← Family IT Guy Podcast
Family IT Guy Podcast
AI Videos vs Deepfakes: How to Tell What’s Real Online (With Jeremy Carrasco)
01:17:23
Loading summary
Transcript91 lines
- [00:02]
A
Jeremy Carrasco, welcome to the Family IT Guy podcast.
- [00:06]
B
Very glad to be here. Thank you so much.
- [00:09]
A
I am so curious to have this conversation today. I've seen you on social media on Instagram and TikTok where you post these really fascinating and well informed videos analyzing what's going on with AI. Like AI generated photos, AI generated videos. You do a really great job at helping people identify. How do you tell? Here's a video that's from the latest AI model that is hyper realistic. What are the tells and what do you look for and the. So I want to ask you a bunch of questions about that today.
- [00:45]
B
Sure.
- [00:45]
A
Before I do that, would you mind introducing what you do and a little bit about your background?
- [00:55]
B
Yeah. So I was a media producer full time before doing this. For the past eight or ten years my main focus has been video production, audio production, leading up to the pandemic. Then during the pandemic I was a live streaming specialist. So I was a multi camera live stream director. Not so much like video game Twitch streaming or Roblox or something, but more like, you know, TikTok does a world series coverage sort of deal. And so that's a lot of what I was doing the past past five years. And that gave me a really in depth knowledge because I was working at a production company, we were for hire. So some days I'd work on corporate things, some days I'd work on very casual types of projects or fashion shows or like sporting events. So I got a very in depth look at so many different parts of the like the social media industry, especially because live streaming mostly goes to social media platforms rather than broadcast tv. So developed a very in depth knowledge of that. And when you're doing live streaming you have to be super aware of everything from the cameras to the audio quality to how it looks to the viewer at home. So after seeing everything that could possibly go wrong with real video, once AI started coming up, it was all new things that were going wrong, all these new little errors that were happening and it just kind of stuck out like a sore thumb to me. So by deduction I was able to figure out how to spot AI videos. And specific to this podcast, I've actually never talked about this publicly is I was also a piano and guitar teacher as a side hustle for three or three years or so. So I was at one point actually when work was down, I was teaching piano to something like 50 kids a week or something like that mostly.
- [02:43]
A
Wow.
- [02:44]
B
Ages 4 to 9 on average. I was never a great pianist, but I was a good teacher and a good person to keep people focused. I went to music school, so I knew all the structure and all that. So, you know, and I'm. Even though I don't have kids of my own, you know, I'm a proud uncle, and I like to think that I'm gonna be my. My niece's favorite, favorite uncle. So that's, you know, very, very passionate about, like, making sure that the Internet's a safe place for kids on a. Both a macro level and also like an individual. An individual parenting level. Even though that hasn't. I haven't personally had that experience yet. I'm looking forward to it, but not yet.
- [03:26]
A
Yeah, well, I mean, the uncle thinks the best. I'm an uncle as well, and it's great. And it really. You know, prior to my siblings having kids and me being, like, directly involved with babies and toddlers and all the things, it's definitely given me a whole new perspective. And then I have a kid of my own now and, you know, again. But there's. It's all the same, man. Like, the deep love for a youngster, for a little one that you just want nothing more than just to. If you could give them everything you had, you would. To help them kind of deal. And so it's beautiful. And uncle is the best. It's like the most fun mode because dad mode is, like, it's all stressful, and then you have to deal with all the ups and downs, and then uncle mode is more fun.
- [04:16]
B
It is. It's been fun. Yeah. My sisters let me babysit by myself a few times, which has been a cool adventure. So I like to think I'm going above and beyond normal uncle status. But, yeah, definitely doesn't stack up to being a parent. I'll cross that bridge one day.
- [04:36]
A
No, that's beautiful. Well, I think your background with production and from what I read about you online, like, you have a. You're steeped in a technical background with video production and as you described as, well, which I think it just makes so much sense from the. Now that I. It gives me so much more context to understand your work with the show Tools AI Channel that you have and really being able to spot, like, okay, a real video wouldn't have that little thing. And so just to start off on some stuff that I've. I've looked at, like, some of your themes and things. Like one of the things you've said, um, so you've said, AI doesn't breathe. The words just keep coming. And that caught my Attention, because if I'm watching an AI video, my eyes are focused on trying to capture the weird things in the AI video. But, but what that tells me is actually I could even close my eyes and just listen to determine if it's made by AI. So would you mind kind of explaining the, your understanding of that and like how that works?
- [05:46]
B
Yeah, AI video only adapted to having audio about not even a year ago. So the way that audio and video interact is very complex in our minds. There's a physical interaction in the real world between something making sound and us hearing it. And it's a very physical process. And in fact, it is in some ways more difficult for. So let me actually step back and just give a picture about how these systems are making videos. I mean, what they're doing is they're trained on a huge, huge corpus of data. They're turned into individual pixel patches for each video. These are then made noisy or they basically become staticky. And the AI's job is basically to learn how to train change a real video from a video to pure static. And it understands all these steps. It's like each step I have to understand what static was added and then how to reverse that later. And it's done with a bunch of other text input, a bunch of other context kind of imbued into that process. So that later when you say make a video of a dog running, it looks at its training set, it doesn't, it's not looking at the original training set, but it looks at all the things that it has learned and recodes that video through pure noise, through pure static. And it's just going step by step, reducing noise, reducing noise until you get to something that resembles a dog running. And very, very complex process that I just ran through in 30 seconds. Right, but the thing that is, the thing that to understand here is that in order to get something meaningful on the output, you have to do a couple things on the input. You have to label it. Well, so in this case, you might have to say, like the dog is running and barking at the same time. For the AI to understand that a bark also needs to happen during the run. Now when it gets all this data in at once, like it might kind of figure out and it's not a human. But let's just, let's, let's just, for now, let's just anthropomorphize it a little bit. It kind of learns like, oh, when a dog runs, it barks. But those text annotations can only be so detailed. So in this case, if you have a person on a video speaking, there's only so much human like detail as of now that's being encoded into it to the point where the AI would intuit when a human would breathe or when a human might make certain mouth sounds and not others. So it might understand that a T sounds like a thing that sounds like a T to it, but it doesn't understand that a T is literally like formed by your tongue pushing against the tip of your teeth and expelling air. Right. It doesn't understand that intuitively. So there are all these things that are tied to us as humans that it is intuiting, but it's not replicating real life physics in any sort of way. And when you are just making pixels happen. So far, the AI video generators have been able to do a good job with pixels and making things look believable. Maybe the physics aren't exactly right, but that's kind of tied to. What I'm saying is the physics aren't right because there aren't any actual physics there. It is combining pixels and learning patterns. And just like a large language model like ChatGPT, what comes out might look like really reasonable text. But the large language models don't have an extremely intuitive understanding of human language in the way that humans do, which is why you can still tell if something is written by ChatGPT. So what has happened is, and we actually have indications from the AI companies themselves, and again, I am oversimplifying this, but they have said that audio matching with video is kind of the hardest thing for them to figure out right now, which is why a lot of the times it's like, why is the person speaking too quickly? Or why isn't their tone matching the scenario? Again, it's because these are extremely nuanced social interactions that as humans we can pick up. The AI has a hard time decoding that in the same way we do or like recode that in a believable way. So even though it requires a little bit more thought, I think it's a very dirty. It's what I would call a durable indicator, something that's likely to stick around for a long time, little while, is being able to really hear and sense, like, is there something off with the way that they're talking with this situation? But I will also say some people hear it and some people don't hear it. So some people, that is not something that they can intuit. But if you hear it, that's good because it's probably going to Stick around a little while.
