OpenAI’s Video Capabilities EXPOSED (Sora Leak)

I don't know if you heard about this one yet, but Sora got leaked. This actually probably benefit OpenAI more than anything, which I find really, really fascinating.

Yeah, it's interesting to see, like, OpenAI really go about product development differently.

Yeah, it looked the most realistic and consistent to me out of anything I've seen.

And so if they were better in from the Turbo model, that's quite impressive.

Hey, welcome to the Next Wave podcast. I'm Matt Wolf. I'm here with Nathan Lanz and. And there has been a lot of craziness in the AI world over the last few weeks. A lot of big things happening. And in this episode, we're gonna just break it all down for you, share our thoughts and opinions, and let's just go ahead and dive right into it.

Using only 20% of your business data is like dating someone who only texts emojis. First of all, that's annoying, and second, you're missing a lot of context. But that's how most businesses operate today, using only 20% of their data. Unless you have HubSpot, where all the emails, call logs and chat messages turn into insights to grow your business, because all that data makes all the difference. Learn more@HubSpot.com.

We're going to set the stage and tell you a little bit about what's been going on in the world of AI over the last couple of weeks. Last year, around November, AI news just sort of died off. Right when you get around the holidays, you don't really get a ton of news because even all the AI companies sort of take the time off as well. The biggest news that came out of November of 2023 was Sam Altman getting booted from OpenAI and then a few days later coming back to OpenAI. But other than that, there was really nothing going on in November around the world of AI. That doesn't feel like the case this year. In November, it seems like there's really been not much of a slowdown at all. Like, there's a lot to talk about. We wanted to start with a very interesting thing that happened with OpenAI and Sora, because Sora got leaked, and I use leaked in air quotes because I don't really feel like it got leaked. Like, somebody made a Python script, essentially, that linked to Sora's API and a bunch of people were able to basically generate videos really quickly while it was online for, like, the two hours that it was online. So basically a group of, like, artists that got early access to Sora, at least according to this, like, Write up that they did here. They got early access to Sora, but they were frustrated by the fact that OpenAI was telling them what videos they could share with the world, what videos they couldn't share with the world, and they felt like OpenAI was taking advantage of them. Right. They basically said, you gave us access to this stuff, you let us play with it, but if we want to publish a video, it's only videos that OpenAI essentially cherry picks and lets us put out. We're here testing it for you, we're red teaming it for you, we're doing all this work for you for free, and all we're getting in return is access to Sora, which we can't share anything we make without your permission first, therefore, that's why we're leaking it. But again, the leak was really just like they went on hugging face. They put up a Python file. The Python file had an API endpoint that pointed to Sora, which means people were able to go on hugging face for like a two or three hour window, generate videos with Sora and you know, kind of share them around. But then pretty quickly, because they were just sharing an API endpoint, OpenAI was able to go and shut down that endpoint and nobody was able to generate videos anymore. So the loophole was closed pretty, pretty quickly. But in my opinion, this had like the reverse effect of what these people that leaked it were going for. Right. Because they, they leaked it, it didn't stay online for very long. So not many people got access to play with Sora. But everybody's talking about Sora again, right? So it's like this actually probably benefit OpenAI more than anything.

Actually I saw a lot more negative sentiment towards the leaker, you know, than I would expect. I kind of expected to be kind of like both sides kind of stuff, at least on X I didn't see that it was all like pretty much one sided. Like this is kind of petty, like they, they give you access to this really cool tool which you could have personally and professionally benefited from by having early access. And you decided to do what? Like do give them the middle finger and like illegally leak the videos.

