AI NEWS : OpenAI Drops "Blueberry Model?" Metas Stunnign New AI Voice, Sora 2 and more

So there have been many stories this week in AI, and I'm sure because there has been so many different stories, you may have missed a few things.

So let's dive into today's stories which.

Are going to give you an update to what you all missed. So one of the stories that I actually nearly missed was this story about OpenAI revamping Sora AI video. This is an article basically talking about the future development of the app and I think it's rather fascinating because we as individuals haven't actually gotten access to Sora yet, whilst the competition from other models like cling 1.5 Runway lumalabs have intensified. So you can see here that this is the information, a reliable source for AI news leaks especially. And it says between the long awaited Strawberry launch and its monster ongoing fundraising, it might be easy to forget about Sora, a video generating model it teased in February to global fanfare. A number of investors told me at the time that the impressive Sora demos cast a worrying shadow over the gaggle of competing AI video startups that had already raised hundreds of millions in venture capital. These concerns may have eased six months later. OpenAI still hasn't launched Sora, but they've said that that could soon change. OpenAI is training a new version of Sora, so I'm guessing that this is going to be Sora too, that in hopes it will quickly generate clips that are higher in quality, longer than the ones it demonstrated earlier in the year, according to a person who's spoken to OpenAI executives about it. So what's crazy about this is that we didn't get access to Sora and I think there were a variety of different reasons why this happened there.

Number one, Sora wasn't particularly received well.

By the global community. Most people, I would say, like on that were, you know, vocal about Sora actually rejected the release. I spoke to many creatives, I spoke to some individuals and they all pretty much just didn't want this tool. So if those individuals would, I guess you could say, reject the tool.

And considering the fact that OpenAI was.

Going through a period where they are.

Pretty much struggling for compute, the computer.

Was so bad that they didn't even give any to their Super Alignment team, which they pledged that they were going to, and the team ended up disbanding. By compute, I basically mean GPUs for inference for many different things. And anytime you use services like ChatGPT to generate audio to generate images, you know, it uses their inference, it uses a compute, which is pretty taxing considering the fact that they are trying to grow their company and also train new models at the same time. So it seems that what OpenAI I think they're trying to do here is I think they're trying to, like, once again, leapfrog the competition. Because when Sora was released, it was unheard of for AI Video to be that good. But I think they're trying to get to the next stage now where they just completely demolish the competition in terms of what is possible. So you can see here that they are speaking about training Sora, which is where the company collects millions of hours of video for training data. And they said that these videos need to be high resolution and contain a diverse array of styles and subjects. And it says, but why has OpenAI taken so long to make any progress on Sora? And what were the issues that prevented it being released as a product in the first place? Of course, one of the things I did forget was that, you know, the Sora announcement, one of the things that was an issue was that it did actually take really long to generate clips. It took around, you know, more than 10 minutes to generate a short clip of a minute or so. And the crazy thing about this is that if you've actually seen Runway gen 3, it is so quick. In fact, I could probably generate an image right now. In fact, I'm going to show you guys how quick it is because it's genuinely outstanding. So if you literally go to Runway. Okay, so let me just say nighttime city cyberpunk, okay? And I just want to show you guys how quick this is in real time. And the reason I'm showing you guys this is because this is going to show you guys what OpenAI is competing with. So if I put a nighttime city cyberpunk camera pans around cinematic drone shot, and then I'm just going to put. I'm just going to do this. So let's do generate. Okay. Why is. Do I need to add an image? Okay, for example, I'm going to go ahead and use this image and I just want to show you guys how quick this is. So this is a 10 second one. So what I'm going to do is I'm going to set this as the first frame and I'm going to click Generate. So you're literally going to see in real time how quickly this generates. So this is, of course, their quicker version, but I'm not going to skip the video at all. It's now in queue. I think it's only been around 10 seconds. But, guys, look at this. Look at how quick this video is being generated. This is why I'm saying that, like, of course, you know, you could argue that SORA has higher quality, yada, yada, yada. And, you know, of course, OpenAI is more known, but you have to understand that this was, you know, 10, 20 seconds, okay, from start to finish, okay, maybe 30 seconds now that it hasn't been done yet. Okay? So this was 30 seconds, and I'm not sure about the quality, but let's just take a look. So this was 20 to 30 seconds. But the crazy thing about this is that this was incredibly fast. And of course, with better prompting, this video could have been made better. But the thing is, Sora took 10 minutes, okay? And that's not. I wouldn't want to say it's not feasible because, you know, 10 minutes is a lot quicker than going out and filming all that stuff. But if you're trying to, you know, offer a video project, you can't really, really just be offering it at, you know, 10 minutes to generate a short clip of about a minute or so. I don't think that that's going to be something that's remarkably effective. So you can see right here, for instance, one of the filmmakers, Patrick Sederberg, said he had to ask the model to generate hundreds of clips before landing a usable ones. In other words, SORA was arduous to use, basically saying that, look, whilst it's fun, I had to generate so many different times. And this is actually quite true, you know, when you're experimenting with certain AI softwares, it sounds good in theory, but a lot of the times, like, even when I'm messing around with AI voices, you know, you have to generate them, like, few times in order to get, like, a good output that you can genuinely use, which is kind of frustrating. So I don't know if they're gonna, like, have something that generates 100 clips and then the verifier, like, goes and checks for the best one that would even take up more inference and compute. But, yeah, this, this kind of thing has to be useful.

And so I think this part of the video was rather important because, of course, you know, future, the improved Sora model is going to be fixing many of these issues. Soar is one of the biggest things that I think, you know, brought AI into the, you know, immediate mainstream because.

It was just a tool that everybody.

Immediately knew what it was. It wasn't just like text, it was.

Like, okay, text to video. This is insane.

So I think that this is pretty Crazy, but I think Sora is like.

A back end product.

I think it's going to be, you know, for movie studios and stuff. But like I said before, other companies have caught up. Runway, maybe not Pika just yet, but Cling AI are striking deals with these, you know, major companies to get stuff.

That they can actually use on a day to day basis.

And with that kind of speed that you just saw from Runway, it's going to be really difficult for OpenAI to compete. But in some other AI news, there was also the Meta Connect. The Meta connect. I got to be honest, this was like really, really incredible. I just was blown away by what Meta has done because llama 3.2 was overshadowed by a few other things going on at OpenAI such as the advanced voice mode and a few other, you know, notable AI developments such as the.

Drama going on at OpenAI.

But Llama 3.2 was a great release because this 70B model actually now includes vision in it and it's actually good vision and it's an open source vision, which means that we're going to have developers that also do some crazy things. So of course we did have, you know, 1B and 3B models that were like on device that were really, really effective. And I think this just goes to show like what is being done in the future in terms of what we're going to be able to get with future AI systems that are going to be really, really effective. I can't wait till certain phones have these models just natively on device when you're able to use them for certain things, just making your phone a lot more effective, like doing certain agentic capabilities as we're seeing in this demo here. I think this is certainly where the future of AI is, provided that these developers manage to take advantage of them and we create an ecosystem where people can build natively on these phones. Now unfortunately, what Meta do have, and I say unfortunately because it's only unfortunate because I live in a location which is highly regulated. This kind of thing isn't available to me yet. So there is this AI voice and it was actually was demoed recently and I'm going to show you that in a moment. But this is kind of like their version of chatgpts. Of course it's not as great, but it's still remarkably effective considering the user base Meta have with their distribution. You have to understand that these guys have like, I think threads has 100 million customers already in terms of, you know, users that have signed up to the platform, Instagram and WhatsApp already have over a billion people. So with that kind of distribution, that's.

Literally like 1/7 of the world's population.

Or 1 8, which is insane.

When is the next total solar eclipse?

The next total solar Eclipse is on August 12, 2026 and will be visible from Greenland, Iceland, Spain, Russia and a small area of Portugal is we are.

Experiment experimenting with automatic video dubbing on reels starting with English and Spanish. So this is going to be a way that you can see more of the content that's out there, no matter what language you speak. And as a creator, this is going to let you reach a lot more people around the world, no matter what language you speak. So now you're going to be able to take your content and it's going to be in your authentic voice, speaking other languages if you want. And we're also doing automatic lip syncing, so it actually looks like you are speaking the other language. So let's check this one out.

With coconut, banana and dulce de. This recipe was one of the first ones I uploaded to social media a little while ago. I rediscovered it and made a couple of modifications.

Now, I think this right here is largely going to be one of the most underrated things in AI, because what we have is like, I wouldn't say it's a unification of many different countries, although some people would see it as that. But I think it's going to be really interesting to see how different cultures connect now that we have AI, because what's going to happen over the next five to 10 years is that AI will be able to translate content easily. There was even a speech, I think it was at some kind of event. I genuinely don't remember the event, but I remember that the translated version got so much more support than the original spoken one because it got translated to English and a lot more people could understand exactly what was going on. And it didn't just sound like, you know, voiceover, it sounded like that person speaking. And I think we're going to be seeing, like, for those of you guys who are like me and you speak English, you're going to be able to see a lot more different concentrators from smaller countries popping up, you know, going completely viral because of this area. So if you're someone who, you know, maybe English isn't your first language and you speak a different language, I would say taking advantage of translating your content to other languages and that broader reach is going to really, really benefit you. And definitely for businesses, definitely for people who are trying to, you know, explore how they can increase their revenue beyond just current Nation. This is going to be something that I think is really, really effective and really underrated because having that translation, I mean, you know, there's so much great content out there that, you know, you could consume and you could enjoy that this kind of, you know, technology enables us to do.

So I think it's going to be, you know, incredible, literally incredible.

But it's very underrated at this current moment in time. And YouTube's going to be doing a similar thing as well. So maybe you'll hear me speak Spanish or French, if that is your native language. I'll take a look at some of more of Meta's announcements. I was even supposed to put these in the original video, but I was just ill at the time, so I wasn't able to do as much as I could. So do apologize for that. But take a look at some more of these announcements because Meta, I think they're probably going to be one of the key industry players that is taking a completely different approach to many of these other companies. Like, while these other companies are focusing on the foundation models, I think Meta is just focusing on user distribution, which I think is probably going to be more important in the future.

So now Meta AI can natively understand images as well as text. Now we're using this new capability to build some features that I haven't seen anyone else build. I mean, this is pretty interesting and novel stuff. Creative tools that we're adding to Imagine Edit. It lets you upload any photo and edit it precisely with natural language in Meta AI across the apps. My family loves this. We spend a lot of time taking photos and making them more ridiculous. You know, one of my daughters actually just wrote a book and we used Med and she used it to illustrate the book. It's just a lot of fun and awesome.

All right.

But moving on to another thing. This one, I think is actually the biggest news that we have today. We are adding natural voice interactions to Meta AI and you don't have to wait a while. We are rolling this out today and over over the next few days. I think that voice is going to be a way more natural way of interacting with AI than text. And probably I think it has the potential to be one of, if not the most frequent ways that we all interact with AI. It is just a lot better. So now you're going to be able to have natural voice conversations with Med AI across all the major apps you got, you know, Instagram, WhatsApp, Messenger, Facebook. We want to make this fun. All right, so in addition to the different system voices that we built in that I think are pretty good, we've also worked with some of the most iconic voices out there to bring them to meta AI, like John Cena, Judi Dench, Kristen Bell, Keegan, Michael Key, and Awkwafina. Should we try it out? All right. Hey, are live demos risky?

Live demos can be risky, yes. They can be unpredictable, prone to technical issues, and potentially embarrassing. However, they can also be engaging, memorable.

But I'm going to do it anyway.

Go for it. Just be prepared for anything to happen. Fingers crossed.

I mean, live demos can be risky, but you can see there, Mark Zuckerberg pulled that one off. And honestly, you just have to feel bad for Google because, like, every time.

They'Ve done a live demo, it's just filled spectacularly. I don't know if the universe is.

Conspiring against them, but yeah, I think.

That just goes to show, like, you know how people are saying that OpenAI.

Have this massive lead. They are really under so much pressure now to really get stuff through the door.

And.

And in another video, I actually spoke about how OpenAI staffers, the safety researchers, are having to work 20 hours per day during tight deadlines in order to get stuff out the door. Because, of course, it's pretty much the last hurdle to pushing these products out there. So, I mean, when we look at meta AI now having their real time voiceover, we had Google have theirs and now literally just last week, we finally had OpenAI rolls theirs out. Of course, we are in this arms race in terms of AI productivity now with OpenAI.

There was really something that I wanted to show you guys, because this is.

Something that I think is changing in.

Terms of how people view AI systems.

And the reason I say this is because if you remember the early days.

Of AI, and I say early days.

When I say early days, really I'm just talking about like two to three years ago.

But there was this researcher from Google.

Blake Lemoine, I'm pretty sure most of.

You guys heard about him, but I'm going to give you guys the too long, didn't read version.

Basically, this guy, Blake Lemoine, he basically was a Google engineer and, and he thought that the AI had come to life. Now, I'm not going to be someone who says it has or says it hasn't. I think these AIs are black boxes.

There's a lot about humans we don't.

Understand anyways, so it's anybody's best guess.

You know, Some people are like, these things are just chatbots. A number of people, like, okay, but.

Like, how do you know that? I mean, we're just simulating certain things in our head. And basically the long story short is that this guy went viral.

There were so many people saying that.

This guy's, you know, a moron for.

Thinking that these chatbots are sentient.

But recently OpenAI have ushered that kind of statement that it may be that these neural networks are slightly conscious. And the reason I said this is so fascinating is because this article was from 2022 when it really went viral. This was even before ChatGPT, which was really fascinating.

But let's take a look at what.

They'Re saying about OpenAI 01 now 01 models.

When we trained it and we actually started talking to it, we started asking it these questions and it was scoring higher on these math tests we were giving it. We could look at how it was reasoning and you could just see that it started to question itself and have really interesting reflection. And that was a moment for me where I was like, wow, like, we've uncovered something different. This is going to be something new. And it was just like one of these coming together moments that was really powerful. So when you read the thoughts, does it feel like you're watching a human or does it feel like you're watching a robot?

It's like a spiritual experience.

It's a spiritual experience.

Experience, but then you can empathize with them all. You're like, oh, that's a mistake that a lot of people would make. Or you can see it sort of questioning common conventions. And yeah, it's, it's spiritual, but like, oddly human in its behavior.

So that brings the question, do you guys think we're going to come full.

Circle on these models and we're going to have people starting to say that, yes, these models are completely conscious. They have feelings.

I mean, there's this entire philosophical debate and I think, think I don't really know. It's one of those things that's kind of confusing because when we start to question what these models are capable of doing, we have to always then kind of question ourselves.

Because one of the things I see.

Time and time again is that, like, oh, these models aren't reasoning, they're not doing this. But it's like so many times these.

Models make mistakes that humans make, and.

So many times these models literally improve.

The same way a human would improve. Like, for example, if you say, show you're working, this is exactly how a human will improve. Like if you give an LLM time to plan, this is exactly how a human would improve their ability to perform any kind of task.

I mean, the whole thing is just.

For like an entire video. I did a video on this a while back. I'll leave a link to that.

But let me know what you guys.

Think, if you think these models are conscious or not. Because I think the debate either way, both of them make sense, genuinely. Because if we're saying that they're not, kind of makes sense. And if we're saying that they are.

Kind of makes sense too. So with that being said, let me.

Know what you think about that. Now, there's also this artificial analysis text to image leaderboard, and basically this leaderboard is where we can see the text to image models in terms of their capabilities. And recently there has been this secret model that has come out of absolutely nowhere. And this model is called Blubri. Guys, this is rather fascinating. Now, at first glance, most people are speculating that this Blueberry model could be referenced to OpenAI Strawberry, but potentially an image generation model that just simply surpasses everything before. I personally wouldn't be surprised if this was just a play on OpenAI Strawberry like a Blueberry model. And they released something that's really cool because SORA does have incredible image generation capabilities. But I think it's surprising that I remember only recently that we got Lux 1.1 Pro that was just so further ahead of ChatGPT and pretty much anything else that it was quite surprising, especially considering the fact that it immediately went open source and then now only, you know, one or two months later we have something called Blubria that is quite anonymous. I mean, it could be Claude, it could be Google. I mean, there are just a variety of different things that we haven't seen yet from these major companies. So I'm truly wondering what company this is. Some people think it's Sora. I think it probably might be OpenAI, considering the fact that they haven't really explored their image generation capabilities and we know they are really good. So I do think this is most likely going to be OpenAI. But let me know what you think about this super secret model, because I think we're probably in for something big if this one already is a high jump amongst the others.