- [10:42]
A
Yeah, that's very interesting. And when you talk about the static thing, are you talking about what we would normally see on, like if you had a cable box back in the 90s and you're trying to watch HBO and it was all snowy. So it starts from that and then works its way into an image of a frame.
- [11:01]
B
Yeah. You know, if you think about every pixel here as, you know, some combinations of numbers, what it's really doing is it takes the video in and just starts randomizing numbers a little bit like step by step. And it's really just figuring out like, how do I reverse that randomization at each step when it's learning so that when it's generating it use what it's learned to say, oh, okay, dog running. I remember the steps it took to turn the dog into noise. Now I have to turn the noise into a dog and reverse the process. So it's a, it's a very mathematical, it's, it's an entirely mathematical process and very, very complicated. And yeah, if anyone wants to learn more about that, that's called diffusion. So diffusion is the process by which you turn real things into noise and noise into real things on real looking images. Diffusion's a much more wide physics mathematics concept than just that. But diffusion image generation is generally the principle here.
- [12:08]
A
That makes sense because I've seen diffusion models and I now realize I didn't know what diffusion meant. That's a great explanation. And so beyond this audio mismatch, like, what else have you found? Like what are some of the giveaways when you're, when you're consuming. Well, shoot. I was going to say when you're consuming AI content, but I guess the bigger picture is that trying to determine like what are you consuming? Are there, have you seen, I imagine it's shifting over time. But do you have like particulars that still that are durable in a sense?
- [12:44]
B
Yeah. And there is, there's no silver bullet, especially because unless you understand a little bit more about what they're creating and why it might, it might change. So let me give an example. People use the word deepfake a lot. Deepfake is not the same as AI video. So a deep fake. That diffusion process that I talked about, where you're synthesizing the entire frame, deepfakes are targeted, so you might just change what someone's saying. So what a deepfake does is it targets just the face or an object. It doesn't have to be a face. And there's a different machine. It's not really a machine learning process, but a different targeted re encoding process that just changes what they said. So the point there is, you know, for AI videos, oftentimes you're looking at the backgrounds, at blurry parts, at especially people in the background, it still doesn't quite get like, you know, if there are a bunch of people walking in the background rather than them being blurry, they'll kind of merge together is a good example. But if it's a deep fake, it's not touching the people in the background. So the people in the background are going to be fine. Right. But in the deep fake, you're going to get edge artifacts, maybe around the face or the cheek. Or if they turn their head, it might not track well. Or if they put their face in front, their hand in front of their face, it might lose track of where the face is. So, you know, also a deepfake might not have the same audio tells. It might not sound as robotic as an AI video might. The hand motions might not match up in a deepfake, whereas in an AI video, you know, the person's eyes might look kind of glossed over or their mouth might. I mean, in both, the mouth might move kind of funny or like their eye movements might not match their hands. Like, there are some common tells, but the reason that there's not a silver bullet is because the average person can't tell, is this a deep fake or is this an AI video without having the bigger context in mind. So, you know, deepfakes are usually used for targeted manipulations, whereas AI videos are oftentimes used for more, like viral videos or like fake influencers. Like, there are different ways that they could be used. So that's why I say if you're, if you think that, you know, a politician might be deep faking, might be getting deep faked, you're not looking at their background, their background's gonna be fine. You're gonna be looking really closely at their lips. Whereas if you feel like a video might be an AI generated video, you might be looking at the background and seeing if things are warping together or wiggling in a weird way, there's a bunch of little nuances to AI video that are artifacts of that diffusion process and that's really what you're looking for. But I can dive more into each one. But that's the problem. There's no one easy trick anymore.
- [15:23]
A
Yeah, and what a fascinating time where it's even more complicated than my question suggested. Well, what category of fake stuff do you want to talk about this type of fake stuff and that type of fake stuff. And it's hard not to imagine the potential for propaganda. There has to be a whole bunch of propaganda already using these things. A whole bunch of. Every time I see, like when the whole Venezuela thing went down with Maduro being kidnapped, like, yep, all those videos and all those photos of like, which ones are those? Like, have you analyzed any of that type of like, political propaganda? Have you come across anywhere like, oh, come on, you know, European Commission or White House, like, you guys could do better than that. Have they dropped any, any good ones?
- [16:15]
B
Yeah, you know, because that's interesting because I, you know, there are some that were real. There were some that were real and taken out of context. I mean, it just been a Venezuelan crowd celebrating about a soccer game taken out of context. Or it could have been AI generated, you know, and so even, even there, there's a whole spectrum, right. And you don't know until you're, until you're looking at the page that posted it. So there were examples of pages that specialize in AI generated content where they might make AI slope. You know, like, it might be your dog skateboard. I mean, there are actual dogs who skateboard. But it's not a good example. You know, it might be your, you know, but it might be you're just like animals chasing each other. AI video. And then they might have just realized, oh, there's an opportunity to create some things about Venezuelans cheering them Astree. Okay, well, that's probably an AI video, you know, but then there are other ones that were just recontextualized real videos. But that's a great example that you gave because, you know, one type of propaganda might be to, you know, find a image or a video of Nicolas Maduro and deepfake him into saying something that would have in some people's minds, like validated that they kidnapped him. Right. Or that they took him. So that would have been one type of manipulation, right? Yeah, there's another type of manipulation which is like, let's make it look like everyone's ecstatic that they took him. So there are different vectors of how and how and why someone would do a manipulation. But let's just be clear. Most of the AI videos that you see are not deep. I know that they're not the same thing, but most of the synthesized videos are just AI videos. Now deep fakes are relatively. People understand that those are bad. You're not supposed to deep fake people. And then there is some crossover with OpenAI. Sora where it uses AI video technology to deepfake people. People, you know. So there is crossover. But a lot of the AI videos people are seeing or that you'll just see virally online are what people might have called AI slop, which is just like sensationalist videos made with AI technology. So I think when people ask about how to spot AI video, that's really what they're looking for is, is like, how do I get the AI slop out of my feed? Because I think that's what people are seeing too much of.