Yeah, and they were all under NDA too, so they all broke. So whoever leaked this broke NDA to do it. And then basically OpenAI shut down the Sora like creator program. So all of the creators that had access to Sora that were able to use it behind the scenes all lost access. Nobody. It's like, all right, since you, since somebody leaked it, we don't know the exact source of the leak. We're closing it for everybody. Nobody gets to use Sora anymore. Right. So it's like all of the, all of the creators that got access behind the scenes, most of them were frustrated that this happened because they lost access to Sora for themselves. And yeah, it does feel kind of petty. I mean, you got access to this tool. I think more of what the negative sentiment around it was that OpenAI was trying to super strongly control the narrative of like, what gets shared. So, like all the videos you're seeing are only the cherry picked videos. And, and this leak kind of showed that Sora has a lot of the same flaws that some of the other video generators have. Right. But at the end of the day, we're on a podcast talking about Sora and we hadn't talked about Sora in months because it's been irrelevant. Well, now all of a sudden SORA is relevant again. So.

Oh yeah, that was a conspiracy theory. It's like they did it on purpose. But I'm like, I just don't really believe that. I don't see why they would do that. Maybe you believe that. I don't know.

No, I do not believe that this was OpenAI, like leaking Sora to get people talking about it again. That wouldn't really make any sen. I also think, you know, OpenAI probably lost a little bit of money. I mean, in the grand scheme of things, probably not too much money, but like, you know, making an open, like putting this API endpoint on the web where anybody could use it for hugging face. I don't know what it cost to generate one of these videos, but people were able to do a whole bunch of them for free in this small window that it was open. Right. So it's like, yeah, it definitely had a slight negative impact.

Yeah. One thing worth noting is I think someone showed from the API calls that it was actually a model, a turbo model, which likely means that it was a slightly lower quality, faster model. So I would say that if that's the case, all the videos that I saw, the examples, they were not dramatically better than the models out there that I've seen, like from Runway and others, but they were better and so they were better. And the Turbo from the Turbo model, that's quite impressive.

I mean, I think when it comes to people, I think it is probably the best model for generating people like walking. It looked the most realistic and consistent to me out of anything I've seen, you know, that already exists out there. So I, and I agree that the Turbo model. Supposedly it's a lot faster to generate. It can only generate up to like 1 minute videos total. Maybe it was only 10 seconds, I don't know. But it had like a shorter generation time than like the main Sora model. But yeah, at the end of the day I don't think it was actually smart by the leaker.

It.

It pissed off more people than like then people actually liked it. I think the only people that are really happy about it were like the, you know, the AI art community that is super anti AI, right? There's been a few new announcements that came out of Anthropic. This week they introduced their new model Context protocol. Says today we're open sourcing the model Context Protocol, a new standard for connecting AI assistants to the systems where data lives, including content repositories, business tools and and development environments. It aims to help frontier models produce better, more relevant responses. So kind of what I'm taking away from this is that it's sort of like a enterprise feature, I guess, or like a feature to use like internally in your company to sort of connect it to your data sources so that Claude has that context of your data sources connected to it. Is that kind of what you're taking away from this as well?

I haven't had time to like really deep dive into it yet, but like from, from reading on Reddit and like X on places, the big takeaway seems to be that yeah, you could kind of do this stuff before, but you need to create a lot of custom software to connect the LLM to your private databases and information. And it sounds like now it's, it's dramatically easier because it kind of just takes care of that for you.

Yeah, I mean, honestly, this is a little out of my wheelhouse. It's not something that like I personally use or know a whole lot about. You know, I tend to play more on the creativity side and the like large language model side of like, you know, getting it to generate stuff that's useful for me. I don't really have any extra like data stores I'm trying to connect to right now. So it's not something I've personally tested. There is another announcement that came out of Anthropic, which I have tried, so I can talk a little bit more intelligently about that. They just rolled out a new feature where you can create your own personal style. And so what this is is if you go into Claude, there's a dropdown that says Choose Style. And under the Choose Style, you have the ability to create a Custom style. And the way you create a custom style is by uploading a whole bunch of text that you've written. So if you have written articles or written a book or have blog posts or, or even if you take transcripts from your videos or something like that, you can load all of this text in there and it learns your speaking style, your typing style, whatever you want to call it. It basically learns on that text to try to duplicate your style. So now when you go to Claude and you give it a prompt, you can actually select, make it sound like me essentially. And the response it'll give you will be sort of in the style of how you would write, or if you put transcripts, it would be sort of in the style of how you'd talk. Right. I played with it. It's really easy to use. It's pretty cool. But I didn't really feel like it sounded like me when I had it generate. Like I, I gave it three of my YouTube video transcripts. So I basically gave it 90 minutes worth of transcripts to try to learn how I talk. And what it gave back to me had a whole bunch of like emojis integrated in it. It would. And it said stuff like, what's up, guys? I got something killer for you today that you're gonna think is really spicy. And then it had like a rocket emoji and it was like, what? Like this? Like there was no emojis in the transcripts that I uploaded. Why are there emojis in the style that I got back? But the cool thing about it is you can actually sort of continue to fine tune it. So when you, when it gives you a style back, you can go and click edit style. And there's a little chat window just like if you're using a custom GPT where you could give it feedback. So I gave it the feedback. Like I never put emojis in my text. And like it's a little more casual than I actually speak. I speak casually, but this is too casual. And it actually sort of tweaks it and tries to get it even closer and closer to your voice and. But right out of the box. The first time I did it, it was not amazing, but you can actually sort of continue to fine tune it a little bit.

Yeah, it's interesting to see like Anthropic and OpenAI really go about product development differently. Recently Anthropic's been releasing things faster, but some of them are kind of half baked, like the computer use thing and whatnot. Right. Where it's like Cool. You did it first. But people already heard that OpenAI was building something similar for a while and they haven't released it. So it's interesting that they are releasing cool things that don't always purply work. And it does seem that OpenAI is kind of taking the path of like, be more like Apple where you try to wait until things are very good. Same thing. They're doing a Sora, they got a great AI video model, but they're waiting to release it when they feel like it's actually the right time, when it's actually super useful for people.

Yeah, it makes me wonder if they like, if they're just kind of taking two different business approaches. Right. Like Anthropic is like let's put it out before or it's ready but collect a ton of feedback. Let's find out what people like about it, what people don't like about it. Like, I mean that's been a practice a lot of. I mean Microsoft is actually pretty known for doing that. Right. Like the first version of the Xbox they came out with like was didn't work very well and they waited for feedback and then like made better versions based on feedback. So I mean like the big tech companies have been doing this kind of thing forever. Like let's put something out that we know is not ready yet, but let's see what people say about it.

I ended up meeting Kevin Bacchus, but from Microsoft because of the Xbox One thing, because I sent him emails about the Red Ring or whatever. I got that like two times in a row. I was living in Florida and when you buy the Xbox and every time you'd buy it, like it get hot one day and then the thing would just permanently die.

Yeah, I agree too. But like I feel like it's slightly different with software versus hardware. Right? Like if you buy an Xbox One and it's like broken out of the box, you're pissed off. Right? Because it's not like, like you probably have to replace it if it's like Red Ring of Death. A firmware update is not going to fix that because you can't boot it up to get the firmware update installed. Right. A software thing like Anthropic can push out a new feature and know that it's not fully what they want it to be, but then pretty quickly push out new updates to get it up to where people want it. Right?

Yeah, but both approaches make sense for their current state states. Right. Like OpenAI is the leader, they have more eyes on them if you release something bad, people could in theory not come back or just have a horrible experience. And it's like, I'm never going to try that again.

Yeah.

Whereas anthropics in catch up mode to anything they can do for attention is smart.

I agree. Yeah. I think that's kind of the thing right now is that I feel like the narrative among AI people, right? The people that are sort of paying attention and talking about AI all the time, the sort of narrative is that like, OpenAI shares a lot of cool stuff, but never really ships. Right. And I think Anthropic might be seeing that going, all right, let's be the company that ships. If we really want to compete with OpenAI.

The feature sounds awesome. I mean, like, the idea to personalize, I want to try. One thing that's really been annoying me with Claude is that, you know, I type really fast and I used to be way more accurate. I feel like as I've gotten older, I still type incredibly fast, but I do make more errors. And I liked with ChatGPT that I could still type incredibly fast. And if there's a typo, who cares? Because you always just. It would know what you. You meant. And I love that. That's like a magical thing for me because, like, type super fast, make a mistake, who cares?

And you feel like, Claude doesn't do that.

Claude corrects me and I find it super annoying. I'm like, I did not want. It's like, I don't want this thing called Claude, like correcting me. Like, oh, you, by the way you type. You know, you spelled it this way or whatever. It's like, whatever.

But Google's been doing that forever, right? Like, did you mean. And then.

Yeah, but OpenAg does not do that, right? So I just like. Or ChatGPT does not do that. So I think that's actually been a really annoying thing for me with Cloud. Like, literally when that happens, I'm like, screw this program.

I don't want to use this. I've actually never noticed that. I don't know if I've ever seen it correct me like that. But moving along, I know V0 has some updates. In fact, the last episode we put out, we actually played around with V0 a little bit. But I don't know much about the update. But you mentioned before we hit record that V0 has some new updates. So.

And this shows you how fast things are moving. AI like, probably 30 minutes after we recorded our episode, they released a new update to V0 that everyone's been talking about. And it, you know, we showed in the last episode that you could clone a website pretty easily with V0, but it wasn't perfect. There's a lot of things that were different. You know, it was probably like a 70% match or something like that. I would say they're like in the realm of like a 95% match now, which just, you know, like, for anyone who's creating a new website now, you could go to V0 and show it a UI that you like or design. It's also, I think you could just.

Give a URL too. Can't you just say, I want a site that looks like this URL and it'll actually look at the URL for you?

Well, what it's doing is it looks at the URL and it takes a screenshot.

Yeah.

And then it puts that into the system to the LLM or whatever. And so, and so that's what they're doing. So like you could do it. If you did a design, then obviously the same thing would work. You design it, you put in there and it would give you the basic code for that design as a website. So like unlocks so many new opportunities in terms. Well, you know, you could have people copying people, which, you know, I'm not really a big fan of, but that, that will definitely be happening. So if you're in business, be aware that people will now be able to just literally type in your URL and say, I want that, give me that. And I don't, I don't have to pay anyone for it. I don't have to hire developer or anything. I just want that. And you'll get like 95% of the way there. The other big thing that, and the design is pretty good. I would say V0 has got the best UI and design out of any of them.

Yeah, I agree with that.

Yeah, yeah, I mean, it makes sense. Like the founder I've actually met in know him, Guillermo Ross. He's a great guy. And, and he's really design focused. You know, he's always been like a huge fan of like Apple and he loves Steve Jobs and so it makes sense that he's super UI and design focused. So they definitely have the best design out of any of the like code generators with AI. But also apparently it got way better on the back end. Like apparently now when it not only is creating the landing page for you, but if you want some kind of back end. I don't think it does everything yet, but apparently it makes it way easier to connect it to something right now. And you could even. It'll. It'll help you to generate a database as well. And you could connect it yourself and it does things even like, like, you know, secret keys and stuff like that. If, you know, if you have like a, like a, like a password for a database, you can even in store the environment variables and stuff like that, which is the kind of thing where when you're coding, you might want something in there, but you don't. You don't. You don't put the password in the code. You would have like a, like a secret that it handles all that for you now and helps you do it.

Yeah. I wonder how many times people just put an API key directly into their code and then throw that software on GitHub and now everybody has your API key. It happens pretty often, I'm pretty sure.

Yeah. I mean, so they, if they keep improving this fast, I mean, it seems like it's totally possible that within, like, three months you're going to be able to, like, basic SaaS apps, probably copy most of their, their features, including the back end and everything, and the ui. And so that's a. It's a new world to be in. I mean, you'll definitely see. You know, there used to be those guys. I forgot the guy's name, the two brothers in Russia. But there used to be these two guys who were notorious. Like, Silicon Valley, hated them because they literally, like, they just watched, like, tech. They'd read TechCrunch and like, any cool startup that was, like, seemed to be getting traction, they just copy them.

I remember hearing about that story a while ago.

I forgot. I forgot their names, but they were like, notorious. And so you'll probably say this, like, even like, at a larger scale, right, because you'll, you'll be able to copy anything and put it in another language or whatever.

Yeah, yeah. I mean, I don't really see anything wrong with, like, using another website for inspiration. Right. Like, use it as a starting point, but then, like, completely overhaul it. Right? Like, okay. I really like the way that, you know, TechCrunches blog looks. Make it look like that. Okay, now I'm gonna go change the colors. Now I'm gonna go change the font a little bit. I'm gonna go. And next thing you know, it. You know, it has similarities, but it's not the same website anymore. Like, I feel like that's what it was designed for, but I don't think that's what it's gonna be used for.

Yeah, but I mean, the reality is Though, like, web designers have been doing that for a long time, right? It's like the. You, you, you pay them a ton of money and then like, you actually hang out with designers and see what they do. And a lot of it is they got a bunch of samples of things they like, they put it into a figma and then they copy little bits and pieces and edit some stuff, and then that's kind of what they do, right?

I've hired a whole bunch of designers myself over the years, right? And they always send you a questionnaire, and one of the questions on the questionnaire is like, give me two or three websites that you really, really like so that we could, like, you know, use that as a starting point, right? So they're always asking you, like, what sites do you want your site to be inspired by? So, I mean.

Yeah. So yes, on the positive side, like, if you're a company or an individual looking to create like a new project to test it out especially, look, you know, we're both like creators with decent followings. I think, like, creator's in a great spot because if you want to test out a concept and then share it and just see what happens, it's gonna be easier than ever. It's gonna be easier than even like framer and stuff like that. Because you'll. At some point you'll be able to do like, AI voice too, right? Just like, just chatting with you, like, oh, I wanna make something like this. I'm like, oh, cool, it's done. Five minutes later, tweet it out, see if people are interested. They're not. Who cares? You know, it's like people forget a week later anyways. So it's gonna, it's gonna really change how people create new products, I think.

Yeah, for sure. I don't know if you heard about this one yet or not, but, like, Uber is getting into the game of AI labeling, which, you know, is what scale AI does, right? So basically they're trying to start a like, side hustle kind of thing where anybody can go and look at a whole bunch of AI generated images and then label them or go read a whole bunch of dialogues from chat transcripts and, you know, rank them of, like, how good it did. And you can actually get paid by Uber to actually go and do that. So Uber's getting into the, like, data labeling business as a side hustle, gig economy thing, which I find really, really fascinating.

Yeah, yeah. Well, I mean, I think Uber is in a very, you know, odd position. You know, they, they haven't been really innovative in a long time. You know, they, they fired Travis, the founder, who. Very controversial figure. I think I told you, I met him before and had a pretty negative interaction with him. But he did do an amazing job of leading the company. And ever since he's been gone, like, they, they haven't really done that many new innovative things. And now with, you know, the rise of like Tesla and you know you're gonna have like, you know, AI taxis, you know, driverless taxis are, are coming. Like, they're definitely coming. Especially now with like the new Elon Trump buddy part. You know, whatever, whatever's going on there. Like, they're definitely coming now. And so since that's happening now, if I was Uber, I would be kind of terrified, honestly, because they're obviously behind in AI compared to Tesla, like, probably like dramatically behind. And so this may be an effort to kind of do two things at once. There are a few things. I mean, one is you're like kind of hedging your bets. So if the, if one of your business units fall off, well, you've got this new business unit. Another thing is you're building up technical talent for AI, which they're going to need for driverless. And then at the same time, backup plan, maybe Tesla or someone buys you, you made yourself a more attractive acquisition target because you have a better AI unit and something that could be useful for them. So that's what I think is going on.

Well, I also think on the flip side of it, this sort of like, side hustle, make money side of kind of pushes a little towards like a theory I've talked about in the past of like, I don't know if the government is going to offer like a UBI kind of thing in the future. I think what we might see, at least in like a nearer term future, is like, a lot of these companies are going to figure out ways to do this, like, gig work where people are making money doing stuff like this. So if you're like out of your accountancy job because AI does accountant work now, well, now maybe you're making money by, you know, driving people around until the cars just do it for you or labeling AI data for some of these companies. Right. I feel like that might be where that's like the sort of next, like, economical shift is a lot more like people taking on a whole bunch of like gig work to make their money versus having like a traditional career path. And I feel like this is sort of like a step in that direction, in my opinion.

Yeah, that makes, that makes Sense. You know, it is interesting that Elon Musk has been a proponent of ubi, but now he's really, you know, more on the Republican side, which is very, like, sees UBI as like a socialist concept that you have to avoid. I think the reality is gonna be somewhere in the middle, probably, like, what people actually have to do where, like you probably eventually with AI, when things become more abundant, it probably will make sense for there to be something where people's basic living expenses are covered in some way. But then maybe you do also have a gig economy on top of that, where there's like actual extra incentives. Like, okay, if you don't want to just like, sit around, play video games.

Yeah.

Like, here's an extra incentive to go off and actually do something productive. Right. Yeah.

Well, I mean, even saying all that though, like, the. The sort of conservative, like the Republican Party. Almost all of the top people in the Republican Party were at one time Democrats. Right. So it's just like a sort of like, shifting of that sort of line.

Yeah, yeah. It's just a label. And the label that what it means is.

Has changed. Yeah, for sure. You know, and. And speaking of Elon Musk, this is another news story that just came out is that the rumor has it he wants to create an app to go head to head with Chat GPT. Now, obviously we've got GROK inside of X and you can open up the X app on your phone and use GROK directly inside of your phone. But it sounds like he wants to make a standalone XAI app that is like basically ChatGPT, but using the XAI technology and, you know, supposedly unbiased and all the kind of stuff that he stands for.

Yeah, I mean, we talked about this a little bit off camera, but like, you know, my take on it is, you know, he wanted X to be the Everything app, which he clearly stated was because he saw WeChat in China and he realized that WeChat in China is an app where you do everything. There's games in there, there's shopping, there's social media, there's your banking. It's all in one. I think one thing that he probably didn't really realize is, like, the way users interact with apps in China, in America and Europe and everything, it's all very different. And like in Japan, websites here, like, there's some websites that are so there's so much data on the website that like, America be like, what is this horribly ugly website? Yeah, this is horrible. What is all this nonsense on the. Really, there's too much stuff. It's. It's. I can't. I don't know what to focus on. For whatever reason, people in Japan love that. They love there to be ton information dense. You know, Joey Ito from mit, he's talked a lot about this. Like, Japanese people, whatever reason, they love UIs, where things are information dense, tons of information. And I think that's the kind of problem you run into is like most Americans, for whatever reason, you know, with X, they're going to think of it doing one thing you tweet on there. I think he's going to have a hard time of convincing people that it's a YouTube too. Yeah, right. Yeah, because it's like, no, they've got one thing in mind.

Well, I mean, he's trying to convince people that it's. It's YouTube, it's the next PayPal, the next payment platform, it's the next ChatGPT. It's the, you know, the list goes on and on and on. Like literally the everything app. Like, he wants you to just live on X eventually.

Yeah, yeah. He's like, I said he was modeling off of, you know, China's WeChat, and people are different, have different things that they want. I don't think it'll work in America.

Yeah, yeah.

I think that's a tough seller. Yeah. As a standalone, maybe it could work, like, if he makes it cool, especially if he died, like, I could see it working because, like, like we said in the past, Grok is really good at doing, like, funny images and stuff, stuff that, like, the other ones won't let you do. So, like, if he does a good standalone app that now is all of a sudden the best one at generating, you know, funny AI art. Right. Like memes or whatever, I could see that being super successful.

Yeah, yeah. And I mean, right now, Grok, it uses the Flux 1 Pro, which is probably the best one out there at generating funny memes and, you know, realistic images. So.

Yep, yep.

So lots of interesting stuff going on in the world of AI right now. If you like stuff like this, make sure you're subscribed to this channel on YouTube or Spotify or Apple Podcasts. Wherever you tune into podcasts, make sure you're subscribed and we'll keep on putting out podcast episodes like this for you. Thank you so much for tuning in. We'll see you in the next one Sa.