
The Intersection of AI and Authenticity
Mason Amadeus: From the 8th Layer Media Studios in the backrooms of the deep web, this is The FAIK Files.
Perry Carpenter: Where artificial intelligence meets natural nonsense, and we do our best to sort through all the aftermath.
Mason Amadeus: I'm Mason Amadeus.
Perry Carpenter: And I'm Perry Carpenter, and on today's show, I've got a segment about Meta's new watermarking initiative.
Mason Amadeus: In the next segment, I've got a piece about Spotify and fake artists in response to a voicemail from a listener. And after that, I'm bringing you a segment about a new release from OpenAI, a physics engine.
Perry Carpenter: And then at the end, round out with a Dumpster Fire of the Week. This is the story behind the Stop Hiring Humans billboards that have popped up in San Francisco.
Mason Amadeus: Sit back, relax, and we've been trying to reach you about your car's extended warranty.
Perry Carpenter: We really have.
Mason Amadeus: We'll open up the FAIK files right after this. [ Music ]
Perry Carpenter: Synthetic media is everywhere. Deep fakes are going to be everywhere more and more and more. And of course, organizations that are putting these out and are very consumer focused are trying to solve for the fact that people are being deceptive with the technology. And one of the ways that they try to solve for that is watermarking or providence marking or, you know, something either visibly in the media or embedded in it. And there's a new initiative from Meta where they put forward a tool for watermarking AI-generated videos. This is covered in a TechCrunch article. They mentioned this as "We developed Video Seal to provide a more effective video watermarking solution, particularly for detecting AI-generated videos and protecting originality." And that was from a research scientist at Meta. The TechCrunch article goes on and says it's not the first technology of that kind. Google DeepMind has SynthID that can watermark videos. There's ways that AI audio gets watermarked. Microsoft has its own video watermarking technologies. But all of these approaches fall short, and that's what Meta is trying to call out.
Mason Amadeus: I'm always interested by watermarking in general as a practice because I think it's like --
Perry Carpenter: Yeah, I think it's ineffective.
Mason Amadeus: It's super ineffective, but we've been doing it forever. Like anyone who wants to be deceptive is simply going to find a way to remove the watermark.
Perry Carpenter: Absolutely.
Mason Amadeus: Are they doing something to make it -- like embedded in a way that can't be removed?
Perry Carpenter: What Meta says is while other watermarking tools exist, they don't offer sufficient robustness to video compression, which is very prevalent when sharing content through social media and platforms. They're not efficient enough to run at scale. They weren't open or reproducible, or they were derived from image watermarking, which is suboptimal for videos. So in addition to a watermark, Video Seal can add a hidden message to the video that can later be unconverted to determine its origins. And then Meta claims that Video Seal is resilient against common edits like blurring and cropping as well as popular compression algorithms. So they're trying to solve for all that, and I think that that's good, but one of the things that I point out, whether it's Meta or like what OpenAI is doing with Sora or some of the other providence marketing and tagging that we're seeing come out from different platforms, is that it can be effective for people that don't know that it exists or that people who are lazy.
Mason Amadeus: In thinking in that line, as someone who's been obsessed with steganography and stuff and like hiding, encoding data inside of other data, I think I have a hunch as to what they're trying. Because if you simply added metadata that was at the file level, that doesn't persist as soon as someone's screen records it.
Perry Carpenter: Exactly.
Mason Amadeus: If you add just a visible watermark, you can see that can be cropped out or probably masked out. But there is something you can do. You can take an image and open it up in like a hex editor and change some data inside it to encode some text or a secondary picture or whatever inside the pixels of the image data in a way where it doesn't look distorted, but if you know how to dig through and find that data, you can pull it out. And so I'm wondering if they're actually trying to embed this watermark in the pixels of the video. So that way, if it's screen recorded or captured in some other way, as long as the visual content is present, the watermark is present. But making that robust to compression, that's a giant ask.
Perry Carpenter: Yeah.
Mason Amadeus: The way it works by being undetectable visually is that the distortions to the pixels are almost imperceptible, and that is the kind of detail you lose when you compress a video is those small details.
Perry Carpenter: Yeah, it would be interesting. I assume it's a combination of things like steganography and some kind of maybe frame rate interpolation or something like that, though that wouldn't survive some edits as soon as people know that it exists. So there's -- I think any single method would end up being defeatable. But again, for me, what it comes down to is as soon as I know that Meta has a system like that or OpenAI has a system like that or Google has a system like that, if I'm wanting to use this type of technology for something deceptive, I'll just not use those. I'll use a local model like Stable Diffusion or one of the Chinese video models that are out there that may not have that, or if they do, it's something totally different that's not meant to be out front in the same way that like Meta's would. So I think it's good at keeping honest people honest. The other thing might be repurposing. So with some of the things that happened this election cycle, for instance, there was the fake video ad from Kamala Harris where somebody had cloned her voice and said that she was with her saying that she was like a DEI candidate and --
Mason Amadeus: I remember that.