- [18:37]
A
And how do you go about the world now? Have you found that how as you peruse your own social feeds on your phone?
- [18:48]
B
Yeah.
- [18:50]
A
How often are you. Do you just have skepticism with every single thing you see or are there certain things where you. That'll just grab your attention, you know, has it spoiled your ability to consume content in general? Like knowing what you know or what's that. What's that been like?
- [19:05]
B
Yeah, well, for me it's interesting because I generally understand what is more likely to be AI or not based on my type of content. So I'm a big music guy. I know that AI can't do music performances. Well, now that because I understand what they look like or a lot of the content that I watch are YouTube channels that I subscribe to or that I know personally or I know that
- [19:34]
A
they're on the right, then you know the.
- [19:35]
B
Yeah, yeah, not worried about it. That's my biggest tip for people is the first step is understand who you trust and spend more time watching the people you trust rather than relying on the for you page or reels or whatever the recommended feed is to feed you everything you're watching because you can't reasonably vet every little thing. But also, you know, that allows you to not be so paranoid all the time. And you know, again, you don't need to be too paranoid about everything. I would just be wary of anything that you come across on a new social media feed that really changes your worldview in any serious way. Unless you vet it to make sure that it's actually a real thing and use those AI videos that might be more obvious. If you come across an obvious AI video, maybe go to that page that created it, watch a few more of their videos just to see like what is, what else is out there so that you're aware of it. A good example right now is there are a lot of AI generated influencers that have a very distinct. They have very distinct patterns to them. Like they usually first of all have only Been posting since late last year because that's when the AI video technology got good or they lean.
- [20:53]
A
When SORA came out kind of thing basically.
- [20:56]
B
Yeah, Sora and Then, and then VO 3.1 after it. You know, that's, that's a pretty big tell. Not a lot of adult people just like discovered social media in late 2025, especially if they are attractive and selling supplements. They're, they were probably doing that before 2025, you know, so that's, that's something where you don't need to be good at spotting a video. You can just be like, oh, like why is this person already selling supplements and they didn't post until January? Like that's pretty weird. You know, so, you know, there are other ways to figure it out that aren't being like hawkeyed about everything you look at.
- [21:36]
A
Yeah, yeah, it's an interesting, like, I mean it sounds to me that, you know, critical thinking is, is a high value activity and maybe enough skepticism just, just to think like, okay, if a. Having the awareness to catch that. Oh, that just potentially shifted my opinion on something or shifted my view and then digging, you know, opening the account and actually taking a principled view at like what might be, what the real deal might be.
- [22:06]
B
Yeah. Also just being aware of what's just trying to grab your attention and just grab your attention is a big deal. A lot of creators, when I say creators, I mean like people who are actually in this for the right reasons, they're not just like there to rage, bait you or like, you know, doing just like, no, no offense to like the people who just like eat food on camera loudly. But like, you know, I mean, it's like, I don't know, like there, there's just an element here which is like find, find people who are doing actual creative things and like you're less likely to find AI in those places because
- [22:44]
A
that, that makes sense.
- [22:45]
B
Those are places where the human connection really matters.
- [22:51]
A
So another thing you've said is, so you've said that only the bad AI videos are easy to spot and that the good ones go unnoticed. And so I think in context of parents and you know, parents trying to learn this for themselves and parents trying to set their kids up for success or know, help their kids navigate all this. Do you have any thoughts on, and maybe this is just more philosophical type of a thing perhaps, but any thoughts on, you know, the ones that do go unnoticed, like what the potential effects might be or what that might be doing to us as consumers of these things because you know, you've been, you've lived in the production world for so long and probably are thinking about, like you said, like, what's the, what's the consumer's perspective? Yeah. Do you have any, any thoughts on that kind of, and that kind of thing?
- [23:49]
B
It's all about spending time watching things that matter. And I think we've gotten away from that as a society where it's just about kind of doom scrolling and spending time somewhere where you can kind of turn your brain off. That's the first thing that I think really needs to change here because AI can do your brain wrong. It can do your passive videos better than most humans can at this point. There are some people who literally don't care. I actually don't think that's most people from what I can tell. I think that that's something that a lot of the AI companies assume that people don't care. I think most people do care and they don't want to be tricked. That's the first thing. So if you really focus on engaging with meaningful videos and creators, bloggers, I think you're gonna, you're gonna do a little, you're gonna do well. You know, I, I think it's, it's important to get kids and people just in general just interested in things, like, interested and passionate about things. And if you, if you find those things that you're passionate about, you're gonna find a person who's passionate about it online. Whether it's like, don't get me wrong, there are plenty of AI cooking videos out there, but if you start making the things that they're telling you to make, you're going to realize that it hasn't been very well thought out. Right. But if you're just enjoying it on a pure aesthetic ground, you're like, wow, that's a really pretty slice of cake. You're not going to know the difference between AI and real very well. So it's about finding those deeper connections on social media and, you know, having just general critical thinking skills gets you a lot of the way there, as you suggested. I just don't think, you know, like, what we haven't really said here is like, our current way that most people use social media is just not really ready for this moment. I mean, if your default mode is just scrolling and letting an hour pass by and not thinking about what you're watching at all, I mean, it's, you're gonna have a bad time. But I would say that the, the reason why it matters and the reason why we should Care is because that's not what we should want to spend our time on. We should want to spend most of our time connecting with real people. Even I can give an example of. Recently there was this big snowstorm in Russia. I did a video about this where there was a lot of real videos about it and a lot of AI videos about it. And if you watch the real ones, you saw some really funny cool stuff of like humans jumping out of their windows into this giant pile of snow, or like humans digging other humans out from this mountain of 10 foot snow that had covered this Russian city. If you saw the AI ones, you saw people like skiing off of rooftops down giant snowbanks. Like that didn't happen. That's not the real story there. The story isn't that there was a bunch of snow. The story is that humans were there and they figured out a way to get through it. And it was like in some ways a really difficult moment for that community. Right. So there are videos that are AI that are going to take advantage of that situation and try to pull your attention away from the real story, which is in itself very interesting and engaging. So I think that once people realize, like, oh, I've been kind of taken advantage of, like, I would have just preferred the real thing, I think that's the moment where you realize why it matters that you can detect these things or not. Otherwise your feed just slips into all AI slop. And I just, I think, you know, we shouldn't want like a digital version of Wall E for ourselves. I think we should, we should strive for it, right?
