there are delays in some technologies. Some technologies do require safety requirements and all that stuff. So I always think delays are necessary. But. But an $18 billion valuation, I don't think that that is that crazy, considering the fact that it is Elon Musk as well. It's not like it's some random startup started by a different group of people that even maybe have worked at the top AI labs. But Elon Musk is a household name. You have to understand that if he takes this stock public, okay, let's say he was about to, okay, he has huge, huge, huge influence, okay? Which means that he can command a huge, huge valuation. Now, this is, of course, for X AI, and I think X AI is definitely going to be working closely with Tesla with some of their AI integrations. And I think the scale of money and scale of compute that you do need to compete with the big players just goes to show because if he's raising $6 billion, of course he does have billions of dollars, you know, with his other companies, but those are also burning millions and billions of dollars too. So he's going to need some clear investment here. Because like I said, Meta is not playing any games. Meta have honestly ordered so much GPUs. Google's compute capacity is the biggest at the moment. They have the most compute out of any of the top companies. And OpenAI are partnering with Microsoft on a $100 billion supercomputer that's being built as we speak. So I think if you want to have any chance at competing at AI, this is clearly the way to go. And a lot of people are stating that, oh, AI is in a bubble. Look at this insane valuation. I don't think so. Okay? I think that this is actually where the valuation lies in the sense that so far, what they've actually done, they've done really decent on the benchmarks. And to be honest with you guys, they haven't slacked off either. They're a small, small team compared to the others, and they've moved pretty quickly. But this is, of course, something that we talked about before, but it seems that the deal is probably going to be finalized very soon. So I know a lot of people are, you know, giving Elon Musk flack for this, but it's not something that I think is too crazy. Speaking of chatbots, this actually brings me to today's sponsor, which is Talki AI. Now, this is a rather interesting website where you can actually talk to AI avatars that are many different characters. Now, not only is this actually a web app, you actually do have iOS and Android applications that makes this really cool. Now, there are actually some unique features about this that make it very, very cool to use. The first one is that it is completely free to use the. There's also a nice voice, and there are many different characters that you can talk to, so you pretty much never run out of people to talk to. Now, you can see right here that what we are currently on is we are on the Discover page, and this is where you can discover any character that you want to talk to. Now, I'm someone who, I guess you could say I'm a little bit weird. So I actually spent a lot of my time talking to inatimate objects like Cheese and, of course, inanimate objects like the universe. Hello, I am the universe. Ask any question. Now, there's also a search page, and this is actually where I found some of the more interesting characters. Now, in my free time, I actually do play Call of Duty, So I did actually scroll down to some of these characters, and we've been going back and forth about why I'm actually terrible at the game. And one of the coolest things you can actually do with these characters is that they actually do have custom voices. So it doesn't just feel like you're talking to a normal chatbot. It actually feels like you're talking to a real person. Where is Captain Price? You can see he just asked me where Captain Price is, and honestly, I don't know. Don't lie to me. And then, of course, what's actually cool about this is that if you don't like the character's response, you can actually change this and you can pick a response that you like in order to make the conversation go a completely different way. So you can see here, you can change it to pretty much whatever you want to do. And this is me talking to an Overwatch character that I'm guessing some of you probably know who it is. So, that being said, I'm going to continue talking to strange, different characters and some inatimal objects and let me know if you find this interesting. So, actually, we have something that I've been wanting for ages, and finally, it's fun. Finally here. Basically, they say that AI avatars aren't great actors, but they don't understand what they're saying or how they should say it. Basically there's now Express one where they've trained AI avatars to perform with a tone of voice and facial expressions and body language. So take a look at this trailer because I actually do have a lot to say because I think this is important in terms of, for business use, of course, not for like deepfakes and stuff like that. But I think this is actually important because it brings us to another level of realism.

Hello everyone. I'm very happy to showcase the new features and improvements from the research team for the EPIC Synthesia V4 avatar release. Now focus on my lips. They're snappier and align more precisely with each word I say to you. And my voice, do you hear the difference? I sound more like an engaging, energetic presenter. A noticeable shift from the typical toned down style.

Right.

This is just a glimpse of the advancements we're bringing to the world. We are dedicated to enhancing every aspect of our technology to provide.

Now, I don't think this demo actually showcases how good the technology is.

Very happy.

I think this clip here actually does it. So just take a look at this clip.

I am very happy. I am so upset.

Now, I gotta be honest, the voices don't sound that great, but I think something that I currently don't understand and I guess probably, you know, if we use Occam's razor, it's probably simple that, you know, the software, like eleven Labs and other companies that are in this space, they really, really, really should add some emotions to the text that you have. Because, you know, a lot of the times what allows voice to sound real is emotion. So for example, if I were to say, oh, I'm really, really happy, you know, that sounds genuine. But I was like, ugh, I'm really, really happy. That sounds, you know, there's sarcasm in that. So I think the subtle nuances are going to be pretty difficult for AI to emulate. I mean, I guess it's kind of a good thing. But the AI can't emulate those things because for deepfakes and stuff like that, that's going to be of course, good. But I do think in the future that as someone who does use these AI softwares for business videos and training videos and stuff like that, I think it would just help a lot because the problem is not deepfakes and stuff like that. The problem is that when voices sound monotone, they just lose your honestly attention. And it's honestly something that I try to keep in mind when I'm speaking because I don't want to sound like, okay, this is what happened today. I, I actually have to, you know, speak loud, then speak weird, then sometimes speak loud, then times speak lonely, then sometimes actually slow down, then sometimes change the pitch of my voice so that people actually do pay attention. Then we have Phoenix Generation 7 major improvements all delivered in less than one year. Our next generation of general purpose technology marks an inflection point in task automation. So we can see here that this is their new. So yeah, Sanctuary AI, if you didn't know, are a company that do produce robots that are pretty much general purpose. Now they actually showcased a recent demo earlier this year that I think was quite understated because it did show full autonomy in certain tasks that were pretty good. And I think them upgrading their robot is a pretty good step in terms of where we're trying to get to for general purpose humanoid robots. So I, I do wonder though if they're going to do some more demos and of course combined with their vision system, which is now called Genceller 8, which is now called Generation 7, if they're going to actually add some of the things that some of the other competitors are doing like reasoning via a large language model or using an actual base of the model. Because one of the things that we don't see for Sanctuary at the moment is that we don't see legs in this model. Now whilst, yes, let me actually state this for a fact, the model does actually have legs, but the problem is that we haven't actually seen them walking yet in any kind of demo. So I'm guessing that it might be a little bit harder than it looks. Now this is something that is pretty, pretty crazy in terms of pretty, pretty scary. Now essentially a Baltimore county principal was seemingly caught on recorded audio making blatantly racist and anti semitic comments. But after investigation, the audio was fake AI generated and a plot by the school's four former athletic director. Now I'm not actually going to play this clip here because explicit, I guess you could say. And I don't want the video to get demonetized so you know, you can watch yourself. But the point is, is that I think that this is pretty insane because the, the audio itself was actually realistic. Like when I heard the audio myself, I was like, wow, that actually sounds super realistic. I mean they clearly had a professional edit it because it wasn't just like something that was copy and pasted from Elevenlabs, but. And I'm guessing that maybe they use some open source software because usually what you can do with 11 labs is they can check the history of what people have generated and then they can easily find the person who generated that text. But I think this is pretty scary because you can see here it says it not only led to his temporary removal from the school, but also triggered a wave of hate filled messages on social media and numerous calls to the school. The recording also caused significant disruptions for the staff and the students. So it was pretty, pretty crazy guys. Like this is something that is absolutely incredible because this is one of the first cases that a story has gone viral where AI technology has been used in a way, which is unfortunately something that we have predicted, but it's pretty awful. And this is something that is quite frustrating because whilst yes, people like us just want to use the technology for enjoyment, entertainment, work, whatever productivity, there are people like I've stated before, that will use this for their own gain and for whatever reason it may be. So I think this is going to set a different precedent where we're going to have to use some kind of technology to potentially verify whether or not clips are AI generated or not. And maybe there's going to be some new legislation that any software that operates and broadcasts its services to the United States must have some kind of synthetic ID to where that can be replayed back and can be understood to be AI generated. Because stuff like this, I gotta be honest guys, this just isn't good for the future. Like we can all agree on, you don't want to be someone who's completely innocent and then someone, you know gets a clip of you from a WhatsApp voice note or from like a video, or like a video you made on Twitter or like a Snapchat you had five years ago and then has you fired and your entire life disrupted, which is pretty incredible. Now Perplexity actually added Voice available for iOS and Pro users. And I think this is going to be a really, really good change. And if you don't know what perplexity is this, they're not paying me to say this, but I think this is something that saved me hours and hours and hours and hours of research time. Sometimes it doesn't find the answer. But I got to be honest with you guys, this is a fundamentally game changing tool. I'm not sure why Google hasn't bought them yet because a lot of people who do use them don't really have bad things to say about them. Some people will say that they're just a Google wrapper, but honestly I don't care what they are because ease of product use is the Biggest thing, like, a lot of people will be like, oh, their product is just xyz. It doesn't matter, okay? If the product is easy to use, people love it and people want to use it, then that's it. The product has done what it's needed to do, and it's a product that has really changed the game. So, like I said before, niche products that do wrap certain LLMs will provide a substantial boost for overall companies in the sense that their LLMs get used a lot more. But this is something that I do think showcases how effective LLMs are in research. And it's something that I would say, if you haven't used it, definitely try it out, especially the free version. Then we could see here, and this is a pretty iconic moment. It's kind of like, I guess you could say, a nostalgic moment, because the first Nvidia DGX H200 in the world, hand delivered to OpenAI and dedicated by Jensen to advance AI computing and humanity. Now, this is pretty crazy because there was something like this before, because here we can see where Jensen Huang actually did this a couple of years ago. And you can see that Musk was here, of course, at the time OpenAI was founded. And you can see here it says, to Elon Musk and the OpenAI team, to the future of computing and humanity. I present to you the world's first DGX1. And we can see now that he's doing it again. So, I mean, it's a nice little callback that I think is pretty interesting. And what's funny is that, you know, so many people were asking, where's Ilya Satskova? Where's Ilya Satskova? Because, you know, it's something that is a burning question in many people's minds. But as I've said before, seems that he just doesn't want the limelight. His lawyers are probably said to just stay silent. I mean, I genuinely have no idea. This is literally just pure speculation. It's clear that he may not want to say anything or just honestly have no idea. I really don't want to speculate too much because I don't want certain ideas to get floating around based on where he may not may be, because I.

Think we, like, owe ourselves and the people of the future a better world. I think we should go off and figure out how to give everybody on Earth a great education and cure every disease and have great entertainment and go explore space and discover new physics and create more abundance, because I think we.

Like, owe ourselves so yeah, that was Sam Altman speaking at Stanford. And there was a lot of stuff, you know, that he said at Stanford. But unfortunately, it was like a private talk. And so many people actually did want to go in and talk to Sam Altman when he was there. And it's understandably why, because the future of the world is not pretty much in his hands. But there's a lot of stake here. And he did say that a new model was certainly coming this year. So I think also with along the some of the leaks that we've heard, that June is probably going to be the month that we're going to get a new model. It may be GPT5, it may not be GPT5. But according to Bloomberg and several pretty good sources, that is apparently the date at which we're probably going to see a new model. So that's definitely something to keep your eye out for.