OpenAI Just Killed Sora

Sora was one of those models where when I'm making stuff, it's one of those, I don't think about you at all [laughs] models. The, the Jon Hamm thing in the elevator? The Jon Hamm- Yeah... Mad Men meme.

I don't think about, yeah. Yes. Like, can we edit that meme into this video? [static] All right.

Welcome back to Denoised. How's it going, Addy? I'm back in town. Let's do it.

Back in town. Yeah, you're on a trip in AI Land, right? [laughs] Yes. [laughs] Yeah.

All right. Speaking of AI Land, big model fallen. Sora. RIP Sora.

Yeah. 2024 to 2026. [laughs] Goodbye, Sora. Adios.

Yeah. All right. So, OpenAI announced that they are shutting down the Sora app. Mm-hmm.

But I believe it is also the whole model, the API access, not further developing it. Big news in the AI video world. Though, I don't know, I'm, I, I'm not, not too heartbroken over this one. Sora was one of those models where when I'm making stuff, it's one of those, I don't think about you at all [laughs] models.

The, the Jon Hamm thing in the elevator? The Jon Hamm- Yeah... Mad Men meme. I don't think about, yeah.

Yes. Like, so you and I used the Sora app heavily for, like, those two, three weeks. Not even he- I wouldn't even call it heavily. I, I did.

Like everyone- I-... we got the Sora app, it was a fun novelty, and then, uh, then you kinda stopped using it, right? I think there was- It was a fun joke for a little bit... huge potential.

I mean, I'm not, like, mad about it. I'm kinda sad about it because, um, other companies could have followed suit and figured out a, a genuine TikTok competitor, like, at that scale of billions of, uh, videos and all that stuff. I think the m- move that was, um, the fatal flaw of Sora, the app, was they had the licensed IP stuff in there the first couple of days, right? And you remember my video of the, like, me in my Mandalorian stage.

It was, it was unlicensed IP. [laughs] Unlic- Right, right. It was the Wild West- Unlicensed... for like the first 48 hours.

Yes. And that's when I generated my best stuff. And then as the IP stuff started to get taken away, it started to get more boring and boring, and it's like, okay, how many times am I gonna just generate myself? It's only funny to me, you know?

Like, I wanna... [laughs] Well, then it- And, and you. Like, you were my buddy on it. You're like, "I guess that's, Addy's kinda funny here.

Okay." And, uh, I think the approach they should have taken was what Fortnite did with the seasons. So, you know, when Fortnite drops, like, a Star Wars season, they, they pay Disney for all that stuff. They ingest it into Fortnite, and it's freaking fun for the kids, right?

Like, they get to be Darth Vader and stuff like that. In the same vein of thought, Sora should have done the same. They should have paid for IP. Now, I know it's, uh, it's a difficult foreign concept for AI companies to pay for IP, but this is standard practice for media companies, and I, I think they would have had more longevity there, and for sure more engagement and less of a falloff as the app got older.

I think there's other stuff at play here. So I mean, their, from their own admission statement, they're trying to cut down on distractions, and Sora is a distraction. We talked about this before, like, M&E, the film industry, is a, like, big for us, very tiny, insignificant little sliver for these companies. They're more interested in how can the large language models be an enterprise and be useful and be part of, uh, uh, all these companies', uh, everyday lives, robotics, world models, autonomous vehicles.

Making funny meme videos and then potentially also making Hollywood level animations, tiny market. Good for, good for publicity, tiny market. But the social media app and the platform, that's a big market. Potentially, but that's such a huge lift to then try to go and try to build a TikTok competitor.

I think they had that potential route with the Disney deal, uh, where Disney was, like, investing a billion bucks. They were gonna use the Sora tech and kind of build this Disney Plus, you know, interact with Disney characters. That whole deal is dead as well, so that whole, they're gonna, Disney's getting their money back and whatever was gonna come of that deal is gone. Maybe Disney will do something with another model or build their own thing, which is totally doable and plausible.

They don't really need Sora for that. But I think that- Mm-hmm... that route was potentially there. I think just as a bigger OpenAI strategy, I think they're a little, they were doing, like, a lot of things in a lot of areas, I think getting a little bit lost.

Claude has caught up tremendously, and Claude has kind of taken over in the coding space. Sure. And OpenAI is now putting their resources towards Codex and building these coding tools- Mm-hmm... because that's where, like, people will pay and then companies will pay, and they actually get productivity improvements.

And given the resources- Yeah, the enterprise revenue is in, um, software automation- Yeah... and things like that. Yeah, and I think it's just, like, a resource issue of, like, you have these limited GPUs, and where do you train and how do you, like, allocate that? And obviously video and running these models, these video models, is intensive.

And are you getting returns on that? Yeah. No, probably not. Better to focus on, like, practical uses.

Fair enough. And I, I think Sora, the video model, like for example, the Comfy API node, the Sora node, it's one of, I don't know, 15 that you can use today. Yeah, and that's what I'm saying, when it's like a model picker choice, like- Yeah... sure, there was the Sora app, and Sora was, like, good for these kind of character meme-y things, but, like, the quality wasn't there.

You know, so if I'm trying to generate something just, like, from a basic video generation model, I'm usually defaulting to, like, Kling or Veo- Veo... or, you know, then maybe Luma or Runway or something. Like Sora's kind of like- C Dance too nowadays, yeah. Let's see, uh, uh, yeah, Sora, C Dance, and yeah, we can talk about them in a second.