- [27:29]
A
Yeah. Wall E was a warning.
- [27:31]
B
It was supposed to be.
- [27:35]
A
Well, and there's part of this makes me think, whenever I think about these gigantic problems or these gigantic things that are going on that might have us confused or misunderstood, misunderstanding something or just, you know, viewing something that's false, it's like these things are so big that maybe there's a really positive thing that will come out that will encourage more skepticism, more critical thinking, maybe would encourage taking nothing for granted. Like if it comes to you on a screen that you don't take it for granted, you know, because I think that's a big skill that I think is worth trying to impart on young people, is like, if it comes through the screen, don't assume it's true.
- [28:25]
B
I think, yeah, I mean, I don't want to quite go that far with it because I do think that there are meaningful connections you can make online. Like, I know that, you know, a lot of, a lot of kids spend their time on video games with their friends where that is a form of socialization. Like I, at this point I'd probably, and I understand all the controversy around Roblox and whatever, but I mean if you, if you, if your kid had the choice between being on Roblox with their friends or doom scrolling YouTube shorts, there's no, there's no comparison that they should be on Roblox engaging with real kids.
- [28:56]
A
If you had to pick between the
- [28:57]
B
two for sure, if you had to pick between, between the two and versus like they will not have the media literacy skills to know if they're watching harmful AI slop on YouTube. That's not something that you're going to be able to develop that in them until later in life.
- [29:09]
A
Right.
- [29:10]
B
And so you know, the, that that's one thing. But yeah, the positive I can see from this is generally what you're saying, which is first of all we have to remember that, you know, for me I've seen a good version of the Internet where it was mostly about communicating with friends and making plans to go to go to parties and go to the movies. And you know, that wasn't that long ago. But the difference was that we were sharing photos with each other, we were instant messaging each other. Well, that's what iMessage and WhatsApp do now. So I think, you know, that is what's happened is a lot of, you know, Instagram and Facebook and TikTok and YouTube have all become primarily broadcasting platforms or like platforms for parasocial relationships with people you don't know rather than what they were originally intended to be. Maybe YouTube, not specifically. And TikTok, I kind of really promoted this world of engagement based platforms. But once TikTok became an engagement based platform and it was more about having the algorithm decide for you what to watch and it became clear that was very addictive to us. A lot of things change. I'm not saying it's TikTok's fault. It's think that there's been a slow progression towards just engagement based feeds. But yeah, if you can focus on FaceTime and iMessage and WhatsApp and video games, you're actually interacting with real people or not spending all your time. There's still a follower page on Instagram, there's still a family only page on Facebook. It's just not the default or not the family and friends. You're only the people that you follow. Anything that they've posted, those are just not the default pages anymore. You got to go A little bit deeper to find those now. So hopefully that's one of the changes as we sort of revert back because I just don't think, I don't think that a purely engagement based social media is sustainable in an AI age. I think AI is going to out engage everyone else.
- [31:12]
A
This episode of the Family IT Guy podcast is sponsored by me. IPhone. Parental controls are notoriously difficult to set up correctly. When parental controls are too complicated, parents give up. And when parents don't use the tools, children pay the price. So I created a step by step guide that shows you exactly how to set up an iPhone to protect your child from addiction and from predators. You can disable the Internet, disable the app store and disable the camera. You can make the screen black and white so that the iPhone becomes a tool instead of a toy. You will have confidence and peace of mind and your child can be protected from the most dangerous parts of the Internet. Go to familyitguy.com to learn more. Yeah, which is probably really great. I mean if that, if it, if it resulted in having people pull back from the engagement based stuff and say, okay, yeah, instead of the algorithm just driving me down some random path, yeah, I'll go to the, like, I really, I tell people about that thing in Instagram too where I use that to demonstrate what an algorithm is like. Switch back and forth between the two feeds and see how potentially even like boring the regular one is or, or the following one is versus the regular one that has the algorithm. It's very fascinating, but that's a good point.
- [32:38]
B
I mean you have, it's going to take a few years. That's, I think right now we're in this uncomfortable inflection point where right now the tech companies would like AI detection and proper labeling to save them. I don't think it's probably not going to. It's probably too big of a problem for them as well. We have to figure out as a society how we're going to react to the Internet just being flooded by AI media. It's going to take a couple years to get there. So it's going to be, I think, a very long and painful transition. So still worth paying attention to. And I don't think that it's as easy as just saying, oh, just put your phone down. Because people spend a lot of time on their phones. So, you know, there has to be some sort of middle ground where we recognize that we're going through a change. We have to change our consumption habits and it's going to Take a little bit of time to get there.
- [33:28]
A
Yeah, I think so too. It makes me feel, I find optimism in that. Like it makes me feel like there's, there's a positive path there, that, that's cool.
- [33:38]
B
I agree.
- [33:41]
A
So something that I say about AI and I'm curious for your thoughts on when I talk to parents about AI is I say two things. I say never let kids use AI alone and I state that in a very complete like no nuance. If you are using a chat bot together for school or something and you have to go pee, pause it until you get back, even like that. Just because at the current state of affairs it's, it's all brand new and there's a lot of unknowns. And then the other thing I say is for adults is that you should, and I guess particularly in my context for family IT guy, for parents, you should learn AI as much as you can. So that, I mean A, it's not going anywhere and you probably need to for your own sake at work and whatnot. But then B, so that you know how to navigate like what the limits are, what the boundaries are so that you can help to teach your kids. So yeah, so those are my two kind of core messages. How did those resonate with you? And, and do you find yourself or anybody, do people come to you with similar questions or suggestions of like how to approach these things?
