EP 211: OpenAI's Sora - The larger impact that no one's talking about

This is the Everyday AI show, the everyday podcast where we simplify AI and bring its power to your fingertips. Listen daily for practical advice to boost your career, business and everyday life. I think Most people have OpenAI's Sora all wrong. So if you don't know much about it, OpenAI last week just released a new text to video model called Sora. And it is extremely impressive and everyone has been collectively talking about just that, how impressive the video is. But there's something larger at work here that I don't really see anyone talking about. We're going to talk about that today. Today and more on Everyday AI. Welcome. What's going on, everyone? My name is Jordan Wilson and I'm the host of Everyday AI. Everyday AI. It's, it's for you. It's your daily live stream, podcast and free daily newsletter helping everyday people learn and leverage generative AI. So if you haven't already, make sure to go to your everyday AI.com and sign up for the free daily newsletter. So every single day we not only go over what's happening in the world of AI news, we cover kind of new tools, fresh finds from across the Internet, but we also break down each and every podcast conversation in pretty great detail by me, a human writing this for other humans, right? So we tell you not just what's going on in the world of AI, but a topic each and every day on how you can learn and leverage what we're talking about. I tell people it is a free generative AI university, no matter what you care about. You can go on our website, you, you can go listen and watch to now more than 210 different podcasts, episodes, videos, et cetera, and go read every single newsletter we've ever written. So it is, I still don't know any other source, any other single source that has more free generative AI education across every single medium. It's actually wild how much content we have there. So before we get to that topic, on what's actually the larger impact of OpenAI, Sora, let's start as we do every single day, by going over the AI news. All right, so let's start with Scale AI and the U.S. department of Defense are joining forces. So Scale AI and the US Department of Defense are partnering to develop a comprehensive test and evaluation framework for the responsible use of large language models within the Department of Justice. Department, Department of Defense. So this framework will allow the DoD to deploy AI safely and accurately in military applications. This partnership between Scale AI and the DoD will create benchmark tests tailored specifically to DoD use cases to measure LLM performance and provide real time feedback for war use. So pretty, pretty big news there. Worth taking a look at. Next, Google has released a new AI feature. And just like Gemini Ultra 1.5, this one is also a limited release. So Google is introducing a new feature called Shop with Google AI that allows users to search for products and generate AI inspired outfit variations in a single click. That's interesting. So this will enhance the user experience by providing personalized and effortless shopping options and it will obviously be driving a lot more business and growth for retailers partnering with Google. So right now this is just slowly rolling out to a select few users inside the Google search app. So hey, holler at us if you have access. I kind of want to know, right? Can I use AI to just make myself dress better? That's what it sounds like. All right. And last but not least, the Internet has all at once fallen in love with with a large language model that is not new. So it, this could be a Valentine's Day hangover, but Internet users have all seemingly fallen in love with an older large language model named Grok at the same time. So let me just say this out loud. This is Grok with a K or sorry, not the Grok with a K. See, it's so confusing. So, so Twitter has their Grok with a K. This is Grok with a Q. Okay, so Grok is actually a California based semiconductor company. So Grok with a Q is a generative AI solutions company that has developed a unique technology called lpu which is language processing unit. And this is their interface engine and it's designed to accelerate the processing of large language models. So the reason why I think everyone is all of a sudden in the last 24 hours sharing about this on social media is because last week Grok with a Q won the won a large language model benchmarking competition. So now the Internet is all at the same time going wild over it. And yes, you should check it out. It is very impressive and essentially by using this kind of LPU language processing unit technology versus the general GPU technique that all other large language models are using, it can generate text based output much faster. Right? Is the quality there? You know, maybe, you know, it just runs other, it runs other local models such as, you know, models from Meta. So you know, it, the, the quality is really just obviously dependent on what model that you're using, but the speed is wild. So yeah, it's worth checking out. So we'll be sending a link and probably a, a video of that so you can just see it in our newsletter. So make sure, if you haven't already, go to your everydayai.com and sign up for that free daily newsletter. All right, that's the intro. Now. I'm excited for today. I'm very excited for today to talk about OpenAI Sora and the larger impact that no one is really talking about. And hey, to our live stream audience, thank you so much. You're listening on the podcast. Come join us live. It's a good time. You know, you can network. That's the good thing. There's a lot of networking happening, happening in our comments here at Everyday AI with our daily live stream so you can come and connect with, you know, Tara, who's joining us here from Nashville, and Dr. Harvey Castro, a top voice in AI Rolando. Thank you for joining us. Nancy and Jay Douglas, everyone, excited to have you all on here. But hey, I want to know right now, what are your thoughts? You know, before I get too deep and I'm going to tell you right off the bat. All right, so if you're on, if you're on a limited, you know, time schedule here and you're like, all right, Jordan, get to the point. I'm going to tell you right away, don't worry, I'm not going to drag you on for 20 or 30 minutes. But I