Perry Carpenter: Things like that. It was originally meant as satire and parody, but it then got repurposed by people who removed the satire and parody frame and just started to push it around. And then it essentially became a piece of disinformation. And I think that there's a lot of things like that where watermarking and providence marking can help because that is kind of the lazy way of creating disinformation or creating some kind of social unrest is just take something that was created in one context and repurpose it in another, rather than creating your entire new thing. I think if somebody wants to create disinformation, they're going to use a system that does not have these watermarks in it as long as they know that they exist, and somebody determined would know that.
Mason Amadeus: Kind of just like the cat-and-mouse game of all security, right? You're just throwing hurdles in front of people to make the barrier to doing it convincingly a little bit harder.
Perry Carpenter: Yep, absolutely.
Mason Amadeus: How do you interact with this? How do you watermark your creation using this tool? Do you just run it through it and tell it what you want it to input as watermark data or --
Perry Carpenter: Well, I mean, I think it would be similar to maybe more invasive within the video than like some of the date/time/location watermarking and equipment watermarking that gets put within video and image from several different consumer-grade cameras that are out there. And they're trying to make those more resilient to tampering as well. So it's not just within the metadata of the image. It's provenance that gets much more interpolated within the data of the image. So I think there's more of that that's going on. Of course, anytime you talk about all of this now, there's also talk about blockchain. So can you put providence markers within an image and then link that up into the blockchain somehow so that you understand if that's been altered over time?
Mason Amadeus: Man, blockchain is really a hammer looking for a nail. They're trying to find something to use it for.
Perry Carpenter: Right.
Mason Amadeus: What a what a cool technology with absolutely no real functional use I've found yet.
Perry Carpenter: Yeah. Well, and when you realize the things within blockchain that make it vulnerable to attack, it's still not necessarily something that is foolproof or something that's tamper proof.
Mason Amadeus: I have a fun little fact to end this segment on, which is that I have a subaudible watermark that I like to insert into audio projects I do. That is just a little waveform of my initials.
Perry Carpenter: Nice.
Mason Amadeus: I haven't put it in every episode of this show because, frankly, I forgot to. But it's something that I have done across my work.
Perry Carpenter: Well, and people who have followed some of our work will know that there's been some steganography that we've done within audio before.
Mason Amadeus: Yeah. Anything that you and I are involved in should be suspect to the people who want to dig into that.
Perry Carpenter: Absolutely.
Mason Amadeus: And with that said, we're going to be right back with a segment about Spotify, AI music, and fake artists in response to a voicemail that we got.
Perry Carpenter: Awesome.
Mason Amadeus: Stay right here. [ Music ] So we got a voicemail. Sayhi.chat/faik is a place you can leave a voicemail. This one, Perry, I'm going to share with you is from our listener, Ty.
Perry Carpenter: Awesome.
Mason Amadeus: Let me just pipe this over to you real quick.
Listener Ty: Howdy, howdy. I was just thinking about Spotify, and I was thinking about how there are definitely people out there who are just making AI-generated music and then uploading it onto Spotify for a quick buck. That is almost certainly a thing that is happening. Maybe this is a known thing. I don't know. But thinking about that. Mason, edit this out -- edit out me stumbling over my words. Thinking about that, what are the odds that Spotify, Spotify itself as a company, starts generating music and flooding their algorithm with it, making use of all the metadata that is available to them to feed people stuff that they don't have to pay royalties for? Yeah, I've kind of lost my train of thought here, but I think you get where I'm going. That's all. Keep up the good work on the show. Bye.
Mason Amadeus: Ty, your question sent me down a massive rabbit hole, and I have a lot of information. We're going to go through it really quickly. Try and keep this segment inside of 10 minutes, and Perry, I saw you just pulled up something interesting, too.
Perry Carpenter: Yeah. So this hits one of the points that Ty touched on. I know you're going to go deep dive into one of the other points, which was some of the conjecture. But Ty said, hey, I know that some people probably are doing this and they're uploading AI-generated music. There was actually a man that was arrested on September 4. The press release about his arrest was September 4th of last year, 2024. He was found to have uploaded a whole bunch of AI-generated music, which in and of itself is okay; that's permissible. But he used a ton of bots and fake accounts to do it. So he was essentially creating hundreds or thousands of pieces of music and using fraudulent accounts in order to get royalties on that. And he was also using bots to listen to the songs.