- [35:00]
B
I have a different viewpoint perhaps on the necessity for kids to learn how to use it. I think that it is probably not necessary, but I recognize that that might not mean anything coming from me without kids. But I just think that the tools that you need to use AI wells to be a good writer and a good communicator. And I've just seen too many examples of people in my own lives, grown ass adults losing that ability as they use AI in real time. And I'm just watching it happen and they've already learned it. I think that the incentive to using it at all is just so great. When you're a kid and you're trying to teach them intrinsic versus extrinsic motivation and like how can you make them as intrinsically motivated as possible and giving them, you know, because, because kids lives are already so focused on achieving benchmarks and test scores and all that that is so optimized for knowledge and like you know, they're going to learn pretty quickly that no one can out knowledge. A large language model right now you can outthink it, you can, you can out creative it, you can out human it, but you can't out knowledge it. Like we're just not gonna, we're not gonna be able to do that. Maybe, maybe, you know, I'm not, and I'm not saying that it's, I don't buy the line that it's like as good as a PhD researcher. It's not like it's not there. Like we're not, you know, it's not even particularly close when you. But, but that's the thing is like you even. There's even a fundamental misunderstanding of this with adults that like having an artificial intelligence Test at a PhD level is not the same thing as being a physicist. Like, a physicist is applying and doing practical skills that might not show up on a test that an AI literally cannot do. So you take that all the way down to being a kid where you're just learning how to do things well and you see that AI can do it better than you. I think it would be. There's, oh, well, why wouldn't I just use the AI? So I, you know, I, and then I think, you know, all the skills that I've learned to where I can prompt an AI. I'm very, very aware of when I'm using AI or when I should use AI or not. And I've had it try to write, you know, a lot of what I do is writing scripts and, and hooks and titles and all that. It's awful at it. I just think, I think it's, it's really, really bad at what I want it to do. Yeah. And you know, sometimes, sometimes when I'm doing like data entry and data organization, that's when I find it most useful. Right. I think it's useful if, you know, again, if your kid's trying to digest 200 lines on a spreadsheet. Yeah. Show them that ChatGPT can do it faster than Excel can. That's an automated tool. Right. But I don't think that that's where most kids are until probably high school. And then once they even hit high school, there's questions about what it's doing to their intrinsic motivation to actually learn. So again, I just think that we're so early right now and we don't know, but I do think it's worthwhile if you're an adult, like maybe you understand what all the dynamics are here and you can definitely impart that on kids. I would just say if it were me right now, I would keep them away. It would even go so far as if I had, I would have personally conversations with my kids teacher to See where they are on it. I mean, I think it really goes this deep because just like we are at a inflection point with content and how people are consuming it, we're at an inflection point with education right now and how we are measuring students success and how we are like what the point of higher education is for. I mean these are very, very important questions for us right now. So it's so early for us to be talking about it and that's why at this point I probably would stay away from it. But I also understand the idea that you need to be able to use it well. Again. I think that if you're a good writer and a good communicator, you're going to be able to use AI when you get older. I mean it's not rocket science. It's like you talk to it like a human and if you're a good communicator it will be helpful for you.
- [39:33]
A
Yeah, that's fair. Yeah, I think that's a good assessment. And I think the same thing about it's how early it is because I've been exposed to different technology life cycles throughout my career where as technologies come and go, when they're brand new, they're always in a riskier state. Like if you use the first version of a brand new system, it's likely to have bugs and problems and things and then as it matures and it gets to more like general adoption across, you know, non technical audiences and stuff and then it gets to smooth out and then eventually for the really big stuff, it gets to the point where like everybody, you know, Facebook is a classic example of a product that has mass adoption and is very mature and these AI programs are still in their early adopter phases. Whenever I think about like kids interfacing with tech, my general preference would be that kids, if they're going to interface with technology that it be very mature, like very well understood. You know, we, we've, we've experienced all the problems. The adults dealt with those and then we give the safe version to the kids, you know, something like that. And so we're still in this like really early adoption part of the, the life cycle.
- [40:53]
B
So yeah, just probably a research breakthrough away from that safe version. Honestly, just with the way that large language models work, I mean with the just inherent nature of the probabilistic nature of them, like in order for them to be safe to use for kids so that they don't fall into any one of the many, many traps that large language models can bring you down. We probably need other research, but I mean, at the end of the day, at the end of the day it's, it's funny because it's like, you know, you need determinism, you need good old fashioned programming to probably help with some of that, but that's just not the, that's not the direction we're going right now. We're just, we're, we're currently, as of early 2026, still hurtling towards large language models, doing more and like, we're now starting to even see some of the like, most like Salesforce. I just said it's like, oh, we started doing some deterministic programming with our large language models. It's like, oh, so you mean coding. You mean you went back, you realize that like, sometimes you just need some lines of code.
- [41:59]
A
Regular old software where two plus two always equals four.
- [42:03]
B
Yep, exactly.
- [42:06]
A
Yeah. The probabilistic nature thing is, it's almost funny. It's so odd. You know, you get a different response each time. And I, I, I see it all the time because I use Claude code a lot and I write a lot of software and stuff with it and I have like, rules that it's supposed to follow and it's, it's entertaining to me almost where I see it follow them in a different way every single time because of it's probabilistic. It's like, this is really odd.
- [42:40]
B
Yeah, I learned, I, I learned I'm not fit to use Claud code and it takes a really good engineer to, to pick up when it's not following the rules correctly. So that's not my space anymore.
- [42:50]
A
Yeah, it's, it's still there. It's still in that place where to write. Like you need to have some it, you know, savvy, some background with like. Yeah, like spotting where it's going wrong because it, I, that happens all the time. Like I'm constantly like, oh, don't do that. That's, that's going to be a security flaw in two years if we go that way, you know. Okay, so you've said that you'll never let, and correct me if this is wrong, but what I understand is you've said that you'll never let AI generate your own face. And so if we map that to, again to parents and kids, how might parents learn from that and where they might want to draw boundaries with themselves and with their kids when it comes to this stuff?
