The FART FARTS Files...
Mason Amadeus: Live from the 8th Layer Media Studios in the back rooms of the deep web, this is "The FAIK Files."
Perry Carpenter: When tech gets weird, we are here to make sense of it. I'm Perry Carpenter.
Mason Amadeus: And I'm Mason Amadeus, and this week we've got a fun show for you with a lot of different things. We're going to start it off by talking about how Google has dropped Gemini 3 and a suite of new tools, including a code editor, and it is a pretty remarkable model.
Perry Carpenter: Oh, nice. Then we're going to move on and talk about how AI has been shown to implant false memories in people, specifically AI-edited audio and video.
Mason Amadeus: That is terrifying. Our third segment, we'll move on to something just a little bit weird, sent in by Bullethead from our Discord community, this parallel generation of image and text to make models more able to understand the context of images while they're editing. I'll explain it in the segment. It's called Parallel Multimodal Large Diffusion Language Models for Thinking-Aware Editing and Generation. So we'll kind of try and cut through that a little bit.
Perry Carpenter: That's a little bit of a mouthful, yeah.
Mason Amadeus: Yeah.
Perry Carpenter: And then we're going to end out with a little bit more on the manipulative front, and this is some of the disinformation campaigns that have been coming out of Russia and some that have been coming out of the Middle East, and what that means for people that are analyzing this stuff.
Mason Amadeus: A grab-bag of tech and deception this week. So sit back, relax, and remember, it's not a bubble. It's just a circular investment strategy. We'll open up "The FAIK Files" right after this. So Google dropped Gemini 3, and it is honestly pretty darn impressive, not just Gemini 3, the model, but a whole suite of tools and other things attached to it. And we're going to break down sort of all of it, but I'm going to start in a weird place because I have a demo for this --
Perry Carpenter: Okay.
Mason Amadeus: -- that I think is going to be fun, Perry. So one of the things that they launched along with Gemini 3 is this product called Antigravity, which is their agentic coding platform.
Perry Carpenter: Right.
Mason Amadeus: I actually have it downloaded right here, and I got it all set up so the viewers watching will be able to see this. It is basically --
Perry Carpenter: So this is, essentially, like the equivalent of a vibe coding platform like Lovable or something like that, right, but it's all within Google?
Mason Amadeus: Yeah, and this is an app that you download to your computer, so it's not something you run through your browser --
Perry Carpenter: Nice. Right.
Mason Amadeus: -- although the models do run through the browser. But you can see if I -- this is their agent mode window right now, which is like the custom thing that they've built. If you hit Open Editor, you can see, if you're familiar at all with VS Code, it is -- it is just VS Code because that's open source.
Perry Carpenter: Yup, yup.
Mason Amadeus: So underneath it is a very familiar interface, a Visual Studio Code, but the important part here is the Agent Manager. Now, there's a bunch of different setup steps. We'll talk about that when we actually do the breakdown. But for this demo, I have set it into Turbo Agent Only Mode, where it's not going to ask us for review. It's just going to go off of what I tell it to do and basically come back when it has finished. And so what I thought we could do was ask it to make us a website for "The FAIK Files".
Perry Carpenter: Okay.
Mason Amadeus: So I was thinking something like create a website using Eleventy -- which is a static site generator -- uses JavaScript that I'm familiar with, I've been working with lately -- for the podcast, "The FAIK Files" part of the N2K network. You can find all the information about it online. What do we want for our website, Perry? What else should I put in here?
Perry Carpenter: We should tell it to do bios for us, so mention our names and say that it needs something highlighting the hosts.
Mason Amadeus: Okay, make sure there are bios for Perry and Mason.
Perry Carpenter: Do we want to have it -- have an embedded mini player?
Mason Amadeus: Yeah, that's a good idea, embedded mini player for the podcast RSS feed.
Perry Carpenter: Display and highlight show art.
Mason Amadeus: The show art carousel and any other features you think of that would make the website really, really cool. Let's go ahead and be vague --
Perry Carpenter: There we go.
Mason Amadeus: -- because this new model is much better at kind of gauging your intent from vague instruction.
Perry Carpenter: Okay.
Mason Amadeus: Any other features you think that make the website really, really cool. Make sure it is easy for us to update with new content and that it is responsive across devices. Think like a senior developer. I just -- I've habitually been putting that in my prompts to see if it helps at all.
Perry Carpenter: Okay.
Mason Amadeus: And then I think we'll fire this off, and then we'll talk more about --
Perry Carpenter: Like when you say senior developer, you mean like in rank, not in age, not like somebody that started with Cobalt.
Mason Amadeus: Yeah, yeah. Think like an old guy. Think like a very, very old programmer, who -- I'm add gonna add -- who is intimately familiar with Eleventy, just so that maybe that'll help point it towards more of that. But I think -- I have mixed expectations here. I don't really know what's going to happen with this because it's only been out for a couple days. I think this dropped --
Perry Carpenter: Right.
Mason Amadeus: -- merely two days ago. I've had a little experience using Gemini 3 Pro in the chat interface, and I've been doing some coding tests with it, and I've been really impressed, but I've not tried this yet, so we'll fire it off.
Perry Carpenter: I haven't either.
Mason Amadeus: We'll send it into the background after -- let's just see what it starts to do here. It looks like everything is sort of spinning up. And you know what? I'm not even gonna look. Let's just let it -- there we go. It has started. It's going. It's thinking, and it's gonna work. And because I set it to Turbo Mode, it's not gonna ask us any questions. We'll check back up on that at the end.
Perry Carpenter: I'm so scared.
Mason Amadeus: Yeah, me too. Who knows what it's gonna be like? So --
Perry Carpenter: You may -- you may crash the internet.
Mason Amadeus: Or I might delete stuff off my computer. Who knows? That's kind of the risk too with these agentic systems. There's actually a --
Perry Carpenter: Over the past few weeks, we've had like an AWS outage. We've had a clear Cloudflare outage. You might cause some kind of weird Google backend issue that affects the rest of the world.
Mason Amadeus: Yeah, somehow it's gonna end up reconfiguring DNS settings on some like internet backbone. Who knows?
Perry Carpenter: It's always DNS.
Mason Amadeus: It's always DNS. There's a YouTuber Jeff Gearling, who is a broadcast engineer who has -- he sells shirts that just say "It was DNS," and I really want one.
Perry Carpenter: Yeah.
Mason Amadeus: So Gemini 3, it dropped on the 18 of November, so two days before time of recording, which is the 20th and three days before you'll hear it. And there's a bunch of information about it. There's like a whole product collection page that you can go check out, and we will link it in the show.
Perry Carpenter: You gotta read that title line. A new era of intelligence with Gemini 3.
Mason Amadeus: Gemini 3 is our most intelligent model that helps you bring any idea to life. Yeah, this is littered with that kind of language, including like -- there's something here. Where is the line? It's talking about like this is the next step towards AGI and all of this sort of thing. So it is a bit self-aggrandizing. However, what they have here is pretty remarkable.
Perry Carpenter: Right, right. Well, and we should say -- we were mentioning right before we hit Record that like about a year ago, even six, seven months ago, I think we were dunking on Gemini quite a bit because it just wasn't where it should have been for being a Google product and -- but we had always said this should be way better because it's Google, and we think that it will, at some point leapfrog. And by all estimations and all the other folks that have looked at this, this seems to be the real leapfrog moment, where everybody's looking at it and saying, all right, this is actually the best thing on the market right now.
Mason Amadeus: Yeah, very much.
Perry Carpenter: And it shows that progress for these kinds of systems is not actually slowing down. We've not hit a wall.
Mason Amadeus: Yeah, we have not yet hit a wall in the scaling race and whatnot. And, yeah, we -- I think we have said from the beginning, you know, Google has the infrastructure. They have the data. They have the information. They have the user base. So it was only a matter of time, and this is a huge step towards that. We'll get deeper into the benchmarks in a moment. This overview here is kind of just a lot of texts and bloviating about how cool it is. But I want to get into like the harder numbers.
Perry Carpenter: Let me -- let me make one more remark then while you're -- while you're getting to that part because we talked about not hitting a wall. The other thing is everybody is concerned about power consumption and rightly so. However, there's an interesting thing that came out of OpenAI, where Sam Altman said -- let me get the actual quote -- GPT-5.1 Thinking High is about 300 times cheaper than -- per task than o3-preview.
Mason Amadeus: Oh, really?
Perry Carpenter: Which, yeah. And so that is a year later, their highest model is now 300 times cheaper to run than the highest model a year ago.
Mason Amadeus: Wow.
Perry Carpenter: And that's actually really encouraging because 5.1 is actually faster too. So we're moving in the right direction with some of the power stuff. At the same time, though, everybody's using these more and more, so the demand is still growing.
Mason Amadeus: Yeah, and we need those efficiency gains because, again, the power consumption and water consumption becomes a critical problem when you get to these massive deployments --
Perry Carpenter: Yeah.
Mason Amadeus: -- and especially the training cycles. Honestly, that is why I have been more -- when I use a cloud-based AI, I lean towards Gemini because they're the only one that's been pretty transparent, even remotely transparent, about their like water and power use. I still haven't seen any like numbers from OpenAI about those specific -- like their power consumption during training.
Perry Carpenter: Yeah, want to dig into that.
Mason Amadeus: Yeah. So I pulled up the page for Google Gemini Pro here because it's got some benchmarks for us to look at. And they say here, our most intelligent model yet sets a new bar for AI model performance, and that definitely seems to be true. On Humanity's Last Exam, Gemini3 Pro got 37.5%, whereas the previous leader had been GPT-5.1, 26.5% there. Also on Terminal Bench 2, it got 54.2% compared to GPT-5.1 at 47.6. Seventy-two percent on Simple QA Verified versus Gemini 2.5, the previous leader, the previous iteration at 54.5%.