Mason Amadeus: I bet you that's what they pinned him for.
Perry Carpenter: Yeah, he was uploading slop under fake accounts and fake identities, and then he was using bots to stream those back down and then receive royalties from them, and he received quite a bit of money over time. It says that he fraudulently obtained -- and we'll put this this article in the show notes, too. He fraudulently obtained more than $10 million in royalty payments through his scheme.
Mason Amadeus: So, well, it's interesting because the word fraudulent is doing a bit of lifting. Let's back up a little bit. Spotify's official policy on AI-generated content is that it's allowed as long as you're not impersonating someone. So you can't upload a fake Taylor Swift or fake Drake song or fake whatever, but you can upload any AI music you want. And this -- it's funny. The first thing I found when I started down this rabbit hole was a conspiracy theory posted to Reddit about Spotify making their own AI music, because a lot of people have noticed that, particularly, these sort of background listening categories, mood music and focus music, are dominated by these like first-name/last-name artists who all make these ambient tracks that are kind of similar, same sort of length, and that is where the rabbit hole opened up underneath me, and I fell down. So Spotify says AI content is allowed. The CEO stated they have no intention to ban AI-generated music at large. I found out that Spotify actually owns an AI laboratory in France, the Spotify Creator Technology Research Lab.
Perry Carpenter: Makes sense.
Mason Amadeus: It was open in 2017. Information on what they've kind of come out with is a little bit sparse. There's a quick tangent I want to go on about the person who runs it. The person who runs the Spotify Creator Technology Research Lab in France is Francois Pachet. Pachet, I may have mispronounced his name. I'm sorry, Francois. He's super interesting and cool. Wikipedia describes him as one of the pioneers of computer music, closely linked to artificial intelligence, especially in the field of machine improvisation and style modeling. And he's been in this game since all the way back in 2002 when he created the Continuator, which is an incredibly cool piece of technology.
Perry Carpenter: The Continuator.
Mason Amadeus: Yeah, it's a usable musical instrument. It's a piece of software that you can use to control an instrument like a MIDI instrument or in one demo, a physical piano, where it learns and plays interactively with you and then will mimic your style and just move forward. So --
Perry Carpenter: Move forever. I mean, I can imagine you go to this research lab, and it's just infinite lo-fi beats. Like as soon as you enter the facility, it's just under the current of everything. And they've used the Continuator, and it just goes ad infinitum.
Mason Amadeus: Kind of like a like a John Cage piece but --
Perry Carpenter: Exactly.
Mason Amadeus: It's really cool, and it's based on a Markov model of musical styles augmented to account for efficient real-time learning and arbitrary external bias. It's very cool, 2002, and it's remarkably good. I encourage you to check it out. We'll throw a link to a YouTube demo in the in the description, but back to Spotify. Music Business Worldwide is a publication. It's pretty reputable, and it's pretty on Spotify's bad side. Way back in 2016, they broke a story about Spotify paying artists to create mood music under pseudonyms, and Spotify at the time vehemently denied all of it, despite it turning out that they did. And recently, a different Swedish newspaper whose name I'm probably going to butcher, but it's Dagens Nyheter, D-A-G-E-N-S N-Y-H-E-T-E-R, published a big exposé, and they identified a musician behind the world's most-listened-to network of fake artists on Spotify. This one guy named Johan Rohr, whose music, the report says, has been released on Spotify under 50 composer aliases and at least 656 invented artist names. His songs have been added to well over a hundred different playlists for instrumental music, which together have over 62 million followers. And he has songs on at least 144 official Spotify playlists under various artists' pseudonyms, and in 11 of those playlists, more than a fifth of the entirety of those lists are made up of his songs.
Perry Carpenter: Wow.
Mason Amadeus: Spotify has not denied this. They're so cagey and weird when it comes to this stuff. They said, "This music exists primarily in Spotify's focus hub, which limits competition with artists from traditional genres and popular music. We don't stop artists or bands from releasing music under their own names or under various pseudonyms, and as the demand for this type of music increases, so do the number of rights holders and artists producing this type of content." This guy has made approximately 3.1 million U.S. dollars in 2022 just from royalties. And he's not in jail.
Perry Carpenter: Right? Yeah, he didn't have the multi-year trying to kind of behind-the-scenes fraud campaign. So this is an interesting juxtaposition between these two people, right? Because you do have -- they're doing essentially the exact same thing. One person is doing it in a way that's got a little bit more subterfuge behind it. The other one is fully known by Spotify and encouraged by Spotify to do it.
Mason Amadeus: Yeah.
Perry Carpenter: Either way you can make a few million.
Mason Amadeus: Yeah, and there's a big history of Spotify sort of doing this. I would encourage you to go poke around musicbusinessworldwide.com. Search their articles on Spotify. There's way too much to get into, but basically, yeah, Spotify does pay artists, and in some cases, pays artists a one-time fee to make songs that they can then include, and it's all inside of this background passive listening.
Perry Carpenter: Makes sense. That's a huge category on YouTube too.
Mason Amadeus: Massive amounts of listenership. Massive amounts of money. So to wrap it all back up and bring it back to your question, Ty, does Spotify make and share its own AI-generated music? There's nothing to confirm or deny that, but I would suspect they don't because what incentive would they have to do that themselves when loads of people are doing that for them already, and Spotify gets the majority of the cut of plays from their system?
Perry Carpenter: The incentive to do it is that they wouldn't have to give any of the royalties away. And so if they had an algorithm that could reliably just create music slop that people would actually want to listen to. So I guess it's not slop if people do want to consume it, but if it made listenable music that was based on an algorithm where they know people want to listen to it, engage with it, then they just let the computer run. And then you're looking at the ROI as cost of computation versus like what, what you would have to outlay your royalty payment to somebody else.
Mason Amadeus: Which makes me think that if they were to do it, they would release it as a feature rather than do it in the shadows. I feel like they would introduce an infinite focus mode or infinite chill, and it would be something transparent that you can use on the front because I -- Spotify is not a company I would describe as trustworthy really, but I don't think that they would need to hide this. They, they have a ton of AI generated music. You can upload it, but I think if it was them making it, they would be up front.
Perry Carpenter: It depends. So the reason that I would think you might want to hide it is if you put a name behind it, that then can have its own personality with it.
Mason Amadeus: I would think just because of the area this is in, of the passive listening background stuff, most people just put on 10 hours of lo-fi, you know, or whatever.
Perry Carpenter: Yeah, true.
Mason Amadeus: So I feel like the opportunity would be there, and maybe they are filling it with their own AI-generated slop. God knows there's a ton of playlists full of pretty samey music from these first-name/last-name artists.
Perry Carpenter: Well, and as soon as Spotify does that, then you're going to have the Sirius XM and everybody else are going to do the same thing.
Mason Amadeus: Yeah. I also, from like the perspective, I don't want to get like fully into the philosophy of like, how is this treating artists and stuff? Because that's a whole can of worms you could talk about, but the music industry is not anyone's friend and never has been. I want to name a couple of songs for you, Perry, and if you don't know any of these songs, I'll be surprised. Katy Perry's "I Kissed a Girl" and "Roar," Maroon 5's "One More Night;" Taylor Swift's "Shake It Off" and "Blank Space," "The Weekend," "Blinding Lights; Britney Spears, "Baby One More Time," Backstreet Boys, "I Want It That Way;" Celine Dion, "The Way It Is; NSYNC, "It's Gonna Be Me." You know all those.
Perry Carpenter: All those, yeah.
Mason Amadeus: Those were all written by Max Martin. This one guy, this Swedish songwriter, who's got his fingers in those and a ton of other hits. It's always kind of been a bit like this when it comes to music and money.
Perry Carpenter: He's got his finger on what's going to go like viral.
Mason Amadeus: Yeah. He's the party maker of, of the 2010s and all the way going back to 1999 with NSYNC.
Perry Carpenter: It's going to be Max.
Mason Amadeus: There's a second Swedish guy whose name I can't remember, but the music industry, the recording industry, is kind of a nightmare.
Perry Carpenter: Yeah.
Mason Amadeus: And I just wanted to bring light to that in general in response to your question, Ty, but I hope that helps answer it. I know it's not conclusive. Are they doing that? Maybe, probably not, but they're certainly doing other stuff that's pretty close to it, and their users are doing it.
Perry Carpenter: Yeah. And it could definitely grow to that. I could definitely see people and organizations going that route because the cost of inference for AI-generated music is going to continue to go down. And as we see -- as we use tools like Suno and Udio, the quality and the listenability and the ability for the algorithm to understand what people like is really, really good.
Mason Amadeus: It is, and it's only getting better. All the jingles in this show are Suno AI. Unless, if they're not, I'll say probably, but just assume they are.