- [43:42]
B
First I want to distinguish between private AI use and public AI use. I think that's a Big thing, which is if I know that I have friends who will mess with other friends by putting their face on something. I mean, look, we know it's a joke, right? And I do think that there will be some super anti AI people who would say like, oh, it's bad for the environment. Oh, like you're using stolen data. And like, I understand that argument, but that's not going to stop everyone from just messing with their friends on a text thread. So we can put that aside. Or like, you know, making a creative thing for your kid if they get a lot of joy out of it. I've seen great examples of that and I really think that that is the thing that even the AI companies themselves are promoting, because that's how they would they understand that is like the most ideal way that people could be using this is like, you know, making their kids stuffed animal fly on a rocket ship. Okay, so I want to separate that out. This is, I'm not talking about that. I'm going to talk about public AI use and how you are representing yourself. And I think that the, this, the reason this matters is, is going to be and already is to some extent impossible to check every single video or photo that someone shares to see if it's AI. So what you need to do is make decisions about who to trust rather than what media to trust. And so if I see that someone has a penchant for deep faking themselves into saying things that they did not say so that they can increase their content output, I'm not going to trust anything they really say. That's, that's, that's the decision that I have made because I think that it shows me that I cannot trust the words coming out of their mouth. Now, of course, I can personally tell the difference when someone's deep faking themselves for when they're speaking. And you know, whenever someone's just like, oh, did you notice? I was like all the. This whole time. I'm like, yeah, I actually did. I know a lot of other people wouldn't, but, you know, so like, I can tell the difference, but I still just like, why are you doing that? And again, it's, it's. The incentives right now are to optimize for engagement. I get all that. So people are doing that right now. I think that for me as a creator, showing that I would use AI to deepfake myself to increase my output or something would just kill my trust. And once you've crossed that Rubicon, you really can't pull it back. People are going to stop trusting you once you cross that line. So that's why I say in public use especially, I don't really want to use it in a way that makes people second guess anything I'm saying. Now, there have been times where I will like side by side me in an AI version of me to show a comparison because I am philosophically not going to put a real person into AI and make them do it. So it's like, okay, I'll take the hit, I'm going to do this. But I'm always very clear like, that is not me. I'm never going to pretend. At one of my videos I even said like, if you were waiting for a big reveal where I was AI all the time, it's never happening. There is no big reveal here. It's just me, right? And again, I just really don't want people to lose trust that it is me. So once you cross it, it's really hard to pull back. And that's why I tell people, just don't cross it. I don't really think it's adding that much to your life. If you feel like you need to out content AI, it is a losing game. You are not going to do it. Other AI generated people are going to out content you. So you need to just like lean into who you are if you want to be a content creator. Which I know it's not necessarily what we're talking about here, but even on LinkedIn and people will do us right. So just represent yourself as yourself, be yourself. Because once you cross that line, it's hard to pull it back. People will notice, people will remember.
- [47:36]
A
Right? It brings another angle to focus on the value of honesty and transparency and trust. I mean that's. I know my son and I talk about this on a somewhat regular basis. Same thing where, hey man, you can lose trust like that and trust that took years to build, you know, really easy to lose it. Really, really hard to build it. And I'm the same way where my, I mean everything I do, I'm always thinking long term, I'm always thinking years out. But, but for family IT guy, like I want people to take what I say at face value or at least be able to take it at face value and to never degrade that and just shoot myself in the foot by trying to generate something that isn't real. It's kind of funny actually. I get a, like a, some number of, you know, dozens of comments a month where people think that I'm AI on my videos. I'm like I'm just a computer nerd that talks slow and so might seem like a robot, but like it actually really is me.
- [48:41]
B
You know, A great example though. It's like you have no need to do it. Like, I think this is where people get. It's like, why are you paranoid about the computer? The computer nerd is like talking with real people. It's like, you know, this is, this is what it was supposed to be for. So that's not when you need to be, that's not when you need to be paranoid.
- [49:07]
A
What's one thing? Like if you could choose just one thing that parents should take away from this conversation or just from your work and your research so far, like, what would that one thing be?
- [49:21]
B
I think for parents specifically, no unmonitored algorithmic use for kids on social media is a really big thing. They won't be able to tell what is real or not. It is too advanced. And even without AI, you know, it's just, it wasn't, it was never a good idea. But there is no, there's not really a way to filter out engaging harmful things from engaging good things on social media for kids. The algorithm doesn't know the difference. So like a good example is I used to run these experiments where I had this iPad that I would, I would have it watch like. And so I mean, I would be sitting there and watching. I would watch like 12 hours of Bluey on a fresh YouTube account or like 12 hours of Bluey and Peppa Pig and Cocomelon or whatever. And I was trying to figure out like, at what point do I start getting like harmful AI videos? Because there are a lot of harmful AI videos targeting kids. Especially because the kids YouTube shorts algorithm is one of the most profitable algorithms. I don't know if people know that, but it's very profitable. The views that short form YouTube does in kids feeds specifically are insane. I mean, Mr. Beast numbers for a ton of random channels. And so there's a big incentive for these AI generating pages to do like some sort of like, kind of like edgy children's content that you do not want your kid watching this violent stuff. But kids are watching it all the time. The algorithm doesn't really know the difference between like a harmful AI video for kids and like a casual brain rot algorithm for kids. I at one point had an algorithm that was only trained on Bluey and like very, very beneficial. Like even like educational things like Ms. Rachel, right? And I was mixing some age groups and I would go to the Shorts feed and a Lot of the shorts feed was like arts and crafts or like, you know, like creative fun things like food content or something, right. I searched for ballerina cappuccino once. For anyone who doesn't know who Italian brain rot characters are, AI generated characters for kids, I searched ballerina cappuccino. I watched one Italian Brainrot video. The next video I got was a AI cat video with a sexual assault in it. It's like once the algorithm is like, oh, you like AI brain rot? Let me show you this. And the algorithm doesn't know, right? It just knows that the kids who are watching Italian brain rot might also enjoy watching this really engaging, pretty awful stuff. And from what we can tell, this is still happening. I haven't revisited those algorithms. I've had other journalists independently verify that this is a pattern where the un filtered. So like YouTube, the YouTube Kids app doesn't have the short speed, so it's a little bit better. But it still has other type of like, not great videos. Right? But the YouTube, the regular YouTube app is 23 times more popular than the YouTube Kids app for kids. It's crazy. YouTube on Safari is 20 times more popular than the YouTube Kids App for kids because they just, they just search YouTube on their iPad. They're not, they're not worried about clicking on YouTube kids, right? That's how a lot of this stuff pops up. And that's how kids get upset or like get nightmares about a cat or get nightmares about bugs because a lot of these videos are like bugs eating babies or like these anthropomorphic cats, like abusing their children. I mean, it's really dark stuff, but it looks like a kid's video to them. It just, it looks like a kid's video. So they don't know any better. And yeah, I just don't think that, I just don't think that you can trust that one of those videos isn't going to slip into your algorithm unless they, unless you have like the purest, like, least curious kid ever. But I mean, if your kid has heard about Trala Laro Trala La, they're going to look up trailer Laro Tralala, which is brain rock character, and then they're going to get one of these harmful videos. So that's what I've found. And that is, that is the number one thing I would say is once you understand, but the algorithm doesn't know the difference, that's when it becomes clear why just unmonitored use just should not be a thing. Now, I try not to Give parenting advice in general on those pages because, like, again, I'm not a parent. I know it's not that easy, but if, if it were my kid, that's, that's what I would do.