Perry Carpenter: Okay.
Mason Amadeus: The long context performance on the MRCR V2 -- I don't really know anything about that benchmark, but compared to the other ones, like we're looking at Gemini 2.5 Cloud Sonic 4.5 and GPT-5.1. And the new Gemini 3 got 77%, whereas the previous highest number was 61.6% from GPT-5.1, and Claude had 47.1%. This thing is spicy, multimodal input, text-only output. It has 1 million input tokens, 64,000 output tokens. I do feel like I'm just shouting numbers, you know? And so I want to talk about like --
Perry Carpenter: Well, I think that's, yeah. So it's one thing to see the benchmarks. It's another thing for people to like actually use it and say my vibe feeling about it feels like it's better, and it seems like we're getting that too. Now, I'll be honest. I've not opened up Gemini since Gemini 3 dropped, so I don't really have just a feeling about it yet because I've been traveling and focused on other things. But from what I'm hearing from a lot of the folks that I follow, like they feel a difference, a subjective difference.
Mason Amadeus: There absolutely is. I felt it in the brief time I've been using it. And I looked up -- I like to go to Reddit to see what people are saying because, you know, Reddit is full of some of the meanest and bluntest people you'll find on the internet.
Perry Carpenter: Yeah.
Mason Amadeus: And pretty universally, I saw people saying it was really impressive and a step up, people saying that it solved problems --
Perry Carpenter: Nice.
Mason Amadeus: -- that they couldn't get Gemini to solve before in other models. So yeah, the reception has been pretty good.
Perry Carpenter: Yeah. The other thing that I've heard is really good with this one is things like spatial reasoning and --
Mason Amadeus: Yes.
Perry Carpenter: -- being able to look at and understand what's on a computer screen.
Mason Amadeus: Yes, that is the other thing. With this -- with the Antigravity platform we're running in the background, it can launch a dev server and like interact and run tests on your website or whatever it is you're having it code.
Perry Carpenter: Yeah.
Mason Amadeus: So it's really good at the screen space stuff. It's also got really, really good visual analysis. I saw people saying this is the first model that could read their handwriting, which is fun.
Perry Carpenter: Oh.
Mason Amadeus: So basically like incremental improvements in all arenas --
Perry Carpenter: Yeah.
Mason Amadeus: -- and pretty tangible incremental improvements. Here's a feature that I want to highlight. I want to balance the difference between yelling numbers and like feelings. And this is a very like tangible feature that I think is very cool, and it's something I've only seen Gemini doing. And they had been doing this a little bit before, but they're really leaning into it now, which is generative UI, where --
Perry Carpenter: Ooh, okay.
Mason Amadeus: I don't know how much you poke around with Gemini, but you can ask it --
Perry Carpenter: I had heard about this, yeah.
Mason Amadeus: Before, I had it -- I asked it to make me a quiz about skeletons or like a quiz about bones. It was for a silly Halloween thing, and it presented me this like interactive multiple choice quiz with all of these things. And I thought, oh, that's neat. It like made a little layout and UI, and they've really leaned into that now, where different prompts will elicit the model to essentially immediately vibe code you a little web app --
Perry Carpenter: Yeah.
Mason Amadeus: -- to teach you or to help you. So the examples they give are like if you ask about fractals, it could make you like this cool, artistic fractal interaction thing where you can like click through and learn. Or you can have a training camp for math. And so rather than like vibe coding your own sort of website, it will vibe code its own thing to communicate with you in the chat platform.
Perry Carpenter: Yeah.
Mason Amadeus: And that's really engaging and very cool, as long as the, you know, it gets the facts right and whatever it is you're doing is not full of hallucinations and whatnot.
Perry Carpenter: This is similar to like the direction that Google's NotebookLM is going to. So remember, it started with like these custom audio overviews that were podcast-like, and now they've evolved that where we'll do a custom video overview as well, where it adds like background images and charts. And, I mean, it's very stylized. It gets old a little bit quickly for me. But still, it's leaning into the fact that because when you're -- when you use NotebookLM, specifically, you're source grounding it really well, and so it can pull facts, figures, and things like that fairly accurately, and represent those in relation to one another, and kind of pull out the narrative. So they're -- it's leaning into that. But this one seems like it's taken that at internet scale, which is really cool. And the examples that I've seen, like the one that you get on the screen with the Van Gogh stuff is really cool as well. You know where this is going, though? This is going to application and commerce, where you'll be able to go to a store, and it will know enough about you that it will present you a custom, just for Mason, this is your storefront. And what kind of questions, what kind of things would you like to see? And let me surface those and custom-build texts that will appeal to you based on the nuance of language that it understands about your tastes and needs and preferences.
Mason Amadeus: Oh, and get ready for a wave of that on top of just like print-on-demand sites, where it generates --
Perry Carpenter: Yeah.
Mason Amadeus: -- pictures to print on coffee mugs and crap for you and whatnot.
Perry Carpenter: Yup.
Mason Amadeus: Yeah, so like the thing is, I guess, with -- and it's just a recurring theme in this show is that like a very cool technology. It's immediately going to be used for the most annoying BS on the planet. But at its core, it's very, very neat. Like this, I particularly like the like learning interface, and I know that's a problem --
Perry Carpenter: Yeah.
Mason Amadeus: -- when it comes to accuracy in certain things. The knowledge that this seems to have and like the reduction in hallucinations is pretty remarkable, too. I've got a blog post from Ethan Mollick that I'll touch on, but I really encourage you to check out where he talks about this. But yeah, a last note on the generative UI experiences is like the fact it can make interactive little things for you as a teaching tool, that gets me really excited.
Perry Carpenter: Yeah.
Mason Amadeus: Because primarily what I end up using these things for is learning stuff. I use it for helping me with code, but I try and do it in a way that helps teach me what the code is doing and like how to do things better and new patterns, or like to double-check stuff about electrical engineering or whatever kind of projects I'm working on, and also the potential for making like visual demonstrations because I'm working on the YouTube channel Wicked Interesting, and I need to do a lot of data visualization at different times in different ways.
Perry Carpenter: Yeah.
Mason Amadeus: And so something like this that could make that easier than me going into Blender and manually making every single thing, that, I mean, that's pretty cool. I don't know that I will do that necessarily, but the potential that is on display here is very interesting to me.
Perry Carpenter: And I do think that some like large companies -- like I could imagine IBM or Dell, you know, something that has all these different business divisions and all these different products. You want to have a customer experience on your website where somebody can just go in and ask a question and toggle a few preferences, and then it surfaces in a -- the coolest way possible the information that they're asking for, like directly asking and answering their questions based on the data that you have available and the way that they ask the question.
Mason Amadeus: Yeah, I mean, I really think -- and it seems very obvious to me that the future is these systems we have now are an amazing interface layer between a base of knowledge, data, and information that is maybe structured for a computer or a bit of a mess and a human user because they're a --
Perry Carpenter: Right.
Mason Amadeus: -- great translation layer there --
Perry Carpenter: Yup.
Mason Amadeus: -- and then to visualize it that way. It's so cool. Before we check on our website progress, I do want to just pull up this blog post from Ethan Mollick because I feel like it helps put things in perspective. Because that --
Perry Carpenter: Cool.
Mason Amadeus: I struggle with getting too hyped or too cynical in sort of oscillating ways where like I get very excited about something and then become more and more cynical as I see it used for more and more crappy stuff. And so I think that this post helped put that in perspective. We're not going to read the whole thing, but we'll link to it. This is from Ethan Mollick. If you don't know who that is, I'd be very surprised if you're listening to this show.
Perry Carpenter: Right?
Mason Amadeus: So he says, I've been testing Google's new Gemini 3 model. It is very good, but rather than give you benchmark numbers, I'm going to show you how far AI has come in three years by asking the AI to demonstrate it. He shared a tweet with Google Gemini that he made where he said I posted about how image-generating AI has gotten exponentially better in the last month, while a new text model was released for GPT-3 today. AI can now write rhyming poems and acrostics and limericks and explain how a candy-powered FTL drive can help me escape from otters. So that was back in 2022 and everyone was like, oh, silly. It's like those memes about writing an Applebee's commercial by training an AI on every Applebee's commercial, yada yada. Was very nascent. So he sent that tweet to Gemini and said, show me how far AI has come since this post by doing stuff. And Gemini said, I can definitely show you the difference. The tweet you shared is right on the cusp of the modern AI boom. To demonstrate how far we've come, I'm not just going to write about it. I've built you a fully interactive, playable, Candy-powered FTL starship simulator. In 2022, AI could describe the engine. In 2025, AI can code the engine, design the interface, and let you pilot the ship yourself. And he goes on to say, indeed, it built me a tiny game where I had to use the power of Candy to escape otters, featuring small poems and an ongoing set of amusing updates, which, like it really does --
Perry Carpenter: Pretty cool.
Mason Amadeus: Yeah, right? It's been three years, and like the underlying technology for all of the societal ills atop it, the underlying technology is incredible. And I just feel like that that helps really illustrate it. And he talks about --
Perry Carpenter: Yeah.
Mason Amadeus: -- Antigravity, which we're going to check back in with in a moment. He did this experiment, though, that I want to highlight, where he gave Gemini 3 access to a directory of old files he used for research into crowdfunding a decade ago. It was a mishmash of files labeled things like project underscore final underscore seriously underscore this time, you know? And data in random formats and whatnot. He told Gemini, figure out the data and the structure, the initial cleaning. Get it ready to do a new analysis and find new things. And he says it did. And then he gave it an assignment to create a research paper, an original research paper, using this data.