Perry Carpenter: Yeah, the vast majority, at least. And then a lot of the transitions, the little sounds of the papers and everything else, that's, that's you personally at your desk doing those.
Mason Amadeus: Yeah.
Perry Carpenter: Literally. I'm not saying that sarcastically for listeners. It is literally Mason recording those sounds and finding interesting ways to combine them.
Mason Amadeus: Yeah. I wanted to, for every piece of sound design in the show is either 100% AI generated or 100% recorded by me at my desk. So a fun bit of show trivia, I guess.
Perry Carpenter: That is fun.
Mason Amadeus: Coming up next, I've got a segment about a new release from OpenAI, which seems very promising, but also a bit snake oily. Stick around. [ Papers Shuffling ] So I want to talk about another release from OpenAI's Ship Myths or the 12 Days of OpenAI, and full clarity, this is a re-record from the rest of the episode, because initially we just tried to record a bit too far in advance, and there wasn't as much information handy.
Perry Carpenter: Right.
Mason Amadeus: So taking another run at this during the 12 days of OpenAI, one of the things they announced was this Genesis model, and I feel like it didn't get as much attention as the other ones, but it is simultaneously really cool and a lot of hype. So what is Genesis? It's a physics simulation engine, basically. It's a powerful physics engine capable of simulating a wide array of materials and phenomena coupled with a generative data engine so that you can transform natural language prompts into an interactive scene or a robotic behavior. So you can say generate me -- in the demo video, what they did was generate me a water droplet running down a beer bottle and visualize all the forces and then --
Perry Carpenter: That's a really good ad for Heineken.
Mason Amadeus: Yeah, it was a really good ad for Heineken. It was their bottle front and center the whole time, and it was cool. You saw that video, right?
Perry Carpenter: Yeah. I mean, there was a ton of hype around it, and I think that that's what you're talking about. The first day or two, everybody was showing that and talking about the possibilities, and then it looks like as you dug into it a little bit more, there's some complexity there. There's some assumptions that people were making. Yeah, tell us about that.
Mason Amadeus: So at first I was like, oh, this is just all hype because I still haven't seen many videos of anyone else using it for anything, just the demo videos they came out with. So that's a red flag, right, or a brown flag. And then the language around it, I disliked because they talk about like it creates 4D worlds and all of this stuff, which I'm assuming they mean the fourth dimension as time, which is this stupid colloquial convention I hate because a lot of people assume that for fourth dimension is always time and all this stuff. They could just say it makes 3D simulations that can exist over time. That's what it does.
Perry Carpenter: Right.
Mason Amadeus: And there's just a lot of hyphy language about it, but what it really is, and what I dug into to find was a really, really good physics engine that they rebuilt from the ground up, and it can achieve speeds 10 to 80 times faster than other platforms for simulation, like Isaac Gym or MuJoCo, MJX, which are physics simulation platforms used for developing robotics applications and other things like that. If you've played a video game, you can think of like the physics engine underlying that game that makes all the objects interact properly. It's a super, super fast physics engine written 100% in Python, which I hate, just because Python is my least favorite language. It's fine.
Perry Carpenter: Every time you say that, it's so ironic that you're talking about AI because Python is like the core language that everybody, not everybody but --
Mason Amadeus: Pretty much.
Perry Carpenter: -- most of the AI stuff is using right now.
Mason Amadeus: Yeah. I mean, it's really -- it's easy to learn. People say it's easy to read. I disagree. It's really hard for me to parse the way it looks. I'm sure I'm going to have to get over this hump because I'm working on getting deeper into the nitty-gritty of working with AI. So eventually I'll probably come around and start --
Perry Carpenter: Yeah, you may love it a few months from now. Python is the best.
Mason Amadeus: No. I like my static languages, my statically typed languages. I don't know why it's just, it's so much clearer to me. But, and then again, I write a bunch of stuff in JavaScript, which everything's dynamic and nothing means anything. So that's beside the point. OpenAI's physics engine through Genesis is made entirely in Python, and it brings together a bunch of different physics solvers. And I want to avoid just making this like a big bunch of definitions and stuff, but basically, a physics solver is the actual equation or algorithm that solves what should happen when two things hit each other.
Perry Carpenter: Right.
Mason Amadeus: Or if you think like when a character steps up onto a rock, their foot lands on the rock and their knee bends a certain way so that their leg stays lined up with their body. That's called inverse kinematics, and there's a solver for that. So you can place a point and have the other joints line up to it. All of these things are very important for 3D modeling, simulation, robotics design, all stuff like that. All of that looks awesome, super cool and functional. And that's what they open sourced was this underlying physics engine.
Perry Carpenter: Right.
Mason Amadeus: But what they promoted and what everyone got excited about was the generative layer on top of it, being able to just say, make me this thing, and then it generates this real simulation. That has not been released yet. They say they will, but we haven't seen it, and I haven't seen any other demos of people using it.
Perry Carpenter: Yeah, as I was looking at it, too, what people were saying is that the physics engine is a genuine breakthrough. People are really excited about it, but what it's forecasting is integration possibilities within other engines like Blender and so on, so that whether you're using something generative or something else, you're applying the physics engine onto some other tooling.
Mason Amadeus: I'm so glad you said Blender because I'm dying for when someone figures out how to integrate it into Blender because that's going to be incredible.
Perry Carpenter: Yeah.
Mason Amadeus: Because Blender already you can run Python scripts inside of it. It shouldn't be too hard in theory. I don't think I have the know-how, and I definitely don't have the drive to be the one to create that integration, but I know that somebody will.
Perry Carpenter: No, I know somebody will have already done that by now probably. They just might not have released it widely.
Mason Amadeus: I can't wait because that's going to be so cool. And then the other thing, like you said, with integrations, NVIDIA announced that in early 2025, they're going to lean really heavily into robotics. They've released a new generation of compact computers designed to be incorporated into robotic devices. The line of computers is called Jetson Thor, which is fun. They said that NVIDIA's kind of angle here is to become sort of the OEM for all the other robot makers. So, you know, Tesla wants to make humanoid robots. NVIDIA wants to be the person that makes the embedded computer that goes in that robot to coordinate and control it. And they said that the breakthroughs that have made this possible are the ability to train robots using Generative AI in simulated environments, and that is extremely cool because if this physics simulator can run faster than real time, maintain accuracy, it can be used to train robots to perform advanced behaviors like walking and other things like that. My mind went back to all those videos in the early 2000s of people like "AI learns to walk." "AI learns to play Flappy Bird" or whatever, early 2010s, not early 2000s.
Perry Carpenter: And NVIDIA is making a play like, you know, they are the powerhouse behind most of the Generative AI stuff. And typically, they seem to be happy with being the underlying force that's being used for training or being used for real-time inference or being used for anything else. But at the same time, they are doing a lot of innovation with their own models as well. There was an article that came out just a little bit ago that says, "Bye-bye ChatGPT. NVIDIA Launches Chat with RTX Free for Everyone." And it is a local large language model that seems to have some promise. You can just download it from NVIDIA, run it on a Windows machine.
Mason Amadeus: Oh, wow. And it has RAG in it and everything?
Perry Carpenter: I believe so. I didn't check with the RAG capabilities. I literally just saw the headline. So I may be missing --
Mason Amadeus: I'm looking at the at the page. It says that it -- "Chat with RTX is a demo app that lets you personalize a GPT large language model connected to your own content, leveraging retrieval, augmented generation, tensor RT LLM and RTX acceleration."
Perry Carpenter: Yeah, there we go.
Mason Amadeus: But anyway, I'm excited to see what happens with OpenAI's Genesis as it gets incorporated into other things. It kind of feels like they just needed something for that day, had a cool demo they could put out, and the underlying tech is good, but as far as immediate impact and usability, unless you're very, very deep in a technical niche, there's nothing there.
Perry Carpenter: Talking about physics engines, Genesis answers a question that Sora hasn't yet answered, right? Because they did release Sora, and everybody was saying that Sora is like this complete world model, and we didn't see that in the way that Sora currently handles physics.
Mason Amadeus: No, not at all.
Perry Carpenter: So the reality and the vision are two different things, but maybe Genesis, at some point, converges with Sora and provides some more real-world physics.
Mason Amadeus: That would be interesting if they could use -- I mean, you'd have to use the simulation data to drive a video generation instead of using generated 3D models moving through sort of predefined paths, because it does include its own ray-tracing renderer.
Perry Carpenter: Right.
Mason Amadeus: So you should be able to you get a video out of it.
Perry Carpenter: Well, I mean, the interesting thing and this may be too much conjecture is that a lot of people have been slightly disappointed with the text-to-video from Sora, but they've been really, really impressed with video-to-video with Sora, where you take a video from one source, put it in, and it has some interesting upscaling and some creative possibilities. And when you're thinking about using something like Genesis to really understand the physics, pushing that to a 3D model or like Blender and then outputting that automatically as a prerendered video file into something like Sora and then really skinning it. Well --
Mason Amadeus: I follow you.
Perry Carpenter: -- then the possibilities open up.
Mason Amadeus: Yeah. I follow you now. That would be -- yeah, because then you -- essentially, it would be kind of like I don't know if you -- this going to be a weird, specific pull, but if you watch the special features on like a 3D animated movie like Shrek and they show you the not fully rendered scenes where everything is just kind of blobby and flat.
Perry Carpenter: Yep.
Mason Amadeus: It's kind of like if you fed that into Sora and then had Sora just paint all the details over it, right?
Perry Carpenter: Yeah, exactly.
Mason Amadeus: That's really cool. Man, I want to play with it, but I just don't have either an accessible enough toolkit or the time to get deep enough to build my own integration. So someone out there, please hurry up. Put it in Blender.
Perry Carpenter: There you go. Put it in Blender or send us a link if you've already found it.
Mason Amadeus: Yes, please. Oh, gosh. Yeah, please email us links, hello@8thlayermedia.com. Put F-A-I-K in the subject line. Coming up next, we've got a little Dumpster Fire of the Week. Stick around. [ Papers Shuffling ]
Unidentified Person: The Safe Files. [ Papers Shuffling ] [ Music ]
Perry Carpenter: If you've been in or around the San Francisco area, you may have seen something that is raising eyebrows, and it is a set of billboards that simply says "Stop Hirring Humans."
Mason Amadeus: I saw those.
Perry Carpenter: And has pictures of people on it or AI-generated people.
Mason Amadeus: And people have been vandalizing them, haven't they? People got mad.
Perry Carpenter: Yeah, people are frustrated because you're seeing the AI world talk out of both sides of its mouth, right. It's AI is here. It's here to help humanity. We're going to figure out how to manage things like job loss and everything else, because we do know that there are capabilities that humans do that are going to be replaced or heavily streamlined because of that. And companies being what companies are, are looking after how do we use this to gain more efficiency, not hire as many people, or potentially let people go? And when you have the economy that we've been dealing with and the fact that there are a lot of people that are afraid right now, when they were seeing these signs, it raised a lot of concerns and it sparked a lot of conversation.
Mason Amadeus: Yeah, it's not the most tactful ad campaign; that's for sure.
Perry Carpenter: No, not at all, and even people like -- in a different context, people like Sam Altman and the CEOs of Google and Microsoft for the past couple of years with Gen-AI have been saying how much this augments the workforce, and they've not been necessarily explicitly saying you're not going to need to hire as many people or you're going to get to let people go, and that messaging has started to change over the past couple of months where they are saying that this is really going to help you from a personnel standpoint. So that's coming out a little bit more, but this company, it's called Artisan, is the one that put out these billboards, and actually, on their website, on their blog, they talk about why they did this.
Mason Amadeus: Interesting. I would love to know.
Perry Carpenter: And they lean into the controversy.
Mason Amadeus: Oh, boy. Okay, so they're being little stinkers?
Perry Carpenter: Yeah, they created it to be provocative. And so last year -- actually two years ago now -- in 2023, they had a set of boards that they put out that were like Artisans. So these are AI-generated agents essentially. So "Artisans won't complain about work-life balance."
Mason Amadeus: Oh, my God.
Perry Carpenter: Humans are so 2023. "Artisans won't come into work hung over. Hire Artisans, not humans." And "The era of AI employees are here." They had the appearance of an AI SDR, like a sales rep and people that are really working with customers trying to develop the pipeline.
Mason Amadeus: Gosh, this feels like PodCube. This is like dark satire. That's --
Perry Carpenter: Yeah, it does. It feels like a Black Mirror type of thing, right?
Mason Amadeus: And are they being ironic? They must be. Or do they believe their own sort of messaging?
Perry Carpenter: Well, I mean, that's what they're selling. So they're actually selling the thing, but I think they are -- they're leaning into the dystopia of it all a little bit.
Mason Amadeus: Yeah. How far into their cheek is their tongue is my question?
Perry Carpenter: I have no idea. I mean, they know what they're doing, and they know that they're creating controversy, and so they're capitalizing on it. In fact, their newest billboard says "Stop Hirring Humans," but the word hiring was misspelled. They accidentally put two R's in it, and it went up that way, and it has the hiring has the little red underline like if Microsoft were to do to show you that it's misspelled because their first instance of it was written by a human that accidentally misspelled it, and they're like, oh, that's gold because the AI wouldn't make that mistake.
Mason Amadeus: As someone who is into marketing and clever marketing, there's a tiny part of me that kind of dislike/respects them for that because that's pretty smart.