want to know from our live stream audience, specifically, what do you think of OpenAI's Sora? All right, so let me just, before we get there, ask one more question and I want to know from everyone who's joining us live, give me a yes or a no or if you're joining us on the podcast, I always leave in the show notes, you know, in the episode description. You can email me, you can connect with me on LinkedIn. I read all your emails, all your messages. Sometimes it just takes a minute. But let me know, should Hot Take Tuesday. This is Hot Take Tuesday, right? Every Tuesday I come with a hot take. You know, something that's outside of the normal AI news or bringing on guests, which is what we do the rest of the week. Should this be a call in show, Right? So the, the software we use to stream, called Streamyard, there's, there's a feature where we can kind of have a little waiting room and you can call in just like this was, you know, an AM radio show, right? And you can call in with your opinion. So shout out to Nisiani, who, who dropped this recommendation yesterday. Should we do that for Tuesday? Should you come on and you know you can, you can join and you know, we'll have someone kind of quote unquote take your call, you know, in the waiting room and you can wait in line to come in and drop your hot take. Let me know, yes or no. Does that sound fun? Or is it too early in the morning for. For you to come with some spicy hot takes? I want to know, you know, let me know, yes or no. Should, should this be a call in or should this just be a random. A random rant from me? All right, Woozy says do it old school radio style. I love, I love some, some just call in radio am. Right? All right, let's get cooking in this hot take Tuesday. I promise you we're going to deliver. So let's just go over the facts and then I'm getting straight to the endpoint. All right, so Sora is a new text to video model from Open AI and it is very impressive. All right, it is very, very impressive. But let's just go ahead and tell you the three things that you need to know that are about the larger impact that. And two of them are things that no one is talking about. I'm sure maybe someone is somewhere. I can't find anyone. Ready? Here's. Here's three things that you need to know and two that no one is talking about. So one, Sora is light years ahead of any other text to video platform right now, so including Runway, Pika Labs and also Meta and Google have previewed their new text to video. Right? So Meta's Emu video and Google's Lumiere. All right. It's also worth noting right now, out of the five that I just mentioned, so OpenAI Sora, Runway Gen 2, Pika Labs 1.0, Meta's EMU and Google's Lumiere, only two of them are publicly available right now. So. So only Runway and Pico Labs. Right? So OpenAI Sora is not publicly available. However, people do have access. So kind of red teamers are those people putting it through safety precautions in a select group of visual artists. So yes, it is technically kind of out there, but not really. The only ones publicly available for all of us are Runway and Pika. But OpenAI Sora is. It's in its own, I would say even to put it to say it's in its own category is doing it injustice. It is in its own sphere. You can't compare at least early results. You. You literally cannot compare the quality of, of these products. All right, so that's number one. And we're going to talk about that more. Number two, the timing of Sora, with all that's been going on in February at OpenAI, means that AGI is near. Artificial general intelligence is closer than we all think. Yeah, hot take Tuesday. We're bringing, we're bringing some fire. And you know, I got receipts. Stick around. Right? And then number three, I don't think anyone will catch the combination of Open AI and Microsoft. All right? And I'm going to lay out, you know, a couple scenarios, but I think that they, that Sora and what SORA means specifically as it, as it comes to artificial general intelligence. AGI. Right. If you don't know much about AGI, I'm going to get into it here in a minute. But that's essentially when, you know, when AI systems become smarter than the average human or the smartest humans at general things. Right. You already have narrow kind of AGI or narrow AI that's outperforming humans and specific tasks. But AGI is when, okay, when these AI systems are better than most humans at general tasks. They were closer than we think. And I don't think anyone right now is able to catch the combination of OpenAI and Microsoft. Right? Microsoft reportedly owns 49% of OpenAI with a large, many, many billion dollar investment in the company. I don't think anyone's going to catch them. I really don't. I'm gonna lay it, I'm gonna lay down a couple scenarios in which it could happen, but it's an unfair advantage right now. All right, so let's get to those three things. I cut them down. Sora's light years ahead, number one. Number two, the timing of this means AGI is near. And number three, I don't think anyone will catch OpenAI and Microsoft. So let's, let's look at number one. Let's look at number one first. Right? And if you're joining us on the podcast, I apologize, you, you're not going to be able to see this very well. But I'm going to go ahead and I'm going to show some examples. All right? So make sure to check the show notes. You can come see this. I'm sure other people have done this, but I took four different clips from OpenAI's Sora. Okay. The good thing that I liked in their research paper, OpenAI allows you to download their results as well as see the prompt that they use to generate these results. Results. All right. But there's a lot more that goes on behind the scenes than, than just that, right? We don't know how many attempts, Right? Because if you've used something like Runway or Pika Labs, like I have, sometimes, you know, the tenth attempt is better than the first. Right. Or sometimes after 20 attempts, you'll get something, you're like, oh, this is much better than the first attempt. So we don't know kind of the process of how Open AI generated this, or it could have just been. Yeah, here's literally one prompt, copy and paste, one shot. It could be. All right, But I'm going to go ahead for our live stream audience anyways, and I'm going to share. All right, I'm going to share this video that I did, took the exact same