Perry Carpenter: Oh, wow.
Mason Amadeus: And he said, specifically, with no further hints, I wrote, great. Now I want you to write an original paper using this data. Do deep research on the field. Make the paper not just about crowdfunding, but about an important theoretical topic of interest in either entrepreneurship or business strategy. Conduct a sophisticated analysis. Write it up as if for a journal. And then the AI, after a couple of vague commands, he says like build it out more. Make it better. He got a 14-page paper, and then he reviewed it, and he says, so is this PhD level intelligence? In some ways, yeah, if you define a PhD level intelligence as doing the work of a competent grad student at a research university, but it also had some of the weaknesses of a grad student. The idea was good, as were many elements of the execution, but there were also problems. Some of its statistical methods needed more work. Some of its approaches were not optimal. Some of its theorizing went too far, given the evidence and so on. Again, we've moved past hallucinations and errors to more subtle, often human-like concerns. Now, obviously this isn't universally true, but in this case, he said that it came up with like its own measure of figuring out how unique a crowdfunding idea was using natural language processing to compare its description mathematically to other descriptions, and like these other kind of cool --
Perry Carpenter: Yeah.
Mason Amadeus: -- novel angles, and most of it held some water. It was like far more impressive than just write me a limerick.
Perry Carpenter: Right.
Mason Amadeus: And it was this very last line that made me want to share this, which was, three years ago, we were impressed that a machine could write a poem about otters. Less than 1,000 days later, I'm debating statistical methodology with an agent that built its own research environment. The era of chat bot is turning into the era of the digital coworker. To be very clear, Gemini 3 isn't perfect, and it still needs a manager who can guide and check it, but it suggests that human in the loop is evolving from human who fixes AI mistakes to human who directs AI work, and that may be the biggest change since the release of ChatGPT.
Perry Carpenter: Yeah.
Mason Amadeus: I don't agree wholly with that. I still think that like it should be a much more cooperative thing than just guiding the AI's work. But I see what he's pointing at, and I think I agree with just the scope of how --
Perry Carpenter: Yeah.
Mason Amadeus: -- impressive these changes have been.
Perry Carpenter: I mean, it is -- it is crazy. The other thing is it's like no matter how good it gets, we're still frustrated with it. Like it had come out with something that would have taken us two or three days to pull together and to think through and brainstorm and like write something. We'll look at it and go, God, you're so stupid.
Mason Amadeus: Right.
Perry Carpenter: Like why'd you -- why'd you say this or think this in this paragraph?
Mason Amadeus: How did you mess this up based on my vague instructions that I gave you --
Perry Carpenter: Right.
Mason Amadeus: -- and while I was holding a coffee in the other hand. Yeah.
Perry Carpenter: It is -- it is really funny how we're just like always gonna -- and some of that's like the brilliance that we have within our own minds and like finding connections and surfacing past contexts that we've had where we can naturally find those weak spots. But at the same time, I think we move really, really fast to like discount all of the other stuff that it's done, and we just hone in on the like, oh God, that's wrong.
Mason Amadeus: And I think it's partially because of the uncomfortability. Like we were so keen to tell -- to say that AI was very human-like, until it became pretty close to human-like, and we got more and more reticent --
Perry Carpenter: Yup.
Mason Amadeus: -- started moving the goalposts further and further as to what we consider intelligence.
Perry Carpenter: Yeah.
Mason Amadeus: And I think that's sociologically interesting.
Perry Carpenter: The other thing is I don't know that like when we grew up with scifi shows like Star Trek, and people would just like, go, hey, computer, do this for me. Or in the Jetsons, where they've got the little automated robot that takes care of their house. We never really thought in those situations about the intelligence and the social and ethical issues that came along with that because they were just interacting with a machine. We kind of bought into the world, and now we're having to go, oh, wait, this thing could get it right. This could get it wrong. There could be an emerging intelligence in it that we don't understand, that approaches some kind of consciousness. What do we do with that? Or if it's robots, are we just kind of building things that, at some point, could be kind of sentient, and we're enslaving them? You know, what are -- what are all those things? And I'm not saying that we're at that point or anything like that or even that it's going to be possible, but those are the ethical issues that we have to grapple with. And in those situations with those scifi shows, we never saw that grappling with. We just saw the cold interaction of, computer, make me a cup of coffee or, computer, help me brainstorm through this.
Mason Amadeus: Yeah, and I think grappling with that nuance is something that we weren't really ready for --
Perry Carpenter: No.
Mason Amadeus: -- particularly with the socioeconomic place we were in when this technology kind of came out.
Perry Carpenter: Yeah.
Mason Amadeus: So I've checked back up on our website, and I realized it was sitting waiting for me to approve the overall action plan and let it run some terminal commands. So we'll let that cook in the background, and I think at the end of this episode, we can -- we can take a look at what it has cooked up.
Perry Carpenter: Cool.
Mason Amadeus: Because right now it's just, here, I'll give you a quick glance at what's been going on here since this is the segment about it, and we'll clip this out for YouTube. But you should watch Segment 4 if you want to see the result. So it started doing research. It created this implementation plan over on the side with, you know, just various proposed structures and things. I didn't read a single lick of this. I'm just clicking OK, Accept. I guess I'll click Accept even though that's been there for a while. Let's see if that's okay. I wish I could get it to auto-accept, but this is -- I'm doing -- what I'm doing is inadvisable. I will say that.
Perry Carpenter: You need another agent that just continually types in, Go for it!
Mason Amadeus: Yeah, just like -- that looks for a blue button and clicks it no matter what.
Perry Carpenter: Yup.
Mason Amadeus: So we'll let that cook in the background. I will, oh, Agent terminated due to error. Ooh.
Perry Carpenter: Ooh.
Mason Amadeus: All right, well, I'll see what's up with that. We'll move on to a break, and then we're going to jump into our next segment. What you got coming up in our next segment, Perry?
Perry Carpenter: Let me look at the -- look and see. Oh, paper from MIT on AI-edited images and videos that can change our memories.
Mason Amadeus: That's right. Oh, boy, okay, buckle up.
Perry Carpenter: Yay.
Mason Amadeus: That's on the way.
Perry Carpenter: So I want to talk real quick about something because I think we've expected this, and it's interesting to see it in actual confirmed research. So we know that as humans we have very fallible memories, in fact. Even though every court case wants like an eyewitness for something, we've seen over and over and over again that eyewitness testimony is sometimes the least reliable testimony because our memories are fallible. And sometimes the more we try to recall the same thing, the more we subtly change it within ourselves. And we did see, even like in the Satanic Panic of the 1980s and early '90s, all the false memory stuff that was being done because somebody would go in for counseling, and then they would, you know, put them in a hypnotic or semi-hypnotic state, and then subtly suggest things that may have happened to them. I think, all with goodwill and good intent and maybe some sensationalism, for sure, but trying to help people get to the problem or, you know, the root issue of some things, and out of this came many, many false memories and accountings of like ritual abuse and everything else. And then, of course, that caused a panic. And so people would be hearing about it, and somebody else would go, you know what? I've never felt right about my life. Let me go in and get hypnotized, and then somebody would be going, you know, tell me if you remember being in a circle and people chanting around you.
Mason Amadeus: Yup.
Perry Carpenter: And of course, if you're suggesting things like that, then they can start to become part of your thing. Because it is the -- it's the conundrum of telling somebody don't think of the big pink elephant. And in order --
Mason Amadeus: Yup.
Perry Carpenter: -- to not think of the big pink elephant, you have to conjure up a big pink elephant in your mind and kind of cross it out.
Mason Amadeus: And isn't like the whole cornerstone of like those sorts of hypnotic things the idea of suggestibility and like being open to suggestion and whatnot?
Perry Carpenter: Exactly.
Mason Amadeus: Like putting yourself in a state where you're like, well, maybe I do, and then you start imagining it, and you're like maybe this is a buried memory or whatever.
Perry Carpenter: Yeah, yeah.
Mason Amadeus: Basically gaslighting yourself.
Perry Carpenter: And imagination and memory are like very, very mixed, right? Because you can remember the things that you've imagined.
Mason Amadeus: Yeah.
Perry Carpenter: Like if you were daydreaming the other day and trying to postulate something, you know, the beauty of memory is that, as you're brainstorming and postulating, you can go back and remember that later and build on it.
Mason Amadeus: Not to sound like a -- not to sound like someone in a stoner movie, but you just blew my mind, man. You can remember the things you imagine. I -- but like I never thought of it that way. That's actually like, yeah, those --
Perry Carpenter: Yeah.
Mason Amadeus: -- two processes are very interleaved.
Perry Carpenter: And then so like if I was talking to you and say, you know, Mason, imagine your childhood. You remember a time that you were like on a swing set, and you were there, and your parents were near you? Do you -- do you remember that time when you were on the swing set and your parents were near you, and everybody's playing and having a good time? And then like you looked around and nobody was there, all of a sudden? Like you don't remember where your parents were, and like you were a little bit scared, right?
Mason Amadeus: I could -- I could gaslight myself into that because, honestly, I frequently will have people tell me things like, hey, remember when you did this thing? And I will have no recollection --
Perry Carpenter: Yeah.
Mason Amadeus: -- and I'll just have to, you know, I think that's a pretty common experience. And so as you were saying that, I was like, I -- maybe I can actually remember that --
Perry Carpenter: Yeah.
Mason Amadeus: -- you know, even in this moment, yeah.
Perry Carpenter: Because it sounds like just plausible enough, right? Because you do have memories of yourself probably like in some kind of playground or on a swing set, and you have memories of yourself with parents or parental figures in that. And you have memories of times where like you've looked around, and people that you've wanted to be, you know, like your parents or somebody else, like they stepped aside, and you don't remember where they were, and you were like a little bit freaked out, and you can smash all those things together really easily.