Perry Carpenter: Yeah.
Mason Amadeus: I don't love this treatment of this tone when it comes to like a lot of people's anxieties and job security.
Perry Carpenter: Especially in San Francisco. I mean, San Francisco is like this this interesting dichotomy because you do have Silicon Valley, and it is the tech hub of the world and where AI's genesis really is in a lot of ways right now.
Mason Amadeus: But its unhoused population.
Perry Carpenter: Yeah, the homeless population and the people that are just living destitute lives are there, and then they're walking by these things. And some of these people that are homeless, everybody that's on the street has a story. But some people may have been a victim of a job replacement, downsizing, or something else, and within weeks went from somebody who was seen as very successful to somebody who is on the streets. And it has to be painful to walk past these signs and see, potentially, the next wave of job loss being thrown out there is something that's kind of a half joke.
Mason Amadeus: Looking at their website, this is everything I'm not excited about when it comes to AI. Literally, the first one they give as an example is like a customer service AI. That sucks. I don't want to talk to a robot. It's hard enough to get service from any of these companies. And then the next thing they talk about is an automatic lead generator that writes specific emails and scrapes the web to try and promote your whatever.
Perry Carpenter: The thing with AI personas and agents is that we will, essentially and eventually, get to the point where we have our own AI assistants that go talk to everybody else's AI assistants on our behalf, and then they come up with a decision and just let us know.
Mason Amadeus: Yeah, and what are we automating? Why are we doing the things that we do?
Perry Carpenter: Yeah.
Mason Amadeus: I think this is the real slop factory when it comes to AI. It's not people generating pictures. It's this.
Perry Carpenter: I've said the word interesting too many times, but I am really interested to seeing how this turns out for them over the next couple of years.
Mason Amadeus: Yeah. I mean, there's a certain kind of person who believes that any publicity is good publicity, and we certainly have a lot of examples in our culture to back that up. So --
Perry Carpenter: I'm assuming that they're going to generate enough bad press that they go out of business themselves or they sell the technology to somebody else that does it with a clean name.
Mason Amadeus: And I mean, they're all going to walk away with a big old paycheck, probably.
Perry Carpenter: Eventually, the same functionality is going to be incorporated into Salesforce and Google's systems and everything else. So these folks will probably end up getting acquired. The people that founded the technology will sell the technology and sell themselves. Something else, they'll walk away with money and do another startup later on.
Mason Amadeus: I'll tell you, I've done something very clever to sort of preempt this shift. I have 10,000 unread emails. I just don't look at them. [ Laughing ]
Perry Carpenter: You could have a I read those for you.
Mason Amadeus: Yeah, honestly, I would because there's already so much spam, and there's just going to be more of it. People are going to -- the thing about this kind of agent is that it's doing something. If you are not going to take the time to put in the effort to write this text or look at my thing and understand it and ask me for something, why should I put in the effort to read what you send me? Especially in this kind of --
Perry Carpenter: So an interesting way to end this. So Apple Intelligence was trying to solve for just that, and so does Google and Gmail. In your Promotions tab in Gmail or there's a there's a new tab in Apple's mail client that essentially anything that's not human written that is just spam or newsletter stuff or other things like that, go into this tab for you. So what the mail systems are trying to do right now is say, what is human-to-human conversation and then what is system-to-human conversation? And if we can take the system-to-human conversation and turn that into system-to-system conversation, at least we're helping manage our own attention a little bit better.
Mason Amadeus: Isn't that kind of stupid, though?
Perry Carpenter: It kind of it kind of sucks. I mean --
Mason Amadeus: It's like if I started a factory that mails used diapers to your house, so you had to buy a robot to put the used diapers in the trash. You don't want them. I don't care about sending them. Why do we do this to each other? Welcome to capitalism.
Perry Carpenter: Yeah, yeah. It's all trying to figure out how do we how do we get somebody in the funnel so that we can figure out how to make our next buck?
Mason Amadeus: Yeah. Well -
Perry Carpenter: Anyway, that's The Dumpster Fire.
Mason Amadeus: That's that is the Dumpster Fire of the Week. And who knows? Maybe next week I'll be replaced by an Artisan AI and you won't even know.
Perry Carpenter: Ooo.
Mason Amadeus: Thank you for tuning in. Happy FAIK Files Friday. Welcome to the new year. I hope it's going to be a good one. There's going to be a lot of exciting developments.
Perry Carpenter: Yeah. If anything, it's not slowing down. So we will see you next week.
Mason Amadeus: Catch you on the flip. [ Music ]