prompt, downloaded the videos from OpenAI. So we have open AI first, and then we have Runway gen 2, all right? And I want you all to see it, and I want to hear all of your thoughts as well. I'm going to do some light narration here just to tell you what's going on. All right, so our first one here we have woolly mammoths walking through the snow. And again, first we have OpenAI, and second, we have Runway. So with OpenAI's, it is fantastic. So here we go. The video is playing. It looks somewhat real, right? No one knows what a woolly mammoth actually looks like. And then we have Runway version 2. It doesn't look bad, Runway gen 2. But you'll see some of these woolly mammoths are kind of walking backwards, some are missing limbs. Where the one from OpenAI is. Is pretty. Is pretty sound. It's pretty solid. All right, and then we have OpenAI's Sora model with this kind of astronaut theatrical, you know, wearing these red hats on another planet. The OpenAI is fantastic. Also, I'm going to pause here, and I'm going to say one thing that's worth noting that I should have started at the top of the show is what you can produce, allegedly, right? A minute. A minute of video in OpenAI with a single prompt. And you'll see in this example here, this kind of astronaut red hat example from Sora, it actually cuts to multiple shots, which is something that right now you cannot do in Runway gen 2 without regenerating, right? So you have a minute, right? So OpenAI Sora creates a minute, a video with splicing in times, multiple shots together. Sometimes it doesn't. I'm sure there's ways to control that. And with Runway Gen 2 and all the other AI video models right now, it's essentially four seconds. You can kind of stitch them together and extend it. You know, there's kind of some workarounds. You can get maybe up to 16 seconds, but it's all kind of the same. So OpenAI's Sora model. Great. Runway Gen 2 didn't do that good, right? It's just a random, random guy here in a motorcycle helmet. Not really moving, nothing very theatrical. All right, then we have our Sora. This is supposed to be a drone shot over a kind of like a gold mining town from back in the day. And Sora looks pretty nice. It's pretty smooth. It does just that. Runway Gen 2, you know, it has a similar look and feel, but if you look at the, the subjects of the image, it looks like maybe two horses and one person. But if you look kind of the person or the horse morphs into something else at the end. Not super cinematic. Doesn't really seem like a drone shot. All right, and then our last one here, OpenAI's Sora. It is a close up. It is supposed to be a close up of a woman's eye, kind of her stare. The detail on this one is mind boggling. I wish I had more words this early in the morning to describe this one from OpenAI's aura. But you can see the. Oh, gosh. I mean, the details of the eyelashes, the details of the skin, the lighting. Right. From the pupil. There is a reflection in the pupil. With, in. Even the reflection in the pupil seemingly has correct composition. This is, this is wild. This video is wild. Right? And then in Runway gen 2, you know, to runways credit here, this, this doesn't look real, but it's actually a pretty decent shot on this last one from Runway, right? You don't get the detail, but you do get a woman. Her. Her skin tone is, is very. Looks kind of AI generated. I guess it looks either like very AI generated. The skin tones are too smooth, but there are some nice shadows. There's some nice lighting going on with this last one from Runway. So it's actually not too bad from Runway. All right. But long story short, if you, if, if you're not going to watch this one, the comparisons are that's, that's apples in, in. In basketball, right? We're not comparing apples and apples. We're not comparing apples and banana chips. We're not comparing apples and fruit. Sora is in another sphere. It's not even in the same category, right. I have a little bit of a background. I talk about this sometimes in, in Martech communications, but a lot of times that involved creating videos, right? Created a lot of video in my day. So I, I, I have a decent eye, you know, more than, more than the average person. I've spent probably more than a thousand hours safely, I would say multiple thousands of hours either shooting or editing or you know, video production. It's not close. It, it's, it's, it's not even fair. How far ahead for a first generation model, right? For a first generation model, which is what Open AI's Sora is, it should not be this good, right? And that's going to lead us to our second point. You know, even if you want to go back and look when Chat GPT came out, right. So when Chat GPT came out, GPT 3.5, the world was shook. I was not impressed, if I'm being honest, because our team had been using the GPT technology for three years and I thought at the time of its release if you wanted to use the GPT technology to create text from text, right? So a text to text prompt, I didn't think Chat GPT was even a top two option at the time. I thought what was then called Jarvis, what is now called Jasper was better. I thought Copy AI was better. So even OpenAI, right, which they're known for their now I think they're really well known for Chat GPT and kind of text to text and multimodal with text. It wasn't even, I don't think a top use of its own technology at the time. Right. The same thing when Dolly, you know, if you look at Dolly 3, Dolly 3, I'm being honest, is not, I would say in the top three. Maybe it's in the top three. Right. But mid journey is so far ahead of everyone else. Then you have stabilities, you know, image generation, you have Leonardo, you have some other great products. Is Dolly top three? I don't know, maybe, maybe they're number three or number four, but they're not number one. Right. So when you look at OpenAI's first iteration or first attempt at something, it's usually not mind, mind boggling. The SORA model is mind boggling. I don't understand what this means for the future of creative. We're going to have another show on that because that's not what this is about. That's what everyone else is talking about. We'll talk about this. Let me just say this. It is going to be extremely difficult for the average human to understand what is real and what's not. Right? Because we always thought, oh okay, well you know, on photos, you know, if you look at A photo long enough. I think it's actually kind of easier on photos because it's still. And you can inspect it, right? You can look at the fingers. Oh, are there six fingers? Right? Or, oh, is the arm bent in a weird way with video? I mean, some of this video looks so smooth, right. And how your brain processes things. It is much harder if the quality passes a bar, if it's a yes or a no. If the quality passes a bar, it is much harder to detect AI video than it is in AI images. And because we've had this long period over the last probably year now where AI images have been good enough to almost pass for real. So, you know, people out there on social media, your average consumer, etc. We've. We've been exposed to this in a large scale and we've had time for our brains to kind of rewire things and to first see something and be like, oh, okay, is that AI image? Is that an AI image? Maybe there's no warning sign. There's no ramp up here with OpenAI. It is with. With. With Sora. It is entirely too powerful. Yes. There's still some instances where, you know, know, you could clearly tell. Right. But if you can, in one prompt, create a minute video and maybe you generate it two or three times, I can guarantee the. The. At least from samples, we're seeing so far the majority of that one minute. If you cut it up, no one's going to be able to tell. This is coming from someone. Again, I've spent thousands of hours in video photo production. You're not going to be able to tell. And this is the first model. My gosh. Wild times, y'. All. Yeah. What Juan is saying here. Juan, thanks for joining us. Juan says, wow, that's a huge difference between sora in Runway gen 2. Definitely a lot, A lot more realistic for Sora. Yeah, absolutely. Mind. Yeah, Mind blown emoji. A lot of mind blown emoji. If you're listening on the podcast, make sure to check the show notes today, the episode description. We'll leave a link. You gotta. You gotta watch the side by side comparison. All right. It's pretty telling. Hey, here we go. Nancy, Nancy, am I paying you today? Nancy says something. She says this will have to be a premium add on because Tokens. Yes. All right, we're gonna get to tokens and compute because that's a big piece. So let's now talk about number two, which I think is the hottest topic for today's hot top hot take Tuesday Foreign. Are you still Running in circles trying to figure out how to actually grow your business with AI. Maybe your company has been tinkering with large language models for a year or more, but can't really get traction to find ROI on Gen AI. Hey, this is Jordan Wilson, host of this very podcast. Companies like Adobe, Microsoft and Nvidia have partnered with us because they trust our expertise in educating the masses around generative AI to get ahead. And some of the most innovative companies in the country hire us to help with their AI strategy and to train hundreds of their employees on how to use Gen AI. So whether you're looking for ChatGPT training for thousands or just need help building your front end AI strategy, you can partner with us too. Just like some of the biggest companies in the world do. Go to your everydayai.com partner to get in contact with our team or you can just click on the partner section of our website will help you stop running in those AI circles and help get your team ahead and build a straight path to ROI on GEni. I didn't ask for the normal flame emojis, but I saw some, you know, Tara, Tara wanted me to burn this down. We'll. We'll try to keep it real. That's what we do here. But we bring receipts. All right? So Sora. Sora signals that OpenAI is close to AGI, or maybe it has already achieved it. All right, y', all, I don't, I don't say this lightly. Don't worry. I bring receipts, I bring a paper trail. The fact that this is a first model, and for some of the facts that I'm about to lay out, we have to wonder, are we at that point of AGI, right? If you read a lot about AGI like I do, or maybe you don't, a common train of thought around artificial general intelligence. And again, and I'm oversimplifying it here because our, our audience is for the everyday people, right? We're not talking to, you know, an entire audience of people who, you know, build machine learning applications on a daily basis. I'm trying to make this simple to understand. Essentially, artificial general intelligence is different than artificial intelligence. So AGI, this is what OpenAI has been openly working toward for years, right? Meta and Mark Zuckerberg have recently said that they are openly working toward AGI, right? So AGI is kind of scary. It's unknown, right? And that's essentially what happens. To oversimplify things, it's when the AI becomes smarter than us across the board. It's when the AI doesn't necessarily need us. It's when the AI can fix itself and improve itself. Right. Essentially is when it displays intelligence across the general, kind of general field of expertise or a general field of skills. Right. So you have something that's called kind of like narrow artificial intelligence in general. Right. So narrow is fact or, sorry, kind of more skill set based. Right. And in those instances, and I'm going to have a chart here from DeepMind, a very famous chart at the end to explain it. But essentially, in certain tasks, obviously, AI is already way outperforming humans. But when we talk about artificial general intelligence, AGI, that's when AI can outperform humans, the average human on a variety of tests all at once. Not 10 specific individual, you know, tests, but one AI system. Can it outperform the average human on general tasks? Let's get to receipts here. Let's look at the timeline. All right, the last month. And I've been talking about this a lot, so if you follow the show every day, first of all, thank you. I appreciate you guys. Second of all, you'll know I've kind of hinted at this before. The writing's been on the wall for a while now, and I saved it for a hot take Tuesday. All right, but let's just take a look in the last month. So Sam Altman has talked multiple times in multiple interviews about the next version of GPT, you know, presumably GPT 5 or, you know, who knows, we might see a 4.5 first, but the next version of GPT 4 or GPT 5. Sam Altman has talked multiple times about the increased ability to reason which leads to AGI implications. Right. We've talked about agents here on the show. Right. We've known for a long time that OpenAI has been working on agents, but we just saw the first kind of official report. I believe it was from the information. We'll link it in the. In the newsletter. But we saw the first official report on OpenAI's agents. Right. And I talk about it here on the show. Essentially two different kinds. One that can control your device, whether that's your computer or your phone, we'll see. And another agent that can perform actions on your behalf. Right. So it can perform actions on a website or an app. All right, you see how that leads to AGI? Yeah, you can't have AGI without a system being able to perform like a human. All right? So it has to be able to control the device, and it has to be able to perform actions on your behalf, and it has to have a Better model. Right? Okay, so we've crossed those two things out. Number one, there's a new model. Number two, it can perform actions like a human. Number three, compute. Compute. Sam Altman has, you know, it's been rumored and widely reported, and he even joked about it on Twitter. I still can't call Twitter axial. It just sounds weird. So Sam Altman joked, but, you know, it's been reported he's trying to raise $7 trillion, you know, for essentially compute, for energy, for these GPU chips or for, who knows, whatever a chip is after a GPU, he's really doubling, tripling 10xing doubt on compute, on chips, on powering generative AI. All right, so combine those three things. Next model being better at reasoning, number one. Number two, open AI's agents. Number three, raising $7 trillion for more compute. And then last but not least, Open AI previewing sora. All right, so I get what you might be saying, hey, Jordan, you're just a nerd. Sora has nothing to do with AGI. It's video. No, it's not. No, it's not. That's topical. This is not video. Is it video? On the surface, yes. But you have to understand what that means. All right, let's unwrap that, shall we? Because I think most people are so blown away by the outputs, they're not looking beneath the surface. And y', all, guess what? It is literally right in front of us. All right, so Open AI released quite a few things all at once when this SORA announcement came out. And we've now seen that there's been a small team working on this for more than a year. All right? But reportedly a lot of even Internal employees at OpenAI just found out about this right before its release last week. So there's been a small team going stealth on this, investing presumably a lot of time, energy, and resources into this project. So most people just stopped and they only looked at the OpenAI page and they looked at all the videos. Right. I was talking about those videos they downloaded. But there is something, I wouldn't say hidden, but you had to really care to read the research paper. It was a completely separate piece. Right? Everyone just went and, you know, oh, you know, all our Billy boys out there that are just trying to, you know, trick you into, you know, reading their newsletter or whatever that's written by AI, and they're just selling you crap products. They didn't care to look at this. All right, we do. I'm sure someone else on the Internet has, but, you know, this isn't where the conversation is, but you got to read the research paper, y'. All. I think I read the research paper before I even downloaded any of the videos. And I said, okay, here's our hot take Tuesday. All right, so let me just. Let me just read excerpts of this. Okay, so in the research paper, which I recommend you all read, doesn't take a lot. It says video generation models as world simulators. All right? And then I'm going to read the last sentence in one more sentence of this little excerpt from the OpenAI research paper on Sora. So they said our results suggest that scaling video generation models is a promising path towards building general purpose simulators of the physical world. One more sentence. SORA serves as a foundation for models that can understand and simulate the real world, a capability we believe will be an important milestone for achieving AGI. Okay, let me highlight those things one more time in case you're not watching them and you can see them highlighted on my screen. OpenAI is calling this World Simulators Building General Purpose simulators of the physical world. And they're saying SORA serves as a foundation model for models. Or, sorry, SORA serves as a foundation for models that can understand and simulate the real world. A capability we believe will be an important, important Milestone for achieving AGI. Okay, guess what, y'? All? No other OpenAI releases were this direct about their correlation to AGI. Okay? Not the new version of GPT4 Turbo, not custom GPTs, not the new memory feature. Right. That was released last week. Not Dolly. None of these things. None of these things. Did OpenAI draw this strong of a correlation between a new product or a new release and AGI? All right, let's continue to unpack and talk about this. So SORA is a world simulator. I'm going to try to remember this and, you know, actually shout out. I mean, you do have to give a shout out to Runway here as well. Right? Because Runway, quite a while ago, this is many months ago, they kind of introduced what they called general world models. And I do believe that in. In a similar vein, SORA is. Is kind of positioning, or, sorry, Open AI is kind of positioning SORA as such a world simulator. So this is more about understanding how the world works. So we're talking about real world intelligence in three dimensional physics. Okay, this is not. Will it mainly just be used for looking at this cool video I can create with text? Yes. But OpenAI is using it for a much larger purpose and we're going to get a little bit into the technical aspect here. As well. But in short, this is helping OpenAI understand and predict the real world. Let me say that one more time and tie it back to AGI. This helps Sora helps OpenAI understand and predict the real world. What does AGI need? It needs to be able to understand how the world works physically, spatially, physics, relationships between things, etc. Can't do that right now, can it do that with Sora? Maybe that's what it seems like where we're going. All right, so again, let's go into more receipts. All right, let's look at again, Sora's research paper for an example on what this real world understanding is. All right, so if you're joining on the podcast, I'm going to do my best here, but Open AI shared three different examples, three different videos in applying different tiers of compute. Again, if you're a machine learning researcher, I mean, yeah, you can email me and tell me how I'm wrong again, when I'm talking here on Everyday AI, I try to oversimplify things. Okay, so people can understand them, but essentially think of computes as layers of technology that is applied to something. Okay, so in this example, there's three different videos of a cute little dog playing in the snow. Playing in snow, right? So your base level of compute, right? Let's just say if you apply this technology behind Sora on one layer, you can't really tell. You can't really tell, is it a dog, is it in, is it the snow morphing into something? You can't tell. It doesn't even look like a dog. You can barely tell anything on a base level of computer. All right, then you're saying, then they're showing an example of 4x compute, right? So let's just say. So we can easily visualize it, the technology behind Sora. Let's just say you stack it times four, or you run it through it four times, or you apply four times the compute, four times the technology toward it. At this point in the middle here, you have what, you know, looks like a dock, looks like an AI generated document. You know, this is kind of what we kind of have now. I think with the current models. You know, this, you know, you can go look at it. That video looks a little closer. Maybe Runway and Pika are a little better, but it looks a little closer to our, our current day text to video. So who knows what Pika has cooked up for 2.0 or what Runway has cooked up for gen 3, but at least right now, and when you look at 32x compute, that is what is on the right on my screen or if you go look at those three videos, that's what it is. So now think of applying that technology 32 times or applying 32 times the compute power behind OpenAI Soar technology. Then you get a video that looks pretty real, right? So we'll share this one in the newsletter as well. But there's a very famous, you know, famous video. One of the first AI videos that kind of went viral on the Internet was Will Smith eating spaghetti, right? And Will Smith just keeps on morphing and you know, even in every single frame everything changes. And you know, the spaghetti is morphing with his face and his mouth is morphing with his eyes, right? Like it's a mess, right? And that's kind of on the left hand side here. That's when you look at something like a base compute or a 4x compute in this one specific example, right? But this 32x compute, it looks like real life. But again, think this is not just for allowing us all to generate hyper personalized realistic looking one minute videos. This model and as we use it, right? And as millions of people use it, right? When you use a model, people don't know this. Sometimes you get a split screen and it says which one's better? You know, there's always a thumbs up or a thumbs down, right? We are training these models. So we are telling OpenAI and we are telling Sora, this is how the world works. This is not correct. This is correct. And as millions of people use this technology, OpenAI gains a better understanding of the world beyond what models currently can comprehend. And it starts with more compute. We need more computes. Here's another thing to keep in mind. Again I'm speaking in generalities here, but generally current large language models, not just chat GPT but most large language models, they try to use as few tokens as possible while still giving a good answer, right? But it could use if it wanted to, right? And people, you know, people much smarter than me, you know, have examples of this online. You can get much better responses from current large language models if you can fine tune them. To not care about tokens, to not care about memory, right? So the reason why all the big companies do this, it makes sense, right? Because you have to balance cost with, with quality, right? But what I'm saying is current models and future models are obviously capable of much, much more what we get out of them on a daily basis, right? For for most of us, for everyday use, for everyday people, right? I'm not talking about, you know, your LLM hackers out there and your machine learning experts. But for the majority of us, what we see out of large language models is a balance that does not show us the full capabilities of models. These models have to balance compute, they have to balance cost with quality. They're capable of obviously much, much more. So, you know, earlier we had a great comment from Nancy, right? She said something about, you know, tokens and cost. And yes, this will be, this will have to be a premium add on, I would assume, right, I would assume. But I also wouldn't be surprised if this does get released and OpenAI is just for a while taking a bath on this. It's going to be very expensive even if we have to pay $20 a month additionally to use this. And we get, you know, a couple generations an hour, maybe we get one or two. I would assume that OpenAI will still be losing money on this. I will assume that they will not care early on because they want us all to use it. They want us all to help train their model. You know, we might get two different variations and you might choose which one's better. They want to see those thumbs up and thumbs down. They need real world humans to train a model that is for the future, for real world AGI. Right. We have to talk about compute for today versus compute for tomorrow. Have any of. Right. Like we talk about on the show all the time on everyday AI, we always talk about, oh, you know, more, more companies. Nvidia and you know, Microsoft creating their own chips. Amazon creating their own chips. You know, Sam Altman trying to raise $7 trillion, which is an asinine amount. Right. Why? Because all of us right now, are we suffering from this lack of compute power? No, we're not. We can go on and use these models, right? Pretty much, yeah. There's limits, there's throttling, there's capsules, but we can use them essentially fairly well. We're not crippled by today's lack of compute. This is for tomorrow, this is for AGI. Right. There's a reason the two leading voices in the push for AGI, obviously it's Sam Altman and OpenAI so far ahead of everyone else in terms of. They've been on the AGI kind of bandwagon for longer. And recently, you know, Mark Zuckerberg has become more vocal. So, you know, you have Zuckerberg and Meta being very vocal and saying, yeah, we're investing billions of dollars in chips, right? And here Sam Altman, who says, okay, I see your billions. I say 7 trillion or joking around on Twitter. 8 trillion is compute for tomorrow. So let's combine all of that. I know we're still on point two here. We're going to wrap up. Zero point three is pretty straightforward. But combine all of that, everything that we just talked about with AGI and a lot of people don't understand what Open AI can currently do with its different products. Right. You can't use them all in the same interface. A lot of them you can ready. I'm going to run them down. You have GPT vision, so OpenAI can see, you have GPT voice, OpenAI can talk, you have whisper, OpenAI can listen jukebox, which they've been sitting on for years. And it's actually pretty impressive. OpenAI can make music and sing. Data analysis. OpenAI can write code and understand data. Just the GPT4 technology. OpenAI can understand that is your general use case. Then you have Sora, where it can understand relationships in the real world now. And then in the blue, you have two things in the future, right, which I would presume are in the works. You have agents which can perform tasks and you have that $7 trillion of compute, which is essentially unlimited resources, right? Yeah. That could take many months or multiple years to achieve. Are you getting it now? Are you understanding what's happening? Yes. Sora is text, video. But Open AI told us, OpenAI told us just most people don't want to read. This is a step toward AGI, y'. All. You not see it receipts on the board. So right now, OpenAI can see, it can talk, it can listen and understand voices, it can sing and create music, it can write code and analyze data. It can actually understand in future models will be better with human reasoning. It is starting to understand relationships and it will soon reportedly be able to perform tasks like a human. You wanted the hot take. This is a mild take, right? Because I don't want to be, you know, too much hyperbole. I don't want to be, you know, too much sensationalized click, click, click bait. The writing is literally on my wall right here. And they are telling us, they are telling us if you bother to read the research paper and to look at past whatever is going viral on Twitter or LinkedIn, you will see this is about AGI referenced this earlier. So it's important to, you know this, this Google DeepMind a couple months ago. And also if you don't know much about Google DeepMind, I'd say they are by far the leading AI research team in the world. All right? So they released a couple months ago this little chart. I'm not going to spend a lot of time on it. I'm going to give you a high level overview. But essentially it's different levels of AI and different levels of AGI. Right. So you have narrow, which are certain tasks. Right. And right now AI can already perform, when you look at specific tasks, better than any human. Right. You have your level two, which is, you know, average. You have your level three, which is expert, let's just say 90th percentile. And then you have your level four, your virtuoso, which is 99, which is better than almost anyone. So on narrow tasks, individual tasks, AI already wipes the floor. Right. And there's already all these studies. It's already been done. Right. General is a different story. That's what all these big companies are working toward. AI, artificial general intelligence, AGI, that's just better at everything than humans. Are we there yet? Maybe, maybe not. Probably we're close. Will we know when we get there? Not necessarily. Could be happening before our very eyes. People who follow AI and have been following it for much longer than I have, and AGI say it's, it's just going to happen. Right. There's not going to be a warning all of a sudden. It's going to be, oh, okay, yeah, we have AGI. Right. So level one emerging AGI, it's already there. Right. So that's. Is AI better than essentially unskilled humans? Right. There's not a nice way to say that. But, you know, people who are unskilled, uneducated. Is AI in general better than those people? Yes. But you don't really start talking about AGI until you get to level 2, which is competent, which is when at general tasks, AI is better than the average human. Are we there today? No. Could we be there soon? Yes. You know, these predictions go back and look five years ago, five years ago, I'll try to find these studies. I have so many studies floating around in my head. Five years ago, I believe they said, oh, we'll have AGI by 2060. All right, in 2019. They said 2060. You know, they said we were decades out. Today. They're saying, oh, it could be a year, could be a year less, you know, because all of these, all of this new technology in this race right now, people did not five years ago understand how important generative AI would be to the US economy. They did not understand that every single of the largest, like, companies in the US are Investing billions of dollars into AI. We did not foresee that five years ago. Most people did not. That is why the developments are coming way faster than even the smartest researchers five, six, seven years ago could have ever predicted. They said decades. And then three years ago they said maybe a decade. And now today people are saying, ah, it might be a year. It could be as quick as a year. Enough on AGI, let's get to point 3. I don't think anyone will catch OpenAI and Microsoft. It'll take what I'm calling a double acquisition to get close. And if you did listen to my 2024 build predictions, which, hey, you never would have, you never would have thought this a show from two months ago. So many of them have already come true. People were like, no, Jordan, these won't come true. Anyways, I said two or three months ago, I said there is going to be a very large acquisition in 2024. An acquisition that most people aren't expecting. And here's kind of the rationale or the reasoning behind it. Companies now, like now, what you see with Sora and when the general public understands that Sora is about more than video, it's about more than text to video. Once the rest of the world and the tech world starts to understand that, you're going to see pressure. Big companies and their stocks, once they're the analysts and the investors and the general public understands what this means from OpenAI and Microsoft. Again, Microsoft owns 49 of Open AI, so you got to talk about them in tandem. Once the rest of the world catches up, whether that's in two weeks, two months or two years, an acquisition is the only way out for these other companies, period. Whether that happens, like I said tomorrow, who knows, maybe an acquisition is close, maybe it's going to still be a couple months, but a big acquisition is going to happen. There is no other way around it in 2024. All right, so I'm going to categorize these. We have our tech titans and what I call startups that can burst or startups that can ignite. I will say highly flammable startups, right? Startups that once they combine with the tech titan, they can do something big. So our tech titans, we essentially have Amazon Meta Alphabet, which is Google, Apple and Nvidia. Right? So aside from Microsoft, I'd say, and you know, you could throw Tesla in there as well, but they're kind of competing in a different space. So you have Amazon Meta, Google, Apple, Nvidia. All right, and then you have your startups that can burst or your startups that can ignite. You have Anthropic, you have Mid Journey, you have cohere, you have, you have stability AI, you have hugging face, you have Runway, you have pika. Yeah, there's probably one or two more you might be able to throw in there. But I'm not talking, you know, companies that are just like rappers, like GPT rappers. Not talking about this. Yeah, I know there's companies that are valued at billions of dollars that are essentially GPT rappers. I'm talking about companies that have unique technology that they've built in house. Right. So anthropic, mid journey, cohere, stability AI, hugging face, Runway, pika. Yeah, there's probably one or two more. It is going to take a. Either two tech titans combining forces. You know, I think you also have to throw in probably, you know, IBM in there, you know, some of those more that are in hardware as well. Right. But then I think they're, you're gonna have to acquire multiple. Right. If you're Amazon, you might have to acquire Anthropic and Mid Journey. They've already invested heavily into anthropic. Right. If you're meta, I would, I would keep my eyes on meta. Right. Meta, I think from a multi modality standpoint is the one closest to where open AI will be soon, if that makes sense. Right. Meta might have to acquire know as an example, a hugging face and a Runway. Or you know, Apple might have to acquire a pika and a stability AI, I don't know. But it is going to take either multiple tech titans combining forces, which I don't see that happening. You know, I should have thrown IBM and some others in the mix here. Or it will take one of these tech titans acquiring multiple igniting startups to compete with the combination of OpenAI and Microsoft. They are so far, they are so far ahead of everyone else. Right. Nvidia not, you know, I guess Nvidia is also kind of, it's in its own category, right? Because yeah, they just released their chat with rtx. But technically, you know, most of these big tech titans are clients or they pay Nvidia for their chips. So I know Nvidia is kind of on the outside, Tesla is kind of on the outside, IBM is kind of on the outside. But you know, even specifically if we're looking at Amazon, meta, Google, Apple, they're going to have to acquire multiple of these companies I think to compete because geez, look at Gemini 1.5. I'm not super impressed. Look at these. You know these video models, right? Even that Meta and Google and Google have have previewed. Oh, gosh, I feel bad for those very smart people who have spent a lot of time building what are very impressive models. And then you see Sora, right? It's in its own hemisphere. It is in its own stratosphere. It is not close. I want to hear from you. That's all I got for today. It is about so much more to recap, right? It is about so much more than just text to video. You have to look at the bigger picture. All right, Number one, Soar is light years ahead of everyone else with text to video. So if you're just looking at it at that at face value, number two, the timing. You got to look at the timing. That has to mean AGI is much closer than we think. And number three, I don't think anyone right now without doing something drastic can catch OpenAI in Microsoft. I hope this was helpful, y'. All. We spend so much time doing this. People always ask me, hey, Jordan, how can we support Everyday AI? Well, we're going to be officially launching some consulting services soon, but right now, hey, if this was helpful, please share this episode. You know, if you're listening on on LinkedIn, you know, repost it. If you're on Twitter, retweeted or re exit or whatever that's called, share this with friends. Tag them in the comments here. You know, text them, talk to them about it. We are trying to be your best friend in AI. We are trying to cut through the noise, cut through the Billy boys. They're just trying to make a buck off you. They're just trying to lead you down the wrong road. We bring receipts. We bring you the facts. Yeah, every Tuesday we, we spice in some hot takes and we give you some opinions. But I think that that generative AI education is essential for all of us to grow our companies and to grow our careers. I'd appreciate it if you let others know and join us tomorrow. Your voice and your context. How to scale a content engine with AI. I'm excited for tomorrow's conversation. I'm excited. Thank you for joining us. Make sure to go to your everyday AI dot com. This is going to be a big newsletter, so make sure you check it out. Thank you for joining us. We'll see you back tomorrow and every day for more Everyday AI. Thanks y'. All. And that's a wrap for today's edition of Everyday AI. Thanks for joining us. If you enjoyed this episode, please subscribe and leave us a rating. It helps keep us going for a little more AI magic. Visit your everyday AI.com and sign up to our daily newsletter so you don't get left behind. Go break some barriers and we'll see you next time.