Mason Amadeus: Absolutely.
Perry Carpenter: So that's what's going on, essentially, with digital memory recall and -- but what I mean by that is all of us have our cell phones. We're taking pictures of stuff all the time, or we're building, you know, taking videos. And increasingly, on our cell phones and on any device that we have, we have like these AI tools now. I think Google was the first to really add the touch and remove the annoying thing type of function within your photo.
Mason Amadeus: Yes, yeah.
Perry Carpenter: Which, at the time, years ago, I was like that's really cool, but that's also really dangerous because let's say, you know, you have this memory of -- or you have this photo of you and your family on the beach, and there was another person in there, or there was like a seagull or something that was just in the wrong place. You like zap that thing out. Well, now every time you look at that, you're reinforcing the memory that's not actually true.
Mason Amadeus: A modified memory, yeah.
Perry Carpenter: Yeah. Or you're in an alley, and there's a distracting person, or there's a distracting bit of graffiti, and you just tap that and say get rid of that. Get rid of that. And then again and again, you're like, in this scene, man, we were at the beach, and it was just us, and it was perfect, and then 10 years later, that's what you remember. Or in this alley, it was, you know, pristine, and, you know, I don't know what kind of memory you're going to build in an alley, but there wasn't the person over there that had just peed by the dumpster type of thing.
Mason Amadeus: Right, right.
Perry Carpenter: It was, you know, it was clean. And that is becoming like the norm. And so this study came out from MIT. Let me go ahead and share the tab for that.
Mason Amadeus: And that exists on a spectrum from banal to pretty important, right? And you may not --
Perry Carpenter: Oh, yeah.
Mason Amadeus: -- always know what is.
Perry Carpenter: Yeah, yeah. I mean, it's not just all about not seeing the person that's peeing in the corner. It's -- or, you know, getting rid of the, you know, the friend that you broke up with in your childhood so they're not in this image anymore of you with your family that you do still get along with. It's way more devious than that because if I can start to manipulate videos or images and then show those to you more and more times, then I can actually create a false reality for you.
Mason Amadeus: Yeah.
Perry Carpenter: So the MIT study, essentially, what they did is they would show somebody an image, and then they would give them a little bit of time, so a time delay. And then they would ask them about the image. So image time delay and then recall. And this is like when I was doing a lot of mentalism and fake psychic work, one of the things that you would always do at the end of a trick if you'd manipulated something up front is you would try to manipulate their memory by the end of it by saying, hey, remember we did this. And you might put some steps out of order in the way that you recalled it. Like if I -- if I asked you a number, and you said 42, then I'm going to actually write that number in real time in a secret way without you being able to see it. So I might pull out an envelope, and I might be kind of writing through that envelope. I'm not going to tell you how that's done, but writing through that envelope after you said 42. Then I give that to you. Or if I've got multiple predictions in my pockets, like if I had set out five different things and had you pick one, and then on my person, I've got basically envelopes in different pockets. Like my breast pocket would say one. Front pant pocket would say another one, the other front pen pocket would say the other one back pockets would be them. So you've got distribution of five different predictions across your body. But you say, you know, I've chosen the, you know, the AirPods is the thing that I wanted to pick, and I've got that written down in my front breast pocket. So I pick that out and say -- and say, oh, that's interesting. I predicted this before the thing, but then we go through like five minutes later after we've done some other stuff, and I'll say you remember I was holding that prediction, and you said AirPods, and I gave that to you. So I can reframe that memory after a little bit of time delay, and it works really, really well. People give you the memory benefit of the doubt --
Mason Amadeus: Yeah.
Perry Carpenter: -- and when everybody goes out of the theater, that's how they remember. It's like, oh, man, he was holding this envelope the entire time. Had all these objects. Had me just kind of grab one. Whether he was psychic or whether he was influencing me, I don't know, but he knew that I was going to pick the AirPods. And for me, I just had every option covered.
Mason Amadeus: Right.
Perry Carpenter: And I was, you know, going to go into a natural place and grab that out so it felt still realistic for you on the front end.
Mason Amadeus: because it's so easy to gaslight you about the things you don't notice, right?
Perry Carpenter: Yeah.
Mason Amadeus: Like that person probably didn't pay that much attention to the fact you grabbed it from your pocket, right? Particularly if all of those are in normal pockets, you know, it's not like you're pulling it out of your shoe or something.
Perry Carpenter: Yeah, exactly.
Mason Amadeus: It's not memorable. Yeah.
Perry Carpenter: Well, and you can even say with some of the ones that I've done, it's like because I'll usually like have the date or something like that written on it, something that makes it feel like really just part of the actual event. And so after they call out the thing they want, I'll say -- I'll may even go, oh, I actually, I meant to be holding this. I meant to have this out the entire time. Sorry about that. So at that point, I'm still kind of like presuggesting, preconditioning the fact that the way that it normally goes is that I'm holding it.
Mason Amadeus: Yeah, all of that subtle memory modification.
Perry Carpenter: Yeah, yep. So our tools are now starting to do that for us is we're starting to see -- and this is the crux of the MIT study is that you have that original image, you give them a delay, and then you ask them about it. And sorry, you were going to say something?
Mason Amadeus: No, I just -- I laughed because earlier I almost said if maybe -- what if you remove the evidence of a crime in your background.
Perry Carpenter: Yeah.
Mason Amadeus: Like you don't know, but you took the one picture of the guy who did the crime, and that's -- they show that in this study that we're looking at on screen.
Perry Carpenter: Yep.
Mason Amadeus: And so I just laughed because --
Perry Carpenter: Yep. Exactly, exactly.
Mason Amadeus: That's a bit more farfetched.
Perry Carpenter: So, it is, it is. But if you're building disinformation or if you're building cover over something, then -- and you only have like one eyewitness for a couple, and then you're building subtly changing evidence, maybe you can inject enough within their memory that there's reasonable doubt if they're hooked up to like lie detector tests and stuff like that. So they were given these original images, then given a time delay to filter, and then they come back and ask about recall, and they had about 200 participants this control study. And then when they went to the AI-edited images, what they found is a 2.05x more false memories than seeing the original unedited images.
Mason Amadeus: Really?
Perry Carpenter: So they were recalling the things incorrectly more or recalling the change more. And so they went across these like four different stages. One was unedited images. That's their control study. Then they went to AI-edited images, and that's where you got that 2.05 more false memories implanted. And then they went to AI-generated videos of unedited images. So if you generate an image of these people interacting now based on a real image, how does somebody respond? And do they remember those interactions and maybe implant a false belief that way? And then the last one is, you can imagine, is an AI-generated video of an AI-generated image, and that actually had the strongest effect.
Mason Amadeus: Really?
Perry Carpenter: It led people to remember added and removed details, even if those weren't present in the original photo. So you take an original photo. You insert Sam Altman holding somebody hostage. Then you animate that into something, create a situation, and then now, within that person's memory, they can plausibly recall that thing, whatever the thing is.
Mason Amadeus: So they showed them each a set of 24 images, then made them play Pac Man for two minutes. So they had two minutes to look at 24 images, two minutes to play Pac-Man, and then they were shown either the AI-generated videos, AI-edited images, the unchanged images, and AI-generated videos and quizzed on if they remembered them being the same or different.
Perry Carpenter: Yeah.
Mason Amadeus: Interesting. That is trippy.
Perry Carpenter: Yeah, right, I would encourage folks to go look at the study. I, like I said, I need to read it a couple more times and make sure --
Mason Amadeus: Oh.
Perry Carpenter: -- that I don't mischaracterize anything. But it is -- it is something that we need to know about.
Mason Amadeus: And that's a really good -- the one that's on screen right now is what -- the original image is like a wedding scene or something. There's a guy in a tux and a lady in a dress. The guy's like looking down, and the lady's kind of looking into the middle distance. And then the edited image, they're both smiling and looking at the camera.
Perry Carpenter: Yeah.
Mason Amadeus: And so it was like subtle changes like that, that they were making using AI.
Perry Carpenter: Right. Well, and those are the kinds of things that we were doing with Photoshop for a long time, right, is like, oh, I've got a -- and this is one we did with our own family is we were taking family photos 12 years ago, and it was just a -- just Photoshop. And it was increasingly difficult to get everybody to have the right expression at the right time when you have little kids.
Mason Amadeus: Right.
Perry Carpenter: And so I had to take a picture of my son from one photo and then superimpose him where he would have been in the other photo. And I think if you were to ask our kids right now if that was the actual, you know, time that they had, they would probably go, yeah, that seems right.
Mason Amadeus: That's how I remember, yeah.
Perry Carpenter: Me and my wife know that I spent two hours doing that, but at the same time, if I try to recall that memory, that's kind of the picture that sticks in my head even knowing it.
Mason Amadeus: And also the thing is too you spent two hours doing it, and if you just do it in one click with AI without thinking, you're even less --
Perry Carpenter: Yeah.
Mason Amadeus: -- you have less time to catalog that a change was made.
Perry Carpenter: Exactly.
Mason Amadeus: Is a much more passive change, yeah.
Perry Carpenter: Exactly. So I think all this will fit into the last segment as well because we're going to still get into like the disinformational use of AI-generated images and plausibility.
Mason Amadeus: Ooh, fun. And before we get into that --
Perry Carpenter: Yep.
Mason Amadeus: -- we're going to take a little dive into something that's just kind of a kind of a weird curiosity that I want to show you. I think you'll find this very interesting, Perry, and if you're watching the video version of the podcast, you'll get to see this because this is a pretty visual thing.
Perry Carpenter: Nice.
Mason Amadeus: I'll do my best to describe it, so stick around. This is -- this one's gonna be weird. So this was sent in by Discord member and future paper clip, Bullethead. Thank you, Bullethead, for the great submissions in our Discord server. You can join it at fake.2/discord, also a link in the description and the show notes. This is a pretty neat little thing, Perry.
Perry Carpenter: Okay.
Mason Amadeus: It's called MMM-Da Parallel: Parallel Multimodal Large Diffusion Language Models for Thinking, Aware Editing, and Generation. And we're going to take that nasty --
Perry Carpenter: Geez, okay.
Mason Amadeus: -- mouth feel and spit it out, and we'll talk about what it actually is. What you can see on screen is a good demo of what it is. It is generating an image and text in parallel. But unlike in LLM, if you remember several episodes back, we did that thing about diffusion language models.
Perry Carpenter: Yeah.
Mason Amadeus: Do you remember, Perry --
Perry Carpenter: Yep.
Mason Amadeus: -- where instead of sequential tokens, it's like noise tokens, and then the text is getting edited non-sequentially. Like stuff gets inserted in the middle, in the beginning.
Perry Carpenter: Yeah.
Mason Amadeus: And what this is doing is using that text generation through diffusion, well, it's generating an image through diffusion, and it is cross-pollinating those sort of tokens together to improve the image generation process to better understand what the user's intent was.
Perry Carpenter: That is trippy.
Mason Amadeus: It's really cool. Now, this is very nascent. I think large diffusion language models. I might be getting -- you know what I mean -- I might be getting like the order of those words wrong. I think those are still kind of in their baby state because there's not many --
Perry Carpenter: Right.
Mason Amadeus: -- big providers building them. So this is not like a super technically impressive thing, but it is a fascinating one. I'm going to jump to the research paper for a moment because I've highlighted a few things that help break down what's going on. I'm going to start with their introduction. Don't worry, I'm not going to read this entire block of text, but this sets this up very well. Recent advances in multimodal generative models have achieved remarkable progress in instruction-based image generation and editing. We're all familiar with that. You ask ChatGPT or Gemini to make something. Given diverse textual prompts, these models can produce visually coherent and semantically aligned results across a wide range of tasks. However, these models often struggle with complex instructions that require reasoning over world knowledge, frequently leading to incorrect editing and generation. We've all probably experienced that too. You're like make this tree blow in the wind, and it's just the tree again.
Perry Carpenter: Yup.
Mason Amadeus: I tried to get Google Gemini to make me a picture of the Oscar Mayer Wienermobile on fire flying over an overpass, and it just would not -- it could not get it right. To mitigate this problem, a lot of models introduce reasoning steps before the visual generation so it will iterate over your prompt and try and come up with a better prompt behind the scenes to feed to the image generation. But they point out that there is a error propagation that can happen if it doesn't fully understand your prompt, and it's going front to back as it generates the prompt --
Perry Carpenter: Yep.
Mason Amadeus: -- it'll feed to the image thing. Anything it gets wrong just gets propagated, fed to the image thing. So there's no sort of talking back and forth. And they found that on certain benchmarks, the inclusion of reasoning can, in fact, reduce the semantic fidelity of the generated images. They have some examples here, where they show that sequential generation can suffer from vague or incorrect reasoning, whereas parallel generation helps align text and images at each denoising step, which helps reduce hallucinations and errors. Their whole thing here is what if multimodal models could generate text and images in parallel? We prompt a powerful vision language model with data triplets, so input image, edit instruction, output image sourced from widely-adopted image editing and generation datasets. The vision language model is tasked to generate a reasoning trace that explains the edit process. So on one side, it's asked to explain what it's editing and why, and on the other side, it's generating the image, and each token is getting, you know, fed back in real time. So as the image is edited, the reasoning changes. As the reasoning changes, the image editing changes to try and get it closer and closer to what the user's intent was. In a way, that's very cool to watch happen too. And they found that during the parallel denoising process, the image region corresponding to a specific semantic concept is often refined simultaneously with its textual counterpart so that they've gotten this cross-pollination working. This paper is pretty fascinating and pretty well-written. I would definitely encourage you to check it out if you are of the nerdy persuasion. But we're just going to try it. I have a --
Perry Carpenter: Okay.
Mason Amadeus: I played with it a little bit before the show, and I actually ran out of my free Hugging Face GPU credits --
Perry Carpenter: Oh no.
Mason Amadeus: -- so I had to make -- I made a second account -- don't tell Hugging Face -- so that we could do this. I have -- I'm giving it the input image of our podcast show art, so "The FAIK Files" logo.
Perry Carpenter: Okay.
Mason Amadeus: And I'm giving it the instruction. This is the logo for a podcast "The FAIK Files". It is a stylized folder with bold outlines and line work. The background is a blue circuit board with a big fingerprint on the top left. The N2K network logo is in white in the bottom-right corner. We need to keep every aspect of this image the same, except we must change the text to read The FART Farts. Keep everything else as close to the original as possible, except the new text. And so I'll hit Generate, and we'll get to watch in real time as it spins up. It'll take a second here, and then it'll start generating the image on the right and the textual reasoning on the left, and you can see these placeholder noise tokens. So the reasoning starts out --
Perry Carpenter: Oh, it moved to FART Files really fast.
Mason Amadeus: It got The FART Files, and I bet you it's gonna get stuck with that. It did replicate the N2K in the bottom corner, which is cool.
Perry Carpenter: Nice.
Mason Amadeus: You know, it's paying attention to the semantic content of the prompt. So --
Perry Carpenter: At one point it put N2K at the very top of our FAIK Files logo instead of The, it looked like. But it's trying --
Mason Amadeus: Yeah.
Perry Carpenter: -- to stay The right now.
Mason Amadeus: It's choking a little bit on it, but it's coming together pretty good.
Perry Carpenter: Yeah, it is.
Mason Amadeus: It has understood the semantic concepts pretty well. It's really fun to see it work. But so as an example, the reasoning trace says, to transform the question image into the answer image, I need to focus on the following changes. And then it was listing out text change, color scheme, background. And so as it was reasoning through that, it was adjusting those various things in the image, and they were talking back and forth. And you could kind of see it in real time like as the bits of that it was reasoning about were being talked about, those things were being changed on the image.
Perry Carpenter: Yeah.
Mason Amadeus: The result is not very impressive because, again, this is like a very nascent thing. It did say -- it says The FART Files. The top of the A has some nonsense over it. It got the thumbprint. It got the circuit board. It got the N2K logo. Everything that I specifically, semantically specified is right. It didn't copy the visual style all that well.
Perry Carpenter: Yeah, yeah, interesting. I wonder what would happen if we threw that same prompt and image into ChatGPT and asked it -- or Google's Nano Banana.
Mason Amadeus: Yeah, let's do that. Let's go to gemini.google.com, and then we'll go ahead and paste in the prompt. We will -- I'll paste in the image as well, and then let's go ahead and say go, and let's see -- let's see how well it handles that.
Perry Carpenter: Yeah.
Mason Amadeus: Because this is just a total different pipeline, right?
Perry Carpenter: Yeah.
Mason Amadeus: Because when we're talking about the other thing, it was that parallel generation. This is serial generation --
Perry Carpenter: Yeah.
Mason Amadeus: -- auto regressive generation through Google Gemini 3.
Perry Carpenter: Yep, so that -- yeah, that first one is creating an entirely new image. This one is -- there's a chance it's going to just do like inpainting is what it's called where it just goes in and like selectively removes and replaces stuff. Ah, there we go.
Mason Amadeus: And the result -- the result is infinitely better. I will say it got it --
Perry Carpenter: It's perfect.
Mason Amadeus: It got it literally perfect. So it's not like a thing that is going to be replacing the way that we're doing image generation right now.
Perry Carpenter: Yeah.
Mason Amadeus: But as far as like an alternate pipeline that I think maybe if more people are contributing to and like developing upon, we could see other applications for it.
Perry Carpenter: Right. I mean, we've seen these things increase in goodness exponentially over short amounts of time. So I can imagine it's going to be great. Just for the sake of -- I guess for the heck of it, open up ChatGPT and see if you can do the same thing there.
Mason Amadeus: ChatGPT. Okay. I've pasted in the picture and the prompt.
Perry Carpenter: All right. ChatGPT is like infuriatingly slow in how it does its image generation.
Mason Amadeus: It really is.
Perry Carpenter: So we might be here for a minute.
Mason Amadeus: I will say that I was impressed by how fast Nano Banana cranked that last one out too. It really --
Perry Carpenter: That was super fast.
Mason Amadeus: Nano Banana has been somewhat impressive. I did not expect to be impressed by it because, again, Jim and I had been lacking behind for so long, but --
Perry Carpenter: Nano Banana kind of blew everybody out of the water. Like, I mean, to the point where it's now integrated into Photoshop because --
Mason Amadeus: Oh, really?
Perry Carpenter: Yeah.
Mason Amadeus: I didn't know that.
Perry Carpenter: Yeah.
Mason Amadeus: I thought they had -- I thought Adobe had their own models that they were building for that.
Perry Carpenter: They've got their Firefly models, but they're expanding out. So when they realize that somebody does something as good or better than them, they're import, you know, they're just kind of bringing that in.
Mason Amadeus: That makes sense. And I mean, Microsoft same kind of thing, right, with them --
Perry Carpenter: Right.
Mason Amadeus: -- opening up to other models.
Perry Carpenter: Yeah.
Mason Amadeus: We experienced an error when generating images. Sometimes, I think that happens, and then you refresh, and it's fine. Ah, let's try again. Take 2.
Perry Carpenter: Yeah, I think Jim and I struggles less with server load as well.
Mason Amadeus: Yeah, well, ChatGPT has got the Kleenex effect for it now. You know, right now everyone calls AI ChatGPT in like casual conversation.
Perry Carpenter: We would call that the Xerox effect back in my day.
Mason Amadeus: Xerox effect?
Perry Carpenter: Yeah.
Mason Amadeus: I haven't heard Xerox in a long time, you know?
Perry Carpenter: Right? Well, that would be like the, you know, make a copy of something. I guess it could just be the Coke effect too.
Mason Amadeus: Gonna say the Hellman's effect, but nobody calls mayo Hellman's.
Perry Carpenter: Right.
Mason Amadeus: Duke's mayo is better anyway. If you have -- if you live in a place that has Duke's mayonnaise, oh my God, it's the best. Hey, there it is.
Perry Carpenter: Hey.
Mason Amadeus: That's so -- okay. So why don't we go ahead? I will copy this. We'll go back to Gemini.
Perry Carpenter: Okay.
Mason Amadeus: Open image in new tab, and then I'll open the other image in the new tab. So this is the difference between Gemini and ChatGPT's attempt to change our logo into The FART farts.
Perry Carpenter: With The FART fart one.
Mason Amadeus: This one with the correct lightness in the AR, like there is in the AI of FAIK is from Gemini.
Perry Carpenter: Okay.
Mason Amadeus: And the one that doesn't have the contrast exactly as I envisioned it is ChatGPT. Because --
Perry Carpenter: Right. Okay, the shapes are more pleasing in the ChatGPT one. The lighting is more pleasing in the Gemini one.
Mason Amadeus: Yeah. It did definitely do inpainting, though. Look at how none of the background changes whatsoever.
Perry Carpenter: Yeah.
Mason Amadeus: I'm pretty sure there's some quality difference, I think, resolution difference, perhaps, but it looks like it's very inpainted.
Perry Carpenter: Right.
Mason Amadeus: And for the record, this logo for the show is not AI-generated. I drew this by hand painstakingly.
Perry Carpenter: Yeah.
Mason Amadeus: Not The FART farts, but the original one, so I'm very familiar with every inch of this thing. And it got most of it pretty darn right, which is pretty cool.
Perry Carpenter: Nice. Okay.
Mason Amadeus: So anyway, and then again, to compare that to the subject of this segment, which is -- which is this is the one from that parallel generation that we -- that we did, so --
Perry Carpenter: I mean, it's cute. It looks like a, you know, almost a Star Wars -- like a LEGO Star Wars type of thing.
Mason Amadeus: I will say this is kind of comparable to the results you get when you use like a small local model to try and do image generation and editing. And so I think if that's where this is now, maybe just a few more steps of refining and scaling and figuring out how to make it work --
Perry Carpenter: Yeah.
Mason Amadeus: -- maybe this will be better at processing semantic content because their paper certainly seemed to point to noticing some increases, at least over this model that's called BagelGPT and GPT-4o also, so, yeah.
Perry Carpenter: Okay. Yeah, I think, give it a year.
Mason Amadeus: Yeah, give it a year. But these diffusion language models are fascinating, and now seeing them work in parallel like this I think is very cool.
Perry Carpenter: Yeah.
Mason Amadeus: That's all I got for that. Now it's back into the world of darkness, right, Perry? Don't you have something spooky for our last segment?
Perry Carpenter: Yeah. I don't know if it's spooky. It's just more news, disinformation stuff. And, yeah, I mean, that's where we are with life, so --
Mason Amadeus: That's pretty spooky. We'll be right back.
Perry Carpenter: Okay, now returning back to false memories, false narratives --
Mason Amadeus: Yes.
Perry Carpenter: -- and all things like that. So this does tie in with Segment 2, where we talked about false memories and recall as one of the big results from AI-generated videos, of AI-generated images. And so I think it makes sense if we go over to something like this BBC article: Russian Bloggers Push AI Videos of Fleeing Ukrainian Troops.
Mason Amadeus: Oh, wow.
Perry Carpenter: You know, again, every good deepfake or every good deception is based in some kind of plausibility and story, and that has to be something that, you know, has a good story, is cognitively smooth in the way that it's plausible enough to be real. And so, you know, people fleeing from a war zone would make sense. People being fed up with what's going on would make sense. And so here there was, you know, Russian Telegram bloggers on November 11 shared two AI-generated videos purporting to show Ukrainian soldiers, quote, "fleeing," and they had claims that the East Ukrainian town that they were fleeing from was encircled by Russian forces. So one blogger the following day removed the video that he had shared, and another said that it was Ukrainians who had created the videos in order to drown out the Russian army's, quote, "real successes."
Mason Amadeus: Oh my gosh.
Perry Carpenter: Yeah. And then a third ridiculed the video, saying that they were clearly AI-generated. So BBC decided to step in and try to see what was going on. They found a ton of visual anomalies in the videos, including unilateral soldier movements, distorted objects, inplausible details such as a self-moving wheelchair and a soldier carrying -- soldier carrying medieval-style axes including other AI --
Mason Amadeus: Really?
Perry Carpenter: You know, though, in war times when you talk about like some of the poverty and some of the desperation, I don't know how accessible medieval-style axes are, but at the same time, you're kind of like grabbing everything that's available for some of these. So when we talk about when somebody goes, well, clearly there's visual distortion, I always dismiss that a little bit -- Mason Amadeus: Yeah. -- because of our, you know, compression artifacts and other things that we see in legit videos all the time.
Mason Amadeus: A lot of videos where it's just a drone tracking something like that. You know, you know, a drone tracking shot --
Perry Carpenter: Yeah.
Mason Amadeus: -- where it's like encircling an object? A lot of videos that are real --
Perry Carpenter: Yeah.
Mason Amadeus: -- of that, people in like comments would be like this is AI just because that perspective is very common in AI things.
Perry Carpenter: So, yeah.
Mason Amadeus: So, yeah.
Perry Carpenter: Yep.
Mason Amadeus: You got to be careful with that.
Perry Carpenter: Yeah. Yeah, you got to be really careful in like overcalling out stuff as, quote, unquote, "obviously AI." So on November 11, it says that this was -- video was posted to a Telegram channel. I did try to click on that. It went to something that wasn't fully yet because I don't have Telegram on this machine, and I didn't really feel like getting there. But let me just check this real quick, too.
Mason Amadeus: Oh yeah, that link, maybe that link will get us there.
Perry Carpenter: Okay, here we go.
Mason Amadeus: Awesome, okay, without having to join any sketchy channels.
Perry Carpenter: All right.
Mason Amadeus: Cool.
Perry Carpenter: Yes. So we have the video. Let me go ahead and click it. We can give our own commentary on what we think.
Mason Amadeus: Okay, so we have some people walking. There's --
Perry Carpenter: Yeah, I say the physics of the men carrying the gurney looks a little bit off.
Mason Amadeus: Yeah, the -- and the guy --
Perry Carpenter: Like it feels too light.
Mason Amadeus: As someone who recently dislocated their kneecap, that guy's not using crutches right, and he looks like he's supposed to have an injury, but he's just walking with the crutches wrong.
Perry Carpenter: Yeah, and the wheelchair --
Mason Amadeus: You know? The wheelchair's moving on its own.
Perry Carpenter: There's a guy in a wheelchair. Yeah, he's got his hands on the -- on the wheels of the wheelchair. It's moving on its own without the wheels actually revolving or him doing any pushing motions.
Mason Amadeus: And also the mechanics of the wheelchair, it's missing some pieces from the base that connect it --
Perry Carpenter: Yeah.
Mason Amadeus: -- and things like that I can see.
Perry Carpenter: So I agree. I would look at it and say this seems clearly AI-generated.
Mason Amadeus: Yeah, the movements are weird. The facial expressions are strange. Like some of these dudes are like smiling and stuff.
Perry Carpenter: Yeah.
Mason Amadeus: Although that, you could, you know, you could think like people --
Perry Carpenter: Yeah, put a guy in the middle. Like his smile looks unnatural for the situation.
Mason Amadeus: Yeah.
Perry Carpenter: But maybe at the same time, you're seeing somebody with a camera, and you're being performative. So like if everything else was spot on, and one guy was making a weird expression, I would find a reason to write that off.
Mason Amadeus: It's the sum of these things. And also --
Perry Carpenter: Yeah.
Mason Amadeus: -- I am astounded by the attempts, very often, to reverse victim and offender, where like they got called out for posting this --
Perry Carpenter: Yep.
Mason Amadeus: -- and then they're like, actually, it was the Ukrainians to make us look --
Perry Carpenter: Right.
Mason Amadeus: Like it's so -- gosh.
Perry Carpenter: Exactly. But, I mean, what they're trying to do is just get fog of war, right?
Mason Amadeus: Yeah.
Perry Carpenter: So confusion is everything in these. Now, the thing that I didn't like is some of the commentary and the way that the experts approach this, you know, as experts in these fields try to bring clarity, one of the things that we have to do is we have to realize that a lot of people are just -- they're guided by their amygdala or their, you know, their reflexive thought and emotion as they look at these kinds of stories. And humans are just going to be human. Humans are -- most humans are not innately stupid in the way that they engage with stuff. They're innately emotional. And so one of the things that I didn't like about this article is this quote from a war correspondent, and he said he -- ridiculed the video, saying that they were clearly AI-generated. He said part of his job as a war correspondent is to explain to -- here's quote -- the idiots that take AI-generated content at face value that, quote, "they are a flock of dumb sheep, and sheeps need shepherds."
Mason Amadeus: Yikes. Yeah, that's a terrible attitude, isn't it?
