Why Grok 3 should be on your radar. Plus RED's New Z-Mount & Oscar VFX Predictions

Joey and Addy break down three major stories this week: xAI's Grok 3 model emerges as a serious AI contender despite its meme origins, with impressive coding capabilities and real-time data integration via X. RED introduces game-changing Z-mount compatibility, opening up their cameras to Nikon glass and Sony lenses without sacrificing functionality—the first major development since Nikon's acquisition. 

Plus, Addy shares exclusive insights from his involvement in the 2025 Sci-Tech Awards, highlighting innovations like fire-resistant gel technology and advanced VFX denoising, before the hosts offer their Oscar predictions for Sunday's ceremony. A fast-paced, no-BS breakdown perfect for filmmakers and tech enthusiasts.

Transcript

0:00

In this episode of the Denoised podcast, we're going to talk about xAI's new Grok 3 model. RED adds Z mount to their cameras. And we're going to talk about the Academy Awards this Sunday and the 2025 Sci-Tech Awards. Let's get into it. All right. Welcome to the Denoised podcast. I am Joey Daoud. I'm Addy Ghani. Welcome. Hey man, Oscars are coming up. I know this Sunday. So we got some predictions for, uh. We'll do some of our own predictions at the end of the episode, and then we'll see, and this is not betting advice. Yeah, no, it's not betting advice. This is just for fun advice or fun predictions. So again, the Academy Awards and also the scientific and technical awards, which sometimes interesting stuff happens there too, but not a lot of people talk about that. I was blessed enough to be a part of the selections this year, the 2025. So I was involved Last year for it. And the people that have won. Oh my God, I got to see the presentations. There's a phenomenal technology that's being awarded this year. Oh, yeah. Yeah. So we'll get into that. We'll jump into that. Okay. First up Grok 3. from xAI. Uh, so Grok 3, new ish large language model from xAI, Elon Musk's, uh, AI company. First off, I didn't realize this until recently, but xAI is a completely separate company from X, which was formerly Twitter. I thought Grok was Part of just the AI portion of X, which was Twitter also because it is embedded in the Twitter X app and it's gonna get very confusing. Oh, interesting. I didn't know that. Yeah. So if you load up the Twitter app, the X app on the bottom, there's a little like Grok icon and you're always able to access. There is a standalone app now, but before that was kind of the only way you could access Grok their AI chatbot. But it is a separate company. xAI Is a separate company, but they're very intertwined. I believe that employees of xAI are also considered employees of X. Okay. Um, Elon loves the letter X. He really loves the letter X. Model X. SpaceX makes a little bit more sense. But yeah, model. Yeah, that's true. Yeah. Yeah. I mean, he's been sitting on X dot com forever, which is why he changed the name of Twitter once he bought it. But, um, yeah. So anyways, they I sort of Grok was like the language model. I kind of like wasn't really paying attention to because it kind of seemed like a joke ish language model. And it was a meme model. It was a meme model. I mean, it would kind of like give you responses, but like, you know, put like jokes in there. And it's like, if I want the serious stuff, I'm just going to go to clutter. Yeah, never took it that seriously. And they were also way behind, but impressively. So sort of within a year of Grok one and now Grok 3, I keep seeing talk on X about Grok 3 and what people are doing with it. And I think it's worth just the least people should be aware of like this language model is out and people are saying it's pretty good. Like kind of on par with patch abt 01. I'm good at coding stuff. It has really deep research feature and it does some interesting stuff too because it is so embedded in Twitter X and and on live data and posts that it pulls in new information and. Can kind of give you more like very up to date info based on whatever is going around on X. And I've seen a lot of people talk about how good it is at coding tasks as well. Wow, I had no idea. That's all news to me. Again, uh, it's slept on and, uh, ChatGPT gets all the credit. And then, uh, there is also the other two or three. Anthropic, Perplexity, or Gemini. Between those, I think most people are getting their work done there. Yeah. And I, you know, I mean, I think there's other things around, you know, why you may or may not want to use Grok and xAI. But I think what's impressive too, and also kind of in the same vein of DeepSeek, which, you know, kind of blew everyone away because it was like, Oh, look, they did this so inexpensively, but they potentially may or may not have also used chat GPT's training data for DeepSeek. Oh, yes, distillation. Yes. Yeah, that's coming out more and more now. But having said that, deep seek is very much in play and being utilized, right? Yeah. You said Perplexity or they added, you could tap into it on Perplexity. But yeah, that was an example where it did seem like they did come figure out some other ways to computationally train models. and make it more efficient. xAI, it seems like did the same playbook of just throwing a lot of like GPUs at it and building out a data center. So they built out a huge data center called Colossus in Memphis, Tennessee. They used, I believe, 100, chips. Um, and also that, uh, because of Elon Musk's relations with NVIDIA and Tesla, they were able to get so many chips, but they were able to really kind of just do this faster. And the other interesting thing too, is I believe the way they trained, so they used synthetic data sets. for controlled learning. Wow, that that is the holy grail of AI right. AI Training. AI Yeah. And so, yeah, you want to touch more on that of synthetic data? Yeah. So when it comes to, like, training a stable diffusion model, do you give it real world images, photos and whatnot? Or can you actually just generate a bunch of images and then feed that to make a whole new model? Um, and if you do, how does the behavior of this new model trained on synthetic data differ from a model trained on real world data? So these are all the things that are being worked out right now, but ultimately it is the future because the amount of training data sets you need for anything is in the billions and to get licensing rights and things like that for billions of images is increasingly becoming more and more difficult. So the only solution out of that is to just generate it. Yeah. And if you generated the false, you come. don't have any of the issues of rights or sourcing or where you got the information from. But also, does it stay true to the real world sort of parameters, right? Like if I'll give an example, if you are making a stable diffusion model, that's going to look at clouds or you want to generate clouds, but it's only trained on artificial cloud images. Are the output going to look like real cloud? Right. Or is it gonna look like an AI version idea of what a cloud should look like or something? Yeah, this was always struck me. Two, when we started going down the like large language model and training stuff like two, three years ago, because it was like, okay, so if it trained on existing data sets already, but then we start using this stuff and we start pushing this stuff online, the AI generated stuff, and then it's updating its knowledge base and scraping new information. But if it starts scraping stuff that's AI generated, are we going to get into this, like, Circle of like, there's a lip spot where sort of like original stuff stopped being created. And now she's training on AI generated stuff that's on the internet. Yeah. I mean, and then when you hear the founders like Sam Altman and these guys talk about where AI is heading, I think they all sort of allude to just AI data overwhelming the amount of data that exists on the internet already, right? Crazy. Especially when it's like so easy, like, I mean, you could just create automation to like spit out blog posts like every second. Exactly. AI generated, but yeah, that's a world that I don't know what I, how I feel about it, like just inundated with synthetic stuff everywhere, but it's in that direction already. Yeah, the Grok 3 is currently out. You could get it through the X subscription model, which if you pay for like X premium, I believe there's like one level that's 30 bucks a month, another level of 40 bucks a month that they're saying is on par with the open AI, like 200 a month level. The Oh, so it's like a pro, it's like get something this good for this cheap. Yeah, I think I feel like with all AI models from the tests I've seen online or people comparing where it's not like way better than any of them, but on par and it's impressive how fast they develop this and caught up within a year. I feel like it's probably another indicator, too, of like, it's only the stuff's only going to get trained faster and better more quickly. Yeah. Yeah. It's a it's like an exponential curve, right? Uh, you know, I find it odd that it's paired with X, the Twitter platform. Maybe that's why it gets leapt on. I think partly that partly all of the. Chaos around Elon Musk. Probably that that doesn't help. That doesn't help for sure. There is a standalone app now. But yes, originally, I mean, you could also access it through the X Twitter app. I mean, that makes sense though, because that is a very real time source of data and they keep training your model on. So it's like he has access to outside of like Reddit or proprietary news articles. It is not the source of data, but it is a big source of data. It is a good source of real time data. I will say the other interesting thing And how they've integrated Grok in X is, uh, you'll see posts or stuff and maybe it's referring to something or some news story. And then there'll be a little like pillboxes for Grok prompts to be like, tell me more about this. And it'll kind of pull in real time data and kind of give you more context about if someone's commenting about current event or something, but you're like, Oh, what are they referencing? Yeah. It'll like research and kind of fill you in. Oh. on the back. And it's, I mean, it's actually kind of a useful implementation of AI on a social network that no one else has really implemented because it would probably cost them a lot. And it doesn't know what's up to with llama and all that. I've never seen them integrate anything on like a Facebook or Instagram. Okay. Like, Hey, give me like someone's posted about something that's like, Hey, give me like the background context about this post. If anything, I think. Instagram was pushing more towards like AI generated images. And that's the wrong direction. Yeah, they're like, it's kind of weird. And then they like pulled back on that. Yeah, we can have a whole episode about. Yeah, um, yeah, there was a whole thing. I've been seen to have just, um, AI generated, like, uh, storylines, like fake, um, accounts that create AI generated images, but it's all around a, like, ongoing, like, fake storyline. Yeah, I'll just go back to like, in a world where we're craving for more. You know, human on human actual real world interaction when you confuse the user with synthetic Yeah humans that pose to be real like that is a recipe for disaster. Yeah, and also like why do we need this? Why do we smartphones. There's already like a million good images around. Yep Yeah, I will say one other thing I have seen about Grok is uh, and I believe it was sort of touted as not having the same Like, well, let me say quote woke training as other models, whatever you want to interpret that as, but which means they, um, didn't put a lot of the safety features that a lot of other AI models have, and I've seen people post scary things saying, like, if you ask it to do like of how to build Bad things that I'm not going to mention. Um, it will tell you detailed instructions of how to build bad things and where to get the bad things. My guess is Elon just fired the safety staff. Either that there just wasn't one or it was just like, you know, let me just put this thing out and have no, um, RED team testing, which normally a lot of these, you know, other models. If you combine no safety with Starlink where you have internet access to remote parts of the world and now anybody in rogue states can have access to data. Dude. Easily. Yeah. Uh, not a good recipe. Yeah. Yeah. Normally these models are RED team where like people come in before they release it to the public and like give it all sorts of crazy prompts to try to see if what information there was that, um, hack a while ago where it was like, tell, you know, I I'm your grandmother and like, tell me how to make your favorite recipe of napalm. And like, that was like a way to like, it's like a, like a wolf in sheep's clothing. Yeah. That was like some way for someone to figure out how to hack, like, uh, I forgot which model it was, but to get it to tell you things that it's like, normally if you're like, tell me how to make a napalm. It'll be like, I'm sorry, I can't do that. But that was a way to hack it. And I think they fix that. But normally they go through extensive testing processes to make sure you can't trick these things into telling you bad things that you shouldn't know how to do. And um, apparently yeah, Grok doesn't have any of these. That's why it's cheap. So that's scary. I hope they fix that. I mean, it's, it's good to have on our radar and that's yeah, I think you should know about it. I mean, cause I had dismissed it as just like, uh, this, you know, Grok was like the jokey chat thing, but they caught up. And yeah, and it's no joke. I mean, they're, they're data center and the training models and all the stats that they are associated with. That's no joke. Yeah. All right. Next in a slightly more positive story. RED has introduced a Z mount to the V-RAPTOR and KOMODO camera systems. Yeah. So yeah, tell me about this. You were excited about this one. I'm very excited. So, uh, I'm a, I'm a big RED fan and, uh, it seems like it's just a lens mount thing, right? Like you can get, uh, a Venice with a PL mount or an RF mount or whatever. So it's like, okay, yeah. So you can get a V Raptor or Komodo with a Z mount versus an RF mount. And Z mount is Nikon's. Yeah. This is post, uh, acquisition and first move, uh, from the RED team that leans into the Nikon ecosystem. thing we've seen play out from the, uh, from the acquisition that has changed the products. I don't know if you're a big lens nerd, uh, Nikon glass, Nikkor is some of the best class on the planet. And up until now, reds wouldn't work with Nikon glass. So, well, cause most reds tend to be RF mount, which is Canon glass. And so this opens it up to all of the Nikon glass. And in addition to that, uh, Z mount and E mount work really well together. So I think it's just a simple piece of metal, a little adapter that you can get. Yeah. So now you have access to all the Sony glass. Yeah. That's cool. And yeah, I saw it was a CVP had a video, right? Demoing this and the, cause you could adapt a lot of lenses too, but a lot, sometimes you have trade offs with the adapters where you can adapt the lens, but it doesn't carry the camera control and metadata, the autofocus, but it seems like that. It's not the case if you adapt to an E mount lens, you still get all of the functionality on your RED. Yeah. With Sony lenses. Exactly. That you nailed it. Which is cool. Yeah, that's awesome. And now, like, you're, you take this camera, which is, uh, just a sensor, and now your job as a cinematographer, or even just a videographer, is like, How do I pair it with all these lenses that I already have? I don't want to have to sell all of them and buy a brand new kit. And the other thing the RED has releasing is new, uh, adapters to go with the Z mount. So now you have a PL adapter, which is what all of the professional environment teams, you know, like Cook lenses come in PL mounts, you know, stuff like that. And then also ND filters and such. I think this is a huge move to finally put RED back into the pedestal where, you know, Sony and RE are like now RED is it's a, you know, I love the company, but they have fallen a little bit behind on sort of where tier one cinema sits. And so maybe this is a move that will get them back up there. Yeah. I mean, if you could get one camera and open up, be able to use. Z mount, E mount with one body. And I think is it, there are other, Oh, here you can convert existing RF mount bodies to Z mount for either 1, 700 or 1, 600, uh, depending on the camera body. So also that you could retroactively, if you have a camera, you don't have to, you can like these self service. Yeah. Yeah. Just send them. I don't know if you get self service or you send them, maybe you can change it. The more lenses you have access to that makes the camera more powerful. Yeah. Initially when, uh, Nikon acquired RED, the speculation was that. RED 3D format are three d format that they really acquired it for that compression. It is really good. I mean, I think you're also very patent protected. It's very patent protected. It's it's really robust to data loss to write it like rights in multiple folders. So if there is data corruption on one side, it is very okay with it. And the thought was okay. Nikon is just gonna take this raw format and put it into all the Nikon consumer cameras. The millions of them now this this move in addition to that it the RED Nikon marriage is starting to make more and more sense and it is beneficial for both companies. I mean, it made sense with the Nikon to like because they sort of lag behind the whole, uh, cinema camera world. Yeah, I just video period on having their DSLR bodies like with Canon with. The, uh, the 5d mark two and that kind of launched the whole like, Oh, you can shoot really good looking video. It's all these like DSLR cameras that look very cinematic. Thank you. Yes. No, I, I have a really soft spot for Nikon. Uh, my first photography camera, the one I learned photography with was a Nikon D 3100 about 10 years ago. Yeah. I loved it and it was my first Nikon 35mm was my first camera. Okay, there you go. Still, still photography camera. Yeah. Yeah. So I have a soft spot for Nikon. And yeah, like Canon has their C500, C400, C300. They have cinema cameras, obviously Sony, Venice's. Yeah, they successfully went and expanded the product line. The cinema series, cinema ranches and Nikon just never did, but now they have read. Yeah, it was a good step to that. And once it came out, what the acquisition price was really good deal. That's a good deal. I think it was around like 80 million or something. No, I think it was 4 billion. I thought it was like in the, I know, I thought it was way lower. Oh, a million. 80 like million. Okay, well, we'll, we'll get it. All right, now we have to fact that I thought it was like, I was like, I was like, wait, really? I thought it was Surprisingly low. A lot of my co workers at Disguise, uh, are, were from RED and RED is obviously here in Orange County. And, uh, Jim Gennard and some of the founders of RED come from Oakley. So they're like kind of a rock and roll surfer company. They're very cool. Like the, the skull logo and the black and white camera bodies. And I just love the brand. Yeah. Yeah. Yeah, I mean, they've successfully, I remember when they first came out, it was like, we're gonna build a high end cinema camera. It's like, what? What? Digital? Like, you make Oakley's. Yeah. And yeah, they did it. They successfully did it. Yeah. All right. And to wrap up, we got the Oscars coming up. This Saturday, so yeah, we'll do some of our predictions, but let's talk about the scientific and technical awards. Yeah, so SciTech awards are really prestigious, probably the most prestigious MNE technical award that's available today. It takes. A long time to win an award. Yeah, so this is not like the Academy Awards, where it's like all the films and things that happened in the previous year, we're going to give awards out. This is a completely separate process. And it's for technology that could have happened at any point in the past. 100%. It's unlike the Emmys, where the Emmys are, I would say, a much Bigger slate and the slate gets renewed every year and like like disguise won the Emmys couple years ago and you know This is much more exclusive and it does take a decade like not joke Like you can have something innovative today, but ten years from now. Yeah may win an award I mean speaking about the one that stood out to me the most is like I'm like aware of this and they're getting an award Is a Freefly's Movi gimbal system. So the Movi always been a Bennett Yeah. Cause this huge splash. Remember when it came out like 10 or 12 years ago at NAB and it was like, you know, someone holding the, the, the rig and they're like moving it like this and the camera's just staying still and everyone's just like, whoa. And they do this. Yeah. And now it's like, um, you've got like for a hundred bucks, you can buy a pretty decent version for an iPhone. Yeah. And like, that's just like, gimbals are everywhere. Yeah, I think the DJI Ronin is a ripoff of that, right? Uh, the Ronin did replicate the Movi. Same technology. And then, same idea, yeah. The Movi was also kind of innovative too, because it was, you could have someone operating, moving the camera and stabilizing, and then you could have a separate operator with an RC controller. That's right. And to adjust the head. That's so cool. Yeah, and that, these were huge, uh, developments when they came out, and now we sort of just, Like it's pretty standard kit for like everyone like and and and the technology has gotten smaller and smaller and smaller. Um, yeah, I feel like DJI Ronin has probably run away with it a bit like movies still kind of stay and they do their background was like and they basically took the gimbal system from drones and they do a lot of drone stuff. Got it. And then it's like, oh, we have a stabilization. I think that might be paraphrasing here, but they had, but they primarily focus on like high end drones. Yeah. And then it was like, oh, we have the stabilization system for drones. We just We could just hand hold it. Yeah, sure. Um, and same thing with drones. You have someone piloting and you have someone else actually separately operating the camera. So they're getting an award. But yeah, again, this was like that. This is technology that came out 10, 20, 12 years ago, and now they're getting an award and luckily enough for this year. And I don't know if I'll ever go through this again in my life, but I'm so blessed to have gone through the actual process. Yeah. What's that like? Oh man, it was so cool. It was so cool. I got to present in front of the SciTech committee, like a panel. And it was like a very exclusive, you know, I had my suit on and everything. And, uh, I, I got to meet all, all of these people that are on here. Actually, we got a live demonstration of the Naked Burn Gel to Nimi Vaino, who won it, Fireskin 360 Naked Burn Gel. So we walk out of the auditorium and we go to the parking lot where this demo is set up. It was so cool, Joey. This guy puts his gel on with like this paintbrush. It's clear and then it dries in a few seconds, takes a flame torch, puts it on his arm. Nothing happens to him. Crazy. And then he does it with a wet paper towel. He puts on the gel on a wet paper towel and blow torches it and the paper towel is still wet. Not a single burn mark, anything. This product is incredible. It's like the complete opposite of napalm. Yeah. So, uh, I mean, look, this stuff is not new. I'm sure it's, and it was being used for film production for many, many years. And finally they have the achievement recognized. Yeah. Yeah. Uh, I did also notice too, that there are a handful of awards all around, uh, denoising technology. No pun intended. But yeah. So can you tell me about any of the, what has got one for denoising Intel and video and then also Disney's? Um, yeah, denoising is a big deal. Um, this podcast is a big deal, but denoising is a bigger deal. Here's why. Uh, computer rendering is incredibly compute expensive, right? So one frame can take hours to render, but what if you render it with less rays and render it Much, much faster, but the resulting image is just noisier because there's just not enough details filled in. Well, can you just pass that through a Denoisedr and get to your final result? And that's what denoising is. It's a way to shortcut rendering effort. Uh, and the denoising pass itself is much more efficient than the actual rendering. And now I think some of the denoising is using machine learning. So, yeah, this is, this goes hand in hand with where AI and ML is and also where rendering technology has involved. So, denoising definitely deserves to be recognized here. Yeah, and I'm sure down in the future we'll probably see more denoising for AI applications rather than just rendering applications. But yeah, that's awesome. I think the name of our