Yeah. But, um, yeah, like Sora was very bottom of, it was probably like in the Croc category of like [laughs] models that you would try. Alt models. Yeah.

[laughs] Yeah, alt models I would, like, get to for what, uh, for what, you know, we're trying to do here. They sell Sora at Hot Topic. It's alt. [laughs] Bad joke.

I, you know, I think there's two other things. Like for one, I don't fully know what was under the hood and, you know, we're not AI scientists, but, like, you look at Veo and Google, and it's very clear, like, everything-That they have been doing around their entire AI ecosystem is somewhat unified and somewhat part of like a large world model. You know, VEO, Genie, Gemini, they're all like- Yeah... you know, in this process of like a big, large world model.

I don't know if Sora is there. Yeah, and, and quality bar wise, like all the Google products are hitting the same level of quality across the board. Mm-hmm. And you're like, "Huh, how, how is it so consistent across the board?"

So there's must be some grand architecture that we're not aware of [chuckles]. Yeah. And it's like, I don't know if Sora is there, if there was any way that like Sora, besides being a video generation model, has world understanding, is a world model, could integrate with chat, can like integrate with all those things, and then potentially lead to more lucrative product lines. Like, was the plan for Sora ever to be a world model that could potentially work for robotics or, uh, vehi- autonomous vehicles?

Like, it's pretty clear that Google's models and what Luma's trying to do and all these other models, I don't know. I don't know enough about what's under the hood, but that could also just be like it's just they're - like what's the future for where can Sora go? Yeah. Like Google to me seems more like NVIDIA, where they're trying to catch all industries, and they have like a big grand plan.

Uh, whereas OpenAI, I think they - to me at least, the perception is that they're super pursuing AGI, ASI, and then like ChatGPT and Codex and all this stuff is just kind of there to make revenue for today. But like their main goal is to just make AGI. Yeah. I mean, yeah, we've kind of forgot about that too.

Yeah, the AGI. It's the [chuckles] still- Yeah... the, the goal. [chuckles] It's nuclear fusion.

Right. It's always twenty years away. [laughs] Yeah. Yeah, less so with the Sora videos.

Yeah. You know, the other thing I was wondering like, you know, Sora had that deal or to make Critters, which was supposed to premiere at Cannes, be that- That's right... kinda Pixar level animated film for thirty million dollars. Don't know what the deal with that is, but I'm guessing either maybe they finished that up or not, or [chuckles] - It's probably dead, my guess.

I would agree. Yeah, if you, if we haven't heard of it, yeah. Yeah, I mean, if the film got far enough along, might as well finish it. I would still be shocked if like Sora was the one and only model used in that.

Like, I - like it's gotta be a, a combination of models and methodologies. Yeah. Also, like, uh, side note here, the, uh, image... Is it called ImageGen?

Their image generator. I think now Image, ImageGen was Google's original name. One point five. It's now just like- Oh, okay...

ChatGPT image, which is still- ChatGPT image generator. Yeah... uh, that's not, that's untouched. It hasn't gotten better.

Like, they just kept it at the, the quality level from like twelve months ago, right? Uh, ancient. [chuckles] It's ancient in our standards, yeah. No, they did one point five, which I think came out end of last year.

So like- Okay... an update, but nothing as earth-shattering as like every Nana Banana update has been. Yeah. I think OpenAI is sort of, um, they see Anthropic as their main competitor, obviously, like Dario and [chuckles] Sam Altman hate each other.

And, uh, so now they're sort of shaping their company to be more one-on-one with Anthropic. Like if you look at Anthropic, they don't have an image or video model. They're not messing around with M&E. They're focusing on AGI, ASI, and code development.

Coding. And yeah, and OpenAI is like, "Well, that, that works." [chuckles] More of a sort of agentic OpenClot-esque features- Exactly. Exactly...

uh, in their app. I know you're not on as X as much, but like Claude has been shipping a, like, new feature roughly every day, and everyone's like, "What the hell?" Like there's just like it's- Right... they're just on a roll of just like some new feature.

Like even yesterday, the new one was, um, like, uh, C-Claude App can connect to like other apps, and you sort of have title cards in your app, which is sort of this like kinda making Claude this, you know, your central hub- Like an OS... to connect everything. Yeah. And, you know, and, and, and something that's consumer friendly and easy to use.

So y-yeah, I think they're looking at Claude as sort of just been much more focused on what they're trying to do, and- Mm-hmm... OpenAI has been a little bit more all over the place, and I think it's now like, "Hey, we need to like focus on what it is we're trying to do." Ex-exactly. Yeah.

I think they're heavily recalibrating, and it's a good time for them to do so, right? They have a new round of funding coming in, the hundred billion dollars or whatever- Yeah... got funded. So I mean, Disney's one billion is like, "Okay, take...

We're fine without it." And they're gonna just pursue this new direction. I mean, like we, [chuckles] we crap on Sam Altman a lot here. If OpenAI is not a viable business model, then that, that does pop the AI bubble, and it'll kind of have a en-enormous blast radius across the board for all of the AI stuff we're doing.

Like everything we're touching, you know, Kling or Quinn or whatever image model, like everybody will be impacted by the OpenAI collapse. Yeah. I mean, I, I, I don't feel like they're close to collapsing. Do you feel like they're- Yeah.

It is on the news a lot. [chuckles] I mean, I mean, there's like, you know, talk of an AI bubble. But yeah, I, I... It doesn't, it...

I haven't felt that way about... Like, just like a, like a, that there would be an overall OpenAI collapse. Right. So yeah, I think the- We'll soundbite that for when they, uh-...

pivot and adjustment-... when they collapse, like [chuckles] in- Well, we can go back and grab our, grab our soundbites from the Disney-OpenAI deal that they, you know, that was a few months ago. Like, we were super, like, wondering what the hell that was about, right? I thought- Uh, some parts of it made sense, parts of it were confusing or, like, didn't make sense.

Yeah. Like putting the stuff in the Disney+ app or, like, putting some user UGC stuff in the, in the, on the Disney streaming platform, that was like, "Is that gonna work? How do they pay it out?" Yeah.

"How would they..." You know? But, like, something where it's like, "Hey, I wanna, like, make a video of myself with, uh, you know, Mickey Mouse, like, driving, uh, you know, with Steamboat Willie- Steamboat... and driving the boat."

Driving the steamboat? You know, and it's like, okay, that kinda makes sense, you know- Yeah... from just a, like, meme-y, fun, kitschy way to make a silly video with- Sure... characters you like.

I thought, I mean, from day one that OpenAI should have paid Disney one billion instead because this is the most profitable entertainment company in the world, and they have all the cool IP, and you just kinda li- do a big chunky license for one billion dollars, and then you do you. And yeah, I mean, I think, I think I've talked about my conspiracy theory. You know, I, I wasPretty sure that, like, OpenAI's plan when they released Sora and had, like, zero IP restrictions was to more of like a showcase to be like, "Look what we could do."

And then they were like, "Uh, sorry, we'll roll it back." But then they signed the deal with Disney because it was sort of like, "Look what we, what we can do." And I've been wondering if the C Dance rollout was kind of the same playbook where it was like, "Look what we could do. Oh, sorry, sorry, we'll roll it back.

Uh, sorry, but we could do this." Wink, wink. Yeah. For media companies to be in bed with a single AI manufacturer is, is not ideal right now.

There are just way too many people competing at the same level. I don't think there's any advantage. It's like why- Right. Disney's smart.

Uh, Disney can... You know, if, if they're not... I mean, I'm sure they're trying to build out their own models. But if- Oh, please no...

you know, they don't need to rely on Sora when they can just plug and play with other models and build out their own app and not have to share the branding with anyone. Like Sora doesn't have that strong of a brand that there was like a boost from being like, "You can make AI, AI videos powered by Sora." You know, Disney's the bigger brand. Exactly.

And their, and their IP characters. Yeah, also speaking of focus with Blocked, but, um- I was just laughing at the GIF of, uh, Javier Bardem from F1. [laughs] Yeah. Everyone, uh, the internet is very happy that Sora has, uh, has fallen.

[laughs] But, um, they're also, uh, putting their erotic chatbot plans on hold indefinitely. Oh, good. [laughs] So... Yeah.

I mean, for me, that was just like such a big waste of environmental resources. Distractions. Yeah, that is just so much... Like what, what is...

It doesn't sound healthy- Yeah... or useful for anyone to... [laughs] It's bad optics for the entire industry. Oh, my last thought on this was just more of like looking back at the long life of Sora.

It is a bit crazy how much they fumbled it, though, because if we- Mm-hmm... remember back a million years ago, two years ago, when Sora was first teased with the clips and the spaceman and the desert and, like, the girl on the train, and these were, like, mind-blowing images. Like, these were, like, the most realistic, like, best-looking things we had seen. And then they, like, didn't do anything with it for months, like half a year at least, before they, like- Right...

actually released it, and by the time they released it, Kling, Veo, Luma, like, everyone had caught up in quality and had publicly available tools that by the time they came out, it was like, all right, Sora, that thing that exists. And then I, yeah, I, I, I forgot about it. And it was, like, heavily gated and expensive. Yeah, like the whole rollout.

Yeah, you had to be on the pro plan. I mean, they built that kind of tool thing that had some interesting ideas, but, you know, now everything else has it and does it better. So for how much of a splash they caused two years ago, uh, and then seeing it shuttered today, kinda crazy. 'Cause they were, they were like, you know, early movers in that space.

Yeah. Goes to show how effective their marketing team is. Okay. Okay.

Next up, sad news. Bummer news stuff. Yeah. Sad news.

All right, two shutdown, uh, or just layoff-y announcements. Uh, the first one, you sent this to me. Yeah. Uh, Epic Games is cutting over one thousand jobs.

That's a lot of people. That's a lot of people, man. Epic is not that big of a company. Yeah.

Now, they're saying it seems it's all related to Fortnite and Fortnite growth and revenue slowing down. Yeah. Do you wanna read- What's your take on it?... Tim Sweeney's email on this?

Uh, I, I think- Yeah, there's a blog post... um, you know, h-his email to his employees, uh, was k-kind of frank and kind of honest, and I thought, you know, it showed leadership there. It, uh... But, you know, the reality is this sucks for a lot of people, a lot of people I know.

Yeah. Some of the challenges we're facing are industry-wide challenges, lower growth, weaker spending, and tougher cost economics, current consoles selling less than last generations, uh, and games competing for time against other increasingly engaging forms of entertainment. Which is ironic because, like, Netflix is also, remember, maybe not so much recently, but I remember a while ago, they were like, "Yeah, our competition is just- YouTube... leisure time."

And it's YouTube now, but I remember, like, a few years ago, it was gaming. Oh. Like- Yes... when they're competing for time of, like, watch time or leisure time, it, it wasn't against other platforms, it was against, like, gaming and, and, and other...

Now it's YouTube, but before it was gaming. Since it's a thing now, I should note that the layoffs aren't related to AI. To the extent it improves productivity, we want to have as many awesome developers developing great content and tech as we can. So again, uh, he wanted to decouple this layoff from sort of the Amazon layoffs and all the other big tech layoffs that have happened and has, like, the AI- Yeah...

stink, stink on it, if you will. I mean, I honestly, when I saw the headline at first, I was wondering if that was the reason, if it was gonna be like, "We're, we're using AI, and we have more efficiencies. We don't need these thousand developers or people." Mm-hmm.

But yeah, no, he clarified it's specific to Fortnite and just that growth and revenue has slowed down there, and that they need- Yeah... these, these layoffs. And, uh, just wanna kind of add that this is after the seven hundred person layoff that happened, I'm gonna say 2024, right? You remember that?

Mm-hmm. Okay. Kinda. And that was also along the lines of, um, we need to focus the company around Fortnite.

Yes. So we're gonna kinda... Y-you know, that's when a lot of, like, the, the VP talent w-was let go, you know, 'cause U-Unreal was building nDisplay very heavily and all the LED volume stuff. And then we thought that, okay, that adjustment happened, so now Epic's in a really good shape for the next few years, and then now this.

So Fortnite has really probably taken a hit revenue wise, and, um, you know, Tim Sweeney, to me, doesn't feel like one of the regular big tech CEOs. Like he, he's no Sam Altman, you know? Like he's not- Yeah... Andy Jassy or one of those guys.

He comes off as like a regular guy, a regular- Mm-hmm... software developer who happens to just be a multibillionaire and a CEO of a really cool company. Right. So, so I think for him to just make this decision is probably hard, and especially 'cause they're a private company, they can kind of do what they want.

They don't have to answer to shareholders. So for them to do something this drastic meant that there is significant internal trouble. Yeah. Do you think doing that year, multi-year long-Apple battle became a distraction or sucked resources- Absolutely.

I mean, yeah. That- Resources or money... that stuff takes money, it takes, uh, Tim's attention away from other things. But I, I think it was like a David and Goliath thing.

It was a battle of principles over- Yeah... anything else, over revenue, and I kinda like that he fought it and he won. Like it was- No, I did like too, and he was like- Yeah... he was like, "We're the only ones that have resources to fight this, and it's, like, crushing smaller developers- Right...

but no one's gonna be able to- Yeah... fight Apple except us." The 30% thing was ridiculous, right? Yeah.

But, um, what Epic needs now, in my opinion, is, uh, they need another Fortnite. They need another massive multiplayer, multiverse, uh, multiverse [chuckles] metaverse type-y- [laughs]... game. And, um, you know, I know they have had other sort of mild hits along those lines, but if, if, like, Fortnite revenue falls off, you augment it with revenue from an incoming game.

You know, unfortunately, you can't just make a hit up. A hit happens when a hit happens. Yeah, that's... Yeah.

And we know that way too well in the movie in... Yeah. Yeah. And, uh, you know, that seems like a...

It is tied to an age group, and then they age out, and I know you're always trying to get new people in and, like- Yeah... keep it cool, but that, that's tough. [laughs] Well, that's, that's, uh, it's, it's funny. That's what Roblox is, um- Yeah...

like, their current strategy is how to retain their users who started playing Roblox when they were 10, and now they're 20, right? Mm-hmm. And they're starting to fall off. Obviously, it's, uh, like, when you're a 20-year-old, you're not gonna be playing Roblox.

So a lot of what's happening is they're trying to build things inside of Roblox that retains an older audience. One of those things is just having better graphics. You know? Like, uh, you don't wanna play- [laughs]...

mine block, you know, blocky, choppy stuff when you're 20 years old. You want, like, hyper-real stuff like Call of Duty or Grand Theft Auto. So that's one. The other one is, um, how about, you know, meeting your partner, some sort of dating mechanism inside the community.

So there, there's a lot of development around that. I don't know. I, I don't know enough about Fortnite to know if that stuff was happening, but, um, that's how you retain the audience for decades. Yeah.

I don't know either, and I don't know if there was, like, a drop in, 'cause I remember always hearing those crazy stats of, um, like, digital clothing and items that were, like, a billion dollars a year alone in just buying- Yeah... wardrobe. I think I told you that. Yeah.

The, the avatar, uh, the avatar department which makes Roblox, you know, accessories like hair, shoes, clothes, that department's profit alone is one billion a year. Yeah. Which is crazy. [laughs] Crazy.

And, you know, especially if real-world expenses are jacking up. [laughs] Yeah. It's more expensive to, like, just exist and live life in the real world. Digital clothing would probably want to be one of the first things to, uh, not spend as much money on if- Yeah...

your real expenses are, are, are, are increasing. The other weakness that Fortnite has that Roblox doesn't is Roblox runs on any, literally any device. Like, a crappy Android phone can run Roblox, right? Yeah.

Right. Um, with Fortnite, you know, you need a nice PC or a console or something, but it sure as hell is not gonna run on a crappy Android phone. I don't think it is. I mean, correct me if I'm wrong.

And that, that takes away a huge amount of, uh, potential users, right? Like, as, like, the, the quote-unquote gamers, like the hardcore gamers with the NVIDIA GPU and stuff like that, as they sort of get bored of it, I mean, you're not picking up new audience. Yeah. I mean, I feel like it's another sideways related thing to the ideas of the metaverse slowing down or...

Yeah. The, the other big play that Epic did back a couple years ago after, uh, you know, around the time when they did the first round of layoffs was, uh, UEFN, Unreal Engine Fortnite. So it's like a custom version- Okay... of the engine that c- you can build your own islands.

Like builds... For people to build. A little bit more accessible Unreal for people to build. Yeah.

So it's, it's kind of like a equivalent of Roblox Studio where you use Roblox Studio to build- Uh-huh... build your own level and your game. Uh, UEFN is, is that. Um, the challenge was, I think, uh, it was incredibly difficult to put a game together that's under two gigabytes or whatever the, uh, limit was, and it, it's not like two high school kids can do it.

You actually still needed, uh, a pretty solid game developer to come in and then just finish up the last bits. So that never really took off. Like, um, you know, obviously there- Maybe in the future, the solution is something like a cloud code for Fortnite or something that makes it easier- Right... where you don't have to be a developer to build out your world.

You can just kinda- Yeah... describe and build, and then that's where AI comes in under the hood to help polish and clean it up and make it usable. Yeah. And, and Apple and Epic are kinda similar in that [laughs] Okay.

Uh, stick with me on this one. Apple and Epic are- [laughs]... kinda similar in the way that they have, for the most part, have not adopted AI, right? Like- Oh...

you know, if, if Epic was early- In a big way. Yeah. Like, just in a, in a fundamentally changing kind of way. Um, if Epic had put AI...

Let's say AI blueprints or, like you said, AI- Mm... game development and all that stuff into the engine, perhaps maybe, maybe it would've helped. I don't know. I have to imagine stuff like that's coming.

And also, I mean, uh, Sweeney's been... I remember we didn't cover this. I think I had that link saved a- for a while, but we never talked about it, where he, uh... Was it...

Oh, Steam. Steam, the gaming platform, had wanted to enforce a rule that was like, "You have to mark if your game used AI in it or something." And he was like, "That's a stupid label." Uh, like, you know, every...

Like, if it was used to help develop, like, if that's gonna be, like, common for every game, and everything is gonna use AI in some part to just help speed up development, you know, like, where do you draw that label or the line for that? And so it was... I'm paraphrasing with that. I think it was something to that effect.

But- Yeah. It was, like, not made- It's definitely- It's like an organic label, like not made with AI label, something like that? Something like... And he was like, it was a stu- that's a stupid thing to, like, have to label.

Exactly. Uh, so, you know, they're definitely not anti-AI. Maybe it's just y- a matter of rolling it out into the tools. But I, I would have to imagine that there's some blueprintCode, co-sidekick thing or something coming to help- Yeah...

speed that up. Regardless of sort of what's going on in Epic, I mean, I... My heart goes out to the, the, the people were just, like, amazing. Every- Yeah...

Like, if you are working at Epic- Yeah... you are at a caliber, performance caliber, you know, personality caliber. Like, you're an amazing person if you're working there. I think a lot of those folks, I hope they can kinda transition their way into AI because- Mm...

you know, just me having b- you know, been on both sides, and now on the AI side, there is such a room for, uh, people that are creative technologists. Like, there's not enough of us in here, um, that, that understand what a frame is, what frame composition is, what, you know, contrast. [chuckles] Like, basic things, right? Yeah.

And, uh, having sort of that eye and that lens, but also understanding the technology. Um, I, I think if you got laid off out of the thousand people, I- you're probably one of them, and, you know, I'm, I'm rooting for you and hope you have a future in AI if you choose to. Yeah. I mean, I'm excited to see what they would develop 'cause I think you are right that they have the skill set of, like, what the quality we need to be, and just if turbocharged with AI, I think there could be some interesting things.

Yeah. For sure. Other [laughs] part two of the news. Oh, man.

Pixomondo. Yeah. Um, they're calling it winding down. Pixomondo, Sony is winding down Pixomondo, but yeah, they're basically shutting down the studios.

They're winding it down because a lot of the people are contractors working on projects, so they're finishing up the projects. And then it did say that a lot of people at Pixomondo would get rolled into Sony in other roles. So I mean, look, it sucks that, you know, Pixomondo, a VFX house, early pioneer in virtual production stuff is... It's still, honestly it's still fuzzy 'cause they, the brand is winding down.

They did say that a lot of the stuff's just gonna roll into Sony. You know, they had that crazy driving rig product, Akira, that we saw at CES- Oh, yeah. Yeah, yeah, yeah... last year.

Akira. That was in LA, and then they moved it to Vancouver, and it seems like the Vancouver stage is just gonna become a Sony thing. It's still fuzzy, but it sucks. I'm guessing all of their stages are coming down, like the one here in Culver City inside the Sony lot, which I spent some time in, as well as the Vancouver one.

And I'm sure they have... I, I think they built one in Saudi Arabia f-for a customer, if I'm not mistaken. Mm. So I'm not sure who's gonna support it.

You think they're all gonna come down, or maybe they'll just do less, have a few of them still up, and it'll just be under Sony or like, ah, it's- I mean, regardless, uh, so the Pixo news came first before the Epic news this last week, and when I heard that- Yeah... I was like, "Oh, that's, that's crazy," 'cause I was not expecting that. To be under the Sony umbrella and to have sort of the global weight and the resources that Sony has, right? This is one of the most amazing, biggest brands in the world.

I thought they were safe. You know? I was like, "Okay, they're gonna reinvent virtual production. They're gonna kinda take it into VP 3.

0 in the future." But then turns out, no, uh, this is the end for them. Uh, yeah. But I mean, if they're like, "Oh, we're gonna roll people into Sony and whatever," then what's the difference or what's the advantage of just, of getting rid of the Pixo brand and the company if you're just gonna downsize a bit but still kinda keep some of the core stuff and just roll it into the Sony group?

Like, why get- I think it's just, like, dismantling for parts and trying to get the most- Yeah... ROI you can, right? Like, you've already acquired this thing for however many millions of dollars, and now how do you minimize the damage, minimize the loss, um, with this, with this news? So one of the questions I have to ask you, tell me what you think.

Yeah. Did the Sony, and just, you know, hypothetically speaking here, do you think the Sony executives looked at everything that's happening with AI, and then they just looked at their VP business, was like, "Wait, this, these two things are in two different lanes"? That... I mean, if they did, that feels like a big misunderstanding of, like, what the two things do.

Yeah, what is this... Do you think it's like, "Oh, this is jumping the gun," 'cause it's like, "Well, look, AI is really good, and we don't need these big, expensive VFX houses, and we don't need stages"? Like, we're not at that point yet, and I don't... A-and it's not like one or the other.

Like, the two work in hybrid. Well, the, the reason I say that is because this, this was, like, a, like, a really scandalous kinda thing, and it was meant to catch headlines, like, like a few years back, I think back in '24. Tyler Perry was, um... You know, Tyler Perry has that immense campus in Atlanta, right?

Like- Yeah... it's an old military base- Yeah... turned into a studio lot. And, uh, within that, he was gonna build more stuff, including a volume, and that was right around the [chuckles] time Sora and some of the vid-video models were coming out.

And then, like, he put- Oh, that's right... he put everything on ice. He's like, "Wait a minute." That's right.

"I'm not building anything right now until I figure out what this thing is." Maybe, but, you know, we've talked about this. Like, I, uh, there's still uses for this, and, uh, look, I mean, dealing with AI stuff, it's basically a little, maybe a little, like, less people and faster than if you were sh-shooting, you know, like, Star Wars style episode one, two on a green screen and then compositing everything later. You still gotta deal with all the shots afterwards.

And so if you could find a middle ground like House of David where you have a wall, maybe a smaller wall, maybe you don't need a huge massive wall, or a very heavily AI-assisted VAD to speed up that process. There are hybrid worlds where, like, you know, these two could play hand in hand, and you kinda get, can get the best of both worlds. It's not one or the other at all. It als- it also depends on the projects- Mm-hmm...

and what you gotta do. And, you know, potentially projection as well. Yeah. You know, that could still also have- Yeah...

place and, and be an alternative to having to build up a big LED wall. Yeah. For sure. I, I think, uh, the, the pro- Right.

Yeah, I mean, if it went... If e-executives were in the Tyler Perry headspace, and it was just like, "Oh, these models." L- They, they saw the C Dance 2.0, like- Ooh...

fighting on the roof, and they're like, "What do we need all this stuff for?" No, I don't even think it's the finished final pixel stuff. I think if, if you look at Beoble and Switch X and some of the relighting, and you're like, "Wait, that's what a volume does, is it relights you, it puts the background in the right place, and da, da, da, da, da," and now there's a tool doing that. And ultimately, I think the production, uh, always wants to fix it in post anyway.

Like, that's, like, the natural gravity of things in production. Mm-hmm. As you know, you're in production all the time. You're like, "I, I think I'm gonna just spend more time in post than try to do this live here."

I mean, it was also the way we're able to cram, like, you know, we did four pages and stuff, like, in a day, and it was, we were able to do that 'cause we're just filming in a gray box, and we're just like, "All right. Move, move, move," and then we, like, put all the shots together later in post. Uh, butYeah. Post is, you know, less- Post is where it's still at...

people in- And I think AI, post-AI VFX is highly TBD right now. Like, we don't know what that's gonna be. But I bet that the executives are, are forecasting that that's where VP eventually kinda shifts to, is instead of, you know, real time, I see VFX, and all the pressure on the stage at that time, shift that into post, but then use AI tools there. Yeah.

Yeah, it could be. And it could also just be... 'Cause I'm thinking back to NAB last year and sort of what Sony was, Sony was demonstrating in their new... They had new products in virtual production.

They had, they had their own, uh, camera track. They had a new camera tracker that they called- Yeah. Yeah, that was, uh, Ocelus? Ocelus.

Yes. Yeah. But I think the vertical for most of those applications is broadcast and live, uh, live events and live performances. So, you know, that could just also be, uh, you know, Sony's a massive company, like, how they're thinking of virtual production, like, less so in film industry and more so in live broadcasts and, and, and live events.

Yeah. And, and, um, I don't think Pixomondo plays in the broadcast world, and there are established bonafide broadcast companies that have just owned it, like Ross, right, Grass Valley, and Ikegami, and, like, these big broadcast brands that are building the volumes and the tracking systems and all that. Mm-hmm. There's Pixotope who does a good job- Zero Density.

Yeah, Pixotope- Yeah... Zero Density. Yeah. Yeah.

Yeah, where they, like, really just kinda dialed in on the broadcasting space. And- Yeah... and that is still relevant [laughs] and you, like, need it. Yeah, and Pixo's differentiator was tier one film and television, right?

Like, they did- Mm-hmm... the highest end of what we offer as an industry, and, uh, that's just such a small market, unfortunately. Yeah. Going back to Sora.

[laughs] Oh, God. [laughs] Small market. Important for our life. Small market- Okay.

Do we have some good news next? 'Cause- All right. Let's talk about, like-... I'm getting bogged down, man...

yeah, let's talk about new AI stuff. Okay. Okay. So let me try to...

Well, I'm not even gonna try to explain [laughs] this paper, but let me, let me compare it to the meme. So Google came out with this new paper called TurboQuants: Redefining AI Efficiency with Extreme Compression. Nice. We've joked about Silicon Valley and [laughs] Pied Piper.

All the time. We wish for that to come back- So-... all the time... this paper is being described as basically, if you remember later, the later seasons, Pied Piper accidentally came out with the- Mid- middle out compression?...

perfect middle out algo. Yes. Yeah. [laughs] The compression, where in Silicon Valley, in the show land, it was a magical compression algorithm that would compress the video and make it super small without, like, losing any quality, and that was their, like, kinda golden ticket to their, their big breakthrough.

[laughs] So people are describing this paper as basically Google did the real-life Pied Piper middle out algo. Oh, that's great. So [laughs] in the explanation from Zhen Zhou, it's basically for LLMs, it was a way to compress... Obviously a big issue with LLMs, you have your context window, you have your memory.

Once it gets filled up and you're having this kind of conversation, if the context window gets a bunch of stuff in it, then it starts veering, it starts misremembering things, and then it'll start... It'll compress it if you fill it up, and you'll pick up your conversation. But in that compression, you'll be like, "Hey, what about that thing we were talking about?" It'll be like, "I don't know what you mean."

[laughs] It's, it's outside my context window. Yeah. [laughs] Right? It's like, what are you talk...

Yeah, yeah. Yeah. I don't even think it'll tell you that. It'll just be like, "What are you, what are you talking about?"

[laughs] I don't, like, I'm not aware. I told you two days ago. I have a dental appointment. Yeah, these are recurring conversations I have with OpenCall.

Yeah. [laughs] Um, [laughs] so, so this compression is sort of like a LLM compression, six times memory, 6X memory reduction, uh, up to 8X faster attention, and- That's huge... better- Those are big numbers... less segregate, de- degradation.

Okay. And I'm very simplifying [laughs] explaining what this is saying. Yeah. Why don't you do my job, Joey, and explain research papers?

So- See how hard that is? It is a pa- Yeah. Uh, [laughs] so it's basically a, yeah, research paper that can i- in theory take existing con- uh, LLM memory and context windows and compress it- Yeah... without losing quality so then you could fit more stuff into the same LLM.

My M&E brain instantly goes to how we can apply this. A lot of the newer video models are systems where it's not just a, a video generation diffusion model, but also it's tied in LLMs and context windows, memory buffers, frame buffers, all that stuff. I'm guessing this is gonna be super useful for, like, a crazy set piece of, like, you know, 10 sequences where you wanna k- keep a context window across the board of, like, where all the cameras were, which scenes- Mm-hmm... where, you know, within each scene where the stuff is, like, you know, if they have a road full of cars, how many cars were there, and all that stuff.

So I think this stuff will be really important in our world as the video models start to get more intelligent. Or even, like, uh, gaming or, like, Genie, like the generative world where- Yes... the more you move through that world, then the more it can just remember stuff and be persistent. Yeah, I remember wh- when we did the test, uh- Yeah...

with Hileo and, uh- [laughs] Like... [laughs] Did I say that right? Hilea? It's Leia.

Hilea. [laughs] Yes, Hileo is- Hilea... Hilea is Ben Affleck pronunciation. [laughs] Oh, sorry.

And, uh, like, we were taking footsteps, and you look back, and the footsteps were still there. Yeah. You know? Yeah.

And, like, that takes memory- Yeah... and that takes attention- Or-... on the generation side... you're driving in the desert- Yeah...

and then the skid marks are on the road, and then you drive back- Yeah... and they're, like, still there. Yeah. Yeah.

Yeah. And right now Genie is, um, one-minute t- memory duration. [laughs] So... Yeah.

So 8X. I mean, we can get to eight minutes. Come on. [laughs] Yeah.

Come on, Pied Piper. [laughs] So, uh, yeah. Right, as a research paper, but I'm curious when... You know, if this gets adapted in- into actual models, and then the dream is just, like, yeah, you keep talking to the model, and it just has persistent, that there is no context window.

It just, it just knows everything or just remembers everything. Yeah. I think from here on out, we're gonna see these incremental innovations like this that really is, like, a blip on our radar. We will cover it on Denoised, and then we'll forget about it next week.

But then these things sort of- Mm-hmm... accumulate over time, and then by the end of 2026, we're gonna have something amazing. Yeah. Or like we mention these things now, and then, you know, you're working on something or trying to think of something a few months from now, and then it's like, oh yeah, wait, what was that thing, that paper that came out or something, that model that, like, applied to that thing?