Perry Carpenter: Yeah. Yeah, so I don't like that. I don't think that this is the BBC or correspondent. It doesn't sound like the kind of verbiage that the BBC would use for these kinds of things, but the fact that there are people that are calling themselves journalists --
Mason Amadeus: Yeah.
Perry Carpenter: -- that are calling their readers idiots and are saying that their readers are a flock of dumb sheep and that their job is to guide them and be the shepherd, I think that's the wrong way to do this. Yeah.
Mason Amadeus: Yeah, that person is burned out, and like the moment you start regarding the people for whom you're trying to provide a service as idiots or less than you, you need to like step back and reconsider your role. That's something I encountered when I was in IT management with people who struggled to deal with users, you know, I mean, you know --
Perry Carpenter: Right.
Mason Amadeus: -- and the classic stereotype of the IT guy telling everyone they're stupid and whatnot. Like you gotta --
Perry Carpenter: Yep.
Mason Amadeus: -- you don't -- you don't do that. You don't be like that.
Perry Carpenter: Yeah. So you mentioned kind of like the DARVO effect on this. So DARVO is like where you dismiss -- what is it? -- dismiss -- Let me look that up.
Mason Amadeus: Deny, Attack, Reverse, Victim and Offender.
Perry Carpenter: That's, yeah, that's it. So that's actually what's going on with this, right? So when they had the videos, they get debunked, and then the Russians will say that, you know, it was actually the Ukrainians that created these videos, to quote, unquote, "drown out on social media the Russian army's," quote, "real successes in Ukraine."
Mason Amadeus: That's a real stretch of a DARVO too. That's like a real stretch of one.
Perry Carpenter: Yeah. That you have two videos to drown out the Russian army's real successes.
Mason Amadeus: And also, how would it even do that? There should be a DARVO Awards --
Perry Carpenter: Right.
Mason Amadeus: -- like there's a Darwin Awards.
Perry Carpenter: Right? Exactly. And then it says the goal is to drown out genuine footage in a sea of obviously fake content that is deliberately spread by the opponent in order to be easily refuted later. This does --
Mason Amadeus: Yeah, I saw two posts and I was really in a sea the other day, for sure.
Perry Carpenter: You know, this does get to the whole liar's dividend thing, though, right, is that what people are really going for is just the fact that truth is very ambiguous online right now. It's really, really difficult to, for sure, say what's real and what's not, unless you have an obviously fake video. This one kind of moves into the obviously fake video, but I can step into the mind space of somebody that's been living in a war torn country for several years --
Mason Amadeus: Yeah.
Perry Carpenter: -- and whatever they're viewing this on.
Mason Amadeus: Yeah. And I will say like to be a war correspondent is to deal with some of like the darkest things in humanity too. So like --
Perry Carpenter: Yeah. So let me kind of end this with one other -- and so we move from Russia, Ukraine to Israel -- to Israel-Gaza. And this is from the Oversight Board, which I don't know a lot about these guys, how reputable they are, but they purport to be a check on Meta, so the company that owns Facebook and so on.
Mason Amadeus: Okay.
Perry Carpenter: And what they're looking to do in this, specifically, is to address AI-generated content in the Israel-Iran conflict. So sorry, I said Israel-Gaza before. This is Israel-Iran. That originally came from this post that I saw with -- from Andrew Smith -- and we'll put this on the show notes as well. He's talking about the Oversight Board announcing a case where Meta left up AI-generated videos faking extensive damage to several buildings in Israel. So that's just stoke anger, right --
Mason Amadeus: Yeah.
Perry Carpenter: -- and continue aggression while a real armed conflict was raging. The content was shared by a page claiming to be a news organization, got over 700,000 views. So again, every deception fits into narrative, and if it seems plausible and a combination of story and the thing that's being shown, the deceptive artifact, then it can get the emotional outrage or whatever the thing that you're wanting to have happen. It can actually lead to that happening. And so initially, Meta did -- said that this content didn't violate any parts of its misinformation policy, and so they were --
Mason Amadeus: If I had a nickel --
Perry Carpenter: -- fine to stand.
Mason Amadeus: If I had a nickel for every questionable decision, like every obviously questionable, terrible decision Meta has made loudly in public --
Perry Carpenter: Right.
Mason Amadeus: -- I would be drowning in nickels. It's just unbelievable.
Perry Carpenter: Right. Yeah, so then the Oversight Board brought this post specifically to their attention, so a lot of people were reporting it. And then the Oversight Board gets involved and brings in, and then better remove the page and the accounts behind it because of engagement abuse and inauthenticity.
Mason Amadeus: Okay.
Perry Carpenter: So it wasn't a misinformation policy that they leaned into. They're like, no, this is -- this is fine, but we'll, you know, pull it down for engagement abuse.
Mason Amadeus: What does that even mean?
Perry Carpenter: Yeah. So what the Oversight Board goes on to say is this points to the challenges in engagement-based social media companies when their users exploit a real crisis for clickbait, which is -- I guess that's engagement abuse, right, is they're trying to drive the click.
Mason Amadeus: But that's --
Perry Carpenter: And so GenAI has made that much easier.
Mason Amadeus: That screams to me that they were just like we got caught with our pants down. We can't turn around and say --
Perry Carpenter: Right.
Mason Amadeus: -- actually, it is a misinformation because Facebook is engagement abuse. Every post on the internet is engagement abuse.
Perry Carpenter: Yeah.
Mason Amadeus: Gosh.
Perry Carpenter: So then here is -- I'm going to move. This is factcheck.afp.com that is hosting this one, and this has some examples of the videos, or, sorry, the images.
Mason Amadeus: Yeah.
Perry Carpenter: And so you'll see just utter destruction.
Mason Amadeus: Yep, AI-generated smashed landmarks and cities and gyms.
Perry Carpenter: Yep, yep.
Mason Amadeus: Yeah.
Perry Carpenter: Things that would be recognizable. And then you'll see like the AI-generated version, which is just all dilapidated and crushed versus the like satellite imagery of what would have been intact at the time. So that -- like that would feel really visceral --
Mason Amadeus: Yeah.
Perry Carpenter: -- if you're somebody familiar with the area, and you weren't in a position to know better.
Mason Amadeus: Yeah, I mean, if you're seeing images of your hometown destroyed, you're seeing images of the city you live in destroyed --
Perry Carpenter: Yeah.
Mason Amadeus: -- or the city -- probably not the one you live in because then you would know it's not true --
Perry Carpenter: Yep, you would know.
Mason Amadeus: -- but, you know, a city you're familiar with and whatnot.
Perry Carpenter: Yep.
Mason Amadeus: Yeah.
Perry Carpenter: I mean, this would be the equivalent of like what we would have here in the US, where somebody is saying that, hey, this city is a war zone. And then you have people that are living there that are like, wait, I'm walking down those same streets you're talking about, and I'm not seeing any of this stuff.
Mason Amadeus: It's reminiscent of the Portland stuff, yeah, and there was AI disinformation around that.
Perry Carpenter: Yeah.
Mason Amadeus: And then also the flooding that happened.
Perry Carpenter: Yep.
Mason Amadeus: Although, in that case it was --
Perry Carpenter: You know, we see this all the time.
Mason Amadeus: Yeah.
Perry Carpenter: Right? It's exaggerated in some cases, right? So there may be -- like in Portland and in LA, there was like a block or two where there was some protest, and sometimes those protests were very packed, but then people would make it -- or they'd take bits of that out of context, and they would say that that's the entirety of what's going on when, in fact, it was very like self-contained intentionally to certain areas for very real reasons.
Mason Amadeus: Yeah.
Perry Carpenter: But then -- but then also, we see people like take an image from -- and we saw this with Portland, as well -- an image from riots in conflict in 2020 that were more extreme than they were this round and then attribute that to something in 2025.
Mason Amadeus: I was going to say.
Perry Carpenter: And so it can feel real at that point. That's not a deepfake.
Mason Amadeus: There is also -- I forget what incident it was, but there was one where someone used video game footage from --
Perry Carpenter: Yeah.
Mason Amadeus: -- like a modern video game and just crunched the resolution a bit and said this was real too.
Perry Carpenter: Yup.
Mason Amadeus: So like it's just how easy it is now, you know?
Perry Carpenter: Yeah, it's too easy, and it's too easy for it to be just plausible enough for somebody that's not like there on the ground to dismiss it. And again, if it fits a preexisting narrative or preexisting bias that you're likely to believe in, then you're not necessarily in a state to think as critically about it. And so that's not like a -- that's not really a condemnation of the person that falls for it. It's just human reality.
Mason Amadeus: I mean, I think a lot about the -- not Maxim. I think a lot about the razor that you outlined in the book FAIK of like if something sparks a big response in you emotionally, that's a cue to slow down and look for other --
Perry Carpenter: Yep.
Mason Amadeus: -- collaborating or corroborating sources.
Perry Carpenter: Yep.
Mason Amadeus: So like that's really the only razor I can think of to help as a consumer of news and a scroller of the internet is like when you feel that reaction, stop and step back.
Perry Carpenter: Yeah, yeah. And that's -- so one of the things that I have put in a lot of my presentations is this thing that I call the FAIK Framework, the F-A-I-K Framework. So freeze and feel. So as soon as you get that emotional reaction, actually slow down and then try to name the thing that you're feeling. Like is that fear? Is it outrage? Is it pressure?
Mason Amadeus: Disgust?