show, Denoised, comes from stable diffusion. Uh, denoising noise to create images, but this is like a different kind of Denoised. It's like the name of the show. It could be like many applications like denoising is a tech for AI stuff, but it's over like denoising news. And yeah, there are like many ways you could take the name of the podcast. Um, all right. And, uh, for fun, let's do a couple of predictions. We're not gonna do all of the categories, but let's do the technical ish categories for best picture. Uh, let's say best VFX. So the nominees for that one are, uh, Alien Romulus, uh, Better Man, Dune Part 2. Kingdom of the Planet of the Apes and Wicked. Yeah, I know who's gonna win, but I'm gonna tell you who should win. Right. Who do we think strategically is gonna win or do you actually think should win? Okay, what are your I think Dune Part 2 is gonna win. It's just the You do? You think? Because that's the movie that sets the trend for VFX, right? Like, VFX movies will want to emulate the Dune look. From now it's that dirty grungy look that doesn't look like there was any VFX done to it. Lots of sand, lots of dirty lenses. I always wonder with Dune, I love doing, I was wondering if it's one of those things where it's like, uh, all the words are going to hold off until the third one comes out. Like, that's always like, like, like Lord of the Rings. Like it didn't really get any of the awards until Return of the King. Yeah. And it's like this weird kind of trilogy thing where they're just like, well, we're just going to give it all to you once you're done. But my heart, it goes out to Alien Romulus. Oh my God. Like some of the. Best visuals on screen. How about you? I was thinking kingdom of planet of the apes. That one, they shot with the actors live action on location with all the tracking and then replace them with the apes, but we're able to really blend the shooting on location. There's a whole cut of the film. With the actors. Yeah. Without the, before the VFX was applied, compositing in real time doing virtual production Uhhuh, um, on, on a viewfinder perhaps. Mm-hmm . You know. Yeah. So I thought there was a lot of really kind of clever technical innovation and Sure. And, and, and taking the next step and a lot of, uh, tech that we've seen. But Is that Weta? I believe that's Weta. Yeah. I feel like all roads always lead back to like, whatever, . So not possibly that, but, uh, I mean, I wouldn't be upset if Dune: Part Two, one. Okay. All right. Best cinematography nominees were our Nosferatu, The Brutalists, which we've talked about before.

Uh, Dune: 24:33

Part Two, Maria and Emilia Perez. I mean, I have to give it up for Greg Frazier. We're all huge fans of Greg in the virtual production community. Yeah. I mean, I'm hoping Dune wins for cinematography. Yeah, it was amazing. Yeah, it's just beautiful. And, and yeah, talk about something where you just see like. Frames from that film just keep surfacing of just like every frames of painting. Exactly. Yeah, that yeah, it's just beautiful. Nothing. I mean, the brutalist. I like the story of behind, you know, shooting the Vista vision. But yeah, I still just all the visuals in June. Yeah. I think the brutalist is more cinematic than all of the ones here. We're the only one also shot film out of. Yeah. Nosferatu. Do they shoot film? I don't know. Actually, not sure. But uh, yeah, I saw that. Some I need to watch that this week, but Nosferatu. But that's also been my best picture. It looks good. All right. Best picture. So that one's got 10 nominees. Anora, The Brutalist, A Complete Unknown, Conclave, Dune Part Two, Amelia Perez, I'm Still Here, Nickel Boys, The Substance, and Wicked. Who do I think is probably going to win? Probably Anora. That's more what I kind of hear. That's the hype movie for sure. Yeah. Yeah. Uh, and, and I think the movie itself is, uh, is a young filmmaker and. Uh, it was shot with, uh, cheap gear, uh, kind of like the creator and like, it's, it's like where, uh, Hollywood is evolving to. So it kind of alludes to the next generation of filmmakers, be able to pull off a big, something that looks, looks, uh, good with a low budget. Great storytelling. Yeah. My favorite Dune part two and like movie that I've watched multiple times already this year, right. Dune part two. Right. Um. I don't think it will win, but yeah, that would be my pick. Sam, like your, my pick is the brutalist here. It is going to win best picture. Oh, your pick's the brutalist. Okay. Yeah. Okay. For sure. We'll see. We will revisit this after, uh, after the account happens this, uh, this weekend, we're both wrong. And the winner is the substance, which I would not be upset if the substance was a wild, fun movie. Uh, well, cool. That is it for this episode of Denoisedd, uh, as always show notes and everything are at Denoiseddpodcast. com. And please give us a like, follow and subscribe, we're still counting on you for early support at this stage. Thank you so much. Thanks everyone. Catch you in the next episode.