And then maybe that's the solution for, like, the project or the, whatever you're working on at the moment. We should have a giant context window for this podcast. So like- Hey- 'Cause we forget a lot of times what we covered. I, I have been getting transcripts and working on stuff to try [laughs] to help.

Surface things. [laughs] Okay, other Google update. Gemini 3.1 Flashlight can generate websites on the fly.

So basically, imagine navigating a website, but everything is being generated. So this could be the future of some sort of hyper-personalized website or web experience where everything you're navigating to is just being built for you. E-commerce, you're like going to the site and like based on your pr- like what you do, it starts building out a rec- You're not gonna like my answer... recommending products.

I think this is perfect for Nigerian prince scammers. That is the dark side of this, yes. [laughs] Like, yo, ch- you put, put, put the PayPal link in there, and then you see, you know, pay for this thing, and it just auto-generates some forged website. Yeah, that, I mean, that's a problem with all of the generative stuff where you just make fake images of, uh, Brad Pitt asking you for money- And fool people...

that look really good. But th- uh, this is like the real-time video for web design. I think this opens up a lot of new possibilities of just like the very early stages of what if websites and w- the web that we know it is like being generated in real time as you navigate through it versus static or just other, you know, da- SQL, database, other ways that it's being built right now. I don't know if we need it.

I don't know if we need it because websites are curated, designed by professionals that understand the human psyche, our attention, our, you know, all of, all of those things, and I don't know if the thing can replicate it to that extent. I get what you're saying. I get what you're saying. [laughs] I think the- these, uh...

L- look, again, the, the Deep, the Deep Mind as a whole, as an organization is amazing. Like, they just drop hits after hits. What about user testing and design testing? So, like, you're serving a kind of different layout to everyone that goes to the website to, like, figure out which is the best, so it's always learning.

And then you kind of- Like AB testing? Yeah, but where it's like AB to the, like, 10th degree, so everyone's getting like- Right... little tweaks or variations. Sure.

And then it's like- Sure, I can see that... re-designing the original website based on that learning. I think this is still early sta- I, I think this is like how... Yeah, this is still early of like what does the new web look like in an AI world.

'Cause on the flip side, the issue that, uh, I keep hitting and is like a thing with chatbots and agents trying to navigate the web is when they go to a regular website, it's not a good experience for them to do stuff quickly. And so we have MCP and these other ways to like connect to websites, but it's still Wild West of figuring out how do we redo interfaces for both AI to, like, do things on the web without having to, like, rebuild all the websites and then also if our own web experience is, uh...

Hey, I mean, if there's anything that happens, like, new on the web, that's kinda cool because, you know, remember back to the early days of surfing the web and, like- Mm-hmm... GeoCities websites and kind of all these niche websites, and that's kind of been killed and replaced with just algorithm feeds everywhere- I think-... and maybe you click on-... half of our audience-...

look at an article... don't know what GeoCities websites are. [laughs] Um, I feel like there's gotta be more. I feel like our audience here is a little older.

Little older. Like [laughs] like RH and [laughs] I- I, I had me a few GeoCities websites, yeah. GeoCities, Tripod, Homestead was one. What were the other ones?

Those are the main ones. GeoCities was like the main. You wanted to build a website and you didn't know what to do, come to GeoCities. [laughs] Nice.

All right. So you're like you don't see... You're, you're skeptical on this one. I'm meh on this one.

Uh, prove me wrong, you know. Uh, I'm more excited about the, the first one, you- I mean-... the Pied Piper. I will, I will say this is probably one of the most expensive web [laughs] experiences if you're generating...

I know it's a Flash and it's a light model, but this would still be probably one of the most expensive, um, websites to- Hundred percent... to serve, to serve up to people. There's probably like 10 GPUs powering that little window right now- [laughs]... burning like coal or something.

Yeah. Maybe it's more for like, uh, uh, you know, Gucci or Hermes, um, web experience- [laughs]... where [laughs] you're- Well-... you're getting a bespoke web- Yeah...

service served to you. Yeah, yeah. I think for premium brands like that, you, you want like a website that's like a envelope. It's like, "Welcome, Joey," and you click on the envelope.

Right. It unwinds into like a leather, crunchy leather thing, and yeah, it's super luxurious. For that, I think this would be absolutely usable. I agree.

So Nigerian prince- And it's like we know you own, we know you have these bags- Yeah... but we, you don't have these bags- Hermès and Nigerian prince... and we're gonna kind of serve them up to you. Go for it, guys.

[laughs] All right. Good place to wrap it up. Let us know what you think of these new potential features. Do you see, uh, do you see a...

Well, first off, do you remember GeoCities? And then second off, do you, um, see a potential for generated websites on the fly? Shout out GeoCities if you remember it. Come on, guys.

Give us a, give us- [laughs]... a couple of GeoCities comments. Also, um, on the Spotify side, if you're listening and, uh, on the stats we see a lot of people listening who are not subscribers, come on. Just give us one little subscription.

That helps us immensely. Mm-hmm. So thank you in advance. Addy is always in the Spotify, so he's looking.

He's tracking who [laughs] he's tracking who's listening and who's not a subscriber. No, I'm not. I ca- I don't have that visibility. [laughs] It's not that creepy.

He's gonna peer out your window look like the Mark Zuckerberg looking in. [laughs] We'd love to have more subscribers. Links for everything we talked about at denoizepodcast.com.

Thanks for watching. We'll catch you in the next episode.