Perry Carpenter: Is it urgency? Is it disgust? Is it hope? You know, what is that? And then the A is, Analyze the narrative. And then like also look at the emotion again. Say what emotion or what emotional triggers may have been engineered into this thing intentionally? And then investigate the claim sources, and then you're able to know, confirm, and keep vigilant. So those types of ways of thinking about it, you have to say is this falling into a story, and is this trying to invoke some kind of emotional response intentionally? And then you always have to ask the question why does this thing exist? Why is it in my feed? You know, how did it land in front of me? What story is it snapping into? What emotions is it trying to engage? And then, ultimately, what does it want me to do or believe? And once you start to ask those questions critically, you're well on your way to being able to disambiguate reality from falsehood.
Mason Amadeus: I think the trickiest part or the hardest part to put into action is to recognize when something is falling into your own biases that you have. Because it's like --
Perry Carpenter: Oh, yeah.
Mason Amadeus: -- it's really hard to challenge your own worldview even, you know --
Perry Carpenter: Yep.
Mason Amadeus: -- in something you have all these preconceived notions about. It is a challenging thing to do, but it is very effective.
Perry Carpenter: Yeah. Oh, and the thing that I say every now and then too is that if it feels really, really easy to believe, it may have been engineered that way.
Mason Amadeus: Yeah, that's a good way to put that.
Perry Carpenter: You know, one of the -- one of the things that we do like with furniture and stuff is like you smooth it out, you know, certain bits of furniture. You put on a lathe, and you get -- you get rid of every bit of friction so that it feels comfortable. And if something that you're seeing online that is -- that falls within a social or political narrative, if it feels really comfortable to believe, then you need to take a step back and look and say, has this like actually been smoothed out artificially --
Mason Amadeus: Yeah.
Perry Carpenter: -- so that it's easy for me to believe?
Mason Amadeus: Yeah, if it's sanded down to be digestible and to fit right into your particular worldview.
Perry Carpenter: Yup.
Mason Amadeus: Yeah. Yikes. Well, I guess all that's left to do in this episode is to check up on our website progress in a little bonus segment and then close things out.
Perry Carpenter: Pull that thing out of the oven.
Mason Amadeus: Let's go ahead and do that. Here's a transition. So at the start of our first segment, I set up Antigravity in Turbo Mode, Google's new coding tool, and told it to just make us a website without needing my input or anything. And unfortunately, it got as far as installing a bunch of node packages. It installed Eleventy, and it started spinning us up some like -- it made an index page, which honestly is kind of garbage. It doesn't -- none of it's dynamic. It has like hard-coded episode cards here for like Episode 1 of FAIK and things like that.
Perry Carpenter: Oh, okay.
Mason Amadeus: So it was off to a pretty rough start, and then it crashed with an error saying that there was -- let me see if I can get -- it's really small over in the corner, but it says agent execution terminated due to model provider overload. Please try again later. I think either that's a rate limiting, a lot of people are using it all at once, or I ran out of free usage.
Perry Carpenter: Yeah. Yeah, you may -- you may run out of credits with that. We'll have to like do that in a structured segment maybe next week --
Mason Amadeus: Yeah.
Perry Carpenter: -- and see what we can do.
Mason Amadeus: Yeah, or maybe we even make like a separate -- like a video video about it, where we can do a longer-form thing and cut it down or something.
Perry Carpenter: Yeah.
Mason Amadeus: But so that it's not wholly dissatisfying, I'll share the website that I have been building with Gemini that --
Perry Carpenter: Ooh.
Mason Amadeus: -- it was just as an example --
Perry Carpenter: Okay.
Mason Amadeus: -- of what you -- what you can do when you're working in tandem with it. This is a website I've been building to kind of link together and host all of the various things that I'm involved with. And also I was learning this so I could learn Eleventy because I want to make us a better website for "The FAIK Files" so we could host all of our --
Perry Carpenter: Cool.
Mason Amadeus: -- like images and videos and stuff. So this was the bare bones idea.
Perry Carpenter: Nice.
Mason Amadeus: It's a very simple-looking website, but it's deceptively simple-looking. I made a directory listing thing where I manage this website from the command prompt. I built this little application --
Perry Carpenter: Oh.
Mason Amadeus: -- basically that I can launch. It checks everything, handles all these get operations. I can deploy the dev server, yada yada. I manage it all from here. And all I do is add content into this directory, and it automatically gets put into layouts, gets put into folders. So it is like a frictionless-to-update website. You can just drop stuff in.
Perry Carpenter: That's really cool.
Mason Amadeus: Yeah, there's a lot of quality-of-life features. So if I go to like -- here's a -- I made Temptations Brand Cat Cigarettes once. This was like a demo post I put together. It has, you know, support for all sorts of different embeds, and then there's share links that send permanent IDs, and there's like a lot of quality-of-life features built in. And there's a lot under the hood that makes this work. And I won't go on too long because the episode's already too long, but if you go to bodgelab.com, you can poke around it. There's really not much up there yet. But I built this framework in a couple of days with the help of Gemini 2.5.
Perry Carpenter: That's really cool.
Mason Amadeus: And then Gemini, yeah, and then Gemini 3 took it over the edge when I was finally tweaking all of this stuff in the command center prompt thing and the updating workflow that I built. It works amazingly. And now this is not make me a website, here's a single prompt.
Perry Carpenter: Right.
Mason Amadeus: This is me iterating back and forth, like describing design things, going in and editing the code myself. I'm fairly proficient in JavaScript and HTML, but I did lean very heavily on Gemini because I also wanted to see what it could do. And when it switched from 2.5 to 3, just the last -- the stuff I did yesterday at evening --
Perry Carpenter: Yep --
Mason Amadeus: -- it was amazing. The leaps and bounds it made were so rapid.
Perry Carpenter: Interesting.
Mason Amadeus: Yeah.
Perry Carpenter: I'll be interested to see like if some of these other agentic coding platforms, like Lovable move from Anthropic to Gemini at some point because they're always going to be looking for the most efficient models. And to date, they've been most comfortable with Anthropic and you can kind of tell by the way the agent chats back to you that it's clawed in a lot of ways because it uses a lot of the catch phrases that are indicative of it. And I think they're very transparent about which models they're using most the time, too. So I'm not trying to imply that, but I do wonder if once the benchmarks are verified and once there's enough actual people trying these and verifying it if some of these platforms make the cut.
Mason Amadeus: I think they probably will because like from just a non-empirical, just a -- what is the word?
Perry Carpenter: For subjective?
Mason Amadeus: Yes, from a subjective perspective, it was like a leap because the --
Perry Carpenter: Yeah.
Mason Amadeus: -- code base for this site has gotten pretty elaborate behind the scenes. And so I was -- I was doing a lot more than like the AI in terms of managing the whole project. It was giving it bits to like help me do this like, or what's the right approach to this?
Perry Carpenter: Yeah.
Mason Amadeus: What's a good pattern for this? And then when 3.5 dropped, I was like, oh, let me just reimport the repo and tell it to do something and, wow.
Perry Carpenter: Nice.
Mason Amadeus: Subjectively, it was amazing. It caught bugs that I hadn't noticed that had been there for a long time.
Perry Carpenter: Ooh.
Mason Amadeus: It made great suggestions in streamlining, so I think we will see that. I've been very impressed with it.
Perry Carpenter: Let me -- let me throw one more thing on the screen. We're not going to go into it, but I'm going to pull this up. So not to totally dismiss Claude from the coding game because, yeah, earlier this month, Anthropic did disrupt the first reported AI-orchestrated cyber espionage campaign that came out of China, and all that was being run on Claude code --
Mason Amadeus: What? Oh.
Perry Carpenter: -- agentically doing everything. We'll have to touch -- this is longer than we have time to get into, but this was significant. And like everybody that's in the intelligence world and the cybersecurity world agrees that fully, you know, agentic attacks are right now way further ahead and way more destructive than anybody thought they would be at this time.
Mason Amadeus: Woof. I mean --
Perry Carpenter: Yep.
Mason Amadeus: -- if you don't need precision, who cares about hallucinations if you're trying to break stuff, right?
Perry Carpenter: Yeah, well, and in this you still want to do your attacks the right way.
Mason Amadeus: Right.
Perry Carpenter: And so there's enough that they got right to meet their objectives. Now, Anthropic did disrupt the one that they're talking about, but they were uber surprised by the technical sophistication of what was going on there. So we'll take a look at that one next week probably.
Mason Amadeus: Yeah, that sounds really, really interesting. So I guess a teaser for next week's episode. I think that concludes our little bonus segment here.
Perry Carpenter: I think we're done.
Mason Amadeus: Yeah.
Perry Carpenter: Yeah, I think we're done.
Mason Amadeus: Thanks for joining us this week. Make sure you check out all the links and show notes. We put a lot of resources in there so you can go deeper on the stories than we have. If you want to send us stories to check out, if you want to tell us what we got wrong or whatever, join our Discord: fake.to/discord or click the link in the show notes or the description if you're watching on YouTube. If you're watching on YouTube, subscribe, hit the Like button and leave a comment. We want to build our YouTube audience up bigger. If you're listening on the podcast side, you've probably already subscribed, but make sure, for sure, that you are. Perry, you got anything coming up that you want to put a little plug in for?
Perry Carpenter: No.
Mason Amadeus: No? Nothing. Gotcha.
Perry Carpenter: So I, no, I was on another podcast this week called "Control Room" from the folks at Norton LifeLock. Maybe we can find a link to that and throw that in as well.
Mason Amadeus: Oh yeah. I've got the link for it because it got served up to me in the wild, in my -- in my algorithm --
Perry Carpenter: Nice.
Mason Amadeus: -- which is cool.
Perry Carpenter: Nice.
Mason Amadeus: So yeah. We'll share that link too. Make sure you check that out, and I guess, until next week, ignore all previous instructions and have yourself a great weekend, future paper clips. We'll see you next time. [ Music ]