- [53:59]
A
Yeah, the principle is sound. Yeah. I mean, because it's, it's, it's, it's based in, in the logic of, like, well, if you were family values or your personal values that you're trying to impart on your kid are related to health, safety, self respect, self confidence, you know, healthy boundaries, then these are things that you might want to avoid.
- [54:26]
B
Yeah. Well, also, I just think that there's so much incredible and like, socially beneficial videos for kids to watch. Or like, I, you know, when I'm watching Bluey, I know there are a lot of Bluey adults out there. I get it. I don't have kids. And I enjoyed Bluey. I'm like, wow, this is showing them how to play and how to interact and how to interact with adults. And like, these are all great things. And, you know, I even got a message from a parent who said, well, I really wish I could stop my kids from watching YouTube, but the problem is that their favorite show is Roly Polioli, which is. That's a kid's show from when I was a kid and I'm 31, and it's like, that is only on YouTube. That's the place where you can watch it. And it's a really good kids show that, like, kids would really enjoy. Right. So she's just like, so the reason I have YouTube in my house at all is because of older kids shows like that or like reading rainbows on YouTube, for example. I think that was the other example she gave. Oh, so she's like, so I can't. She's like, so I can't completely get rid of YouTube. Cause it's like good free children's content that my kids really enjoy, but I just have to be much more careful to make sure that they're not unattendedly using YouTube. And yeah, I think that's, I think that's the thing that we would all agree with that.
- [55:40]
A
Yes, I think so. And there's this. There's a, in YouTube kids, there's a somewhat not obvious mode that's called approved content only mode.
- [55:51]
B
Yep.
- [55:52]
A
And in approved content only mode, you can approve specific channels, you can approve specific videos and playlists. And then even what I, what I really like about it is that if you approve a channel or a playlist, you can go into that channel or playlist and deselect certain videos like if you don't like those and that's a good mode. And the trouble is though that there's, well, some like TVs that have YouTube built into them. They don't have YouTube kids or you can't get YouTube kids or some platforms, it's hard to get that going up on the TV or get it going on an Amazon tablet versus an iPad. And so it's not consistently available. But where it is, I recommend people look for that. And I wrote a guide on my website on familyitguide.com if you type in YouTube I have a step by step like how to set up approved content only mode because I agree man, there's a bunch of amazing stuff. One of the things that just randomly comes to mind is number blocks to where you've got one and two year old kids learning numbers and math and like, like legitimately, you know, in a really healthy way, it's great.
- [57:02]
B
So Yeah, I think YouTube kids with that mode is great. And unfortunately on the macro level it is YouTube's excuse to not improving some of their algorithms off of YouTube kids. Yeah, separate conversation. They're just like, oh, just use YouTube kids. And then when you see the numbers about how few parents enforce YouTube kids and furthermore how few parents know about the mode you just talked about, which yeah, you know, I would like that. I think again it comes to adult media literacy in that case and educating adults is a whole nother ball game. Anyone you know, you know that. But that's really what this is about.
- [57:38]
A
Yeah. So on those lines of algorithms and rabbit holes, something that's been in the media a lot lately and that's often top of mind for me is the. I don't like saying these words but it's like sexualized content for children or of children. And there's, you know, GROK AI has been taking a lot of heat lately for enabling the creation of some of these things. And so what experience or research have you done around that topic and how that maps over to like some of these content platforms that kids are exposed to or that adults are of course exposed to? The.
- [58:26]
B
The GROK example is important because it is again it points to like the inflection point we're at where it is illegal in many places. It actually theoretically could be illegal with our current laws in the United States. It's just not being enforced with the Take It down act. But.
- [58:48]
A
Oh right.
- [58:50]
B
But look, the tech companies don't always have their eye on the ball even though they should. I Think just the first thing we should understand is a little bit how so I'll use csam Child sexual abuse material as like the acronym if people don't know rather than other. First of all because it's more algorithm friendly and second, because it's actually the technical term. So CSAM had. There are, there are massive corpuses of csam. So I just want to talk about this, which are sanctioned and regulated so that AI, this isn't an AI thing. So that tech companies, social media platforms can detect CSAM on their platforms. It even goes so far as there's a Microsoft products that scans pixels on the screen in the background looking for known CSAM locally on people's computers. Because it's such a serious thing, CSAM is not protected speech. There are very strict rules and regulations. And because you need a corpus or a bunch of CSAM in order to detect csam, it exists, we know how to detect it. And it is like it was a solved problem for a long time. It was taken extremely seriously. Then what has happened is because people are starting to be able to AI generate because it's unknown or new material. And this isn't necessarily, by the way, graphic csam, it might just be suggestive material. So for example, like putting a real, sorry, an AI photo of a boy in just the wrong context rather than it being, you know, graphic or directly sexual in any sort of way. So because the AI systems don't intuit what like context is, it's much easier to make. And on GROK specifically it's a problem because GROK on X. But I guess with GROK specifically it'll just comply. Not all the time. Like it will turn down. Like it will sometimes detect that it's making a photo of a kid, but not always. And then because a lot of these trolls, and I mean not even trolls, but just like bad people are asking Grok to undress people, you know, it just, it won't necessarily turn it down again. Sometimes it does, but a lot of times it doesn't. So we have been seeing undressed kids on X. So that's one side of it. And that's why it happens. Because it is newly generated, it's not detected as CSAM in the way that we have been detecting csam. So because of that, now it should be said that Instagram has rules against even AI generated sexual material. Especially obviously when it is an AI depiction of a kid. But it's hard to enforce. And I've seen them fail to enforce it on Instagram too. One of the tips I would have for parents is do not. And this is, I don't think a controverty would do not create a page posting your kids activities because they show up in CSAM adjacent feeds. And I know this because I've tested it and I have followed accounts that are doing suggestive CSAM adjacent material with AI. And then if you follow enough of those accounts, you do end up with real accounts, oftentimes posted by parents of their kids. In that context, it's not that they're using photos of your kids, just like it'll be a photo of, I'm not going to describe it, but it'll be a photo of sexually suggestive nature created by AI with a kid involved. And then right under it is a real girl performing a gymnastics routine or something like that. And then it's another sexualized version. And so again, the algorithm's just picking up who's watching what. And it doesn't know that one of these is AI, one of these is real. It just knows that there are some predators out there who are watching both of these things in their feeds. And it is a really big problem because it's creating an on ramp for people that did not exist before because a lot of the AI sexualized accounts on Instagram are using hooks or styles of video editing that might be suggestive and sexual in nature involving kids. And if you follow one of those pages that does that and you see one of those, it can devolve further. So I just, I think it's a really scary situation that's happening. I've tested it, I've seen it happen. I've seen these rabbit holes turn very dark and very real very quickly. And the, you know, the, the thing that I have definitely landed on is I now have seen firsthand why you should not post photos of your kids on social media. And that, that has been a very frightening realization for me.
- [64:15]
A
Yeah, it's one of my, I don't know how else to put it, but, you know, least favorite things to tell people. You know, because I love the idea of being proud of your family and wanting to share what's going on and how much you love your kids and how you know all the things that you're doing and, and sometimes even in, in a otherwise relatively private Instagram account. But a lot of those private Instagram accounts could have 100, 200, 300 people that you don't know who's who and who's doing when. You really don't know. Yeah, and, and, and it's it's, it's too bad. And so then, you know, you have to go to. People ask me a lot. Well then what should I, what should I do? I want to share my stuff with my family. So then you got to do it via encrypted sharing things. Like there's a company called Proton, they make a thing called Proton Drive to where Proton can't even see your photos. You can see your photos and if you share them with somebody else, that person can see them and that's it, you know, or share them on. On signal. You know, that way you can see what you're sending, the recipient can see what you're sending. Nobody else, you know, signal can't see it. Nobody else can see it because. Yeah, I agree. There's this stuff, the scale of the problem is so big that. And then we've got schools posting photos of children on social media.
- [65:40]
B
Yeah.
- [65:41]
A
Because they think it, they think it's fun. I mean, and I don't mean that in Jess like they really do. Like this is fun. Look, we're posting a thing of our sports thing or our art event or whatever. But yeah, I agree, man. I think the, that, that's what the grock thing. I made a video the other day on it basically saying that like this is a reminder that we shouldn't post our kids faces and photos on the Internet that we have to grok.
- [66:05]
B
It's very direct too. I mean it's just in the social media app, just like grok remove clothes and that's all you needed to say. And I think that that was, that should be pretty alarming that a tech company would make it that easy without thinking about the implications of it. And I, you know, while, while I do think that there, like you said, there are ways to share it with the people who really care and where it really matters. You know, that's probably again, that goes back to like what social media should be used for and what the Internet should be used for. And it probably shouldn't be for broadcasting. Your kids should be for sharing, you know, cool photos of them that you're proud of them with, with your close family and friends.
- [66:49]
A
And I think it's an opportunity to demonstrate the value of privacy to our kids where we put ourselves in the positions of whether we're, you know, uncles, aunts, parents. We are stewards of their privacy. So they, they are, without them realizing it, they're entrusting us to, to respect their privacy so that when they are adults, even if it's an innocent thing now they don't ever look back and think like, man, I wish that didn't happen. I wish that my photo wasn't posted when I was at Disneyland and I wish my, my pictures of me doing whatever I was doing wasn't posted, you know, because now I don't want. Because it's, it wasn't their intention to do that and it got done on their behalf. And so like, you know, I think, yeah, I think we, we are, us adults are steward. It can, can be and should be stewards of privacy for children.
- [67:44]
B
Agreed. Yeah. And also agreed on Proton. That's what I've been starting to use too. Just out of curiosity, what about imessage? How do you feel about imessage and Apple photos?
- [67:55]
A
I think that generally speaking it's great. I, I think there's, there's probably three buckets of relative privacy. The least private bucket is like literally posting it publicly on social media. And even if you're posting to a private group, whatever, like posting on social media because once it, once something touches the Internet, it's no longer yours.
- [68:26]
B
Right.
- [68:27]
A
That's like the, so that's the most, that's the least private bucket. And then the bucket in the middle that is relatively private is Google Photos, Apple Photos, you know, imessage, perhaps regular text message to a certain extent to where. Yeah, what happens is you. Those photos still don't belong to you anymore because you're placing them on the computers that belong to somebody else and they therefore become theirs. That being said, companies like Google and Apple and Microsoft are some of the best security companies on the planet. Like, they have fantastic security which keeps outside parties from getting into the stuff that's inside their servers for the most part. I mean, Edward Snowden kind of showed us that that's not always the case. But it's not available to the world. It's only available to Google and Apple and Microsoft. And so now can third parties come request those photos for any reason? Especially if it's within a legal boundary? Yes. And then you have the most private. Where you are using an Internet resource, like with Proton Drive or with signal, but it's configured in a way with the right types of encryption. Specifically what it is, is, is in encryption you have this concept of a private key. And imagine, you know, like the, the maybe. I don't know if this is a great example, but like when you go to the bank, if you think of the movies. No. Okay. Think of like nuclear guys that are going to launch nuclear missiles in the movies and they got to Put two keys at the same time and turn them. You have to have both keys to do the thing.
- [70:28]
B
Yeah.
- [70:29]
A
And so if one of those keys is your private key for your data, and you can't access your data without that private key, that's what Proton Drive gives you, your private key. Where when you use Google and Apple and Microsoft products, they have the private key, they don't give it to you. And they do that so that they can manage the data and manage the system, but it also gives them access to the data. So these, when you. When you use a platform where they give you the private key and they don't get it, then you have true mathematical privacy as a result.
- [71:05]
B
That makes sense. Yeah.
- [71:07]
A
Yeah.
- [71:08]
B
So those are kind of, I think what was really interesting to me, and this is an example of. I'm also using this as an example of like, AI detection is when I was visiting my family over the holidays, we realized that my uncle, who had been in the, you know, Apple ecosystem since 2006, had a lot of photos of all the cousins and us as kids. And it was to the point where he could just type in Jeremy Carrasco in his search bar on Apple photos and get a picture of me when I was 10. And I was like, oh, my God, I could not believe how it was able to trace my face back. And I'm actually giving this as an example of like, I think it's actually a relatively good use of that technology that I was able to just search for my face and send me all the photos, which is great because now I have all these photos of me that I didn't know it existed until this year.
- [72:10]
A
Oh, that's cool. Yeah.
- [72:11]
B
It also shows that AI or, I mean, that's not really an AI detection, but like, face detection is a solved problem. If it can run locally on your phone, it can probably run on Facebook servers to detect deep fakes. So I'm hoping that that's. People start understanding that that is a different problem than AI generation.
- [72:33]
A
That's the example I give, by the way. Another thing worth noting on the Apple stuff, specifically, Apple is one of the very few companies, I'd almost say only company, but I'm not 100% sure that offers the choice to where you can ask for your private key and they'll give it to you. So they have a thing that's called advanced data protection mode and you can turn it on in your icloud settings. Now when you do that, I don't generally suggest that everybody do that because if you lose that private key, you lose access to your icloud account. Same thing for Proton by the way. Like Proton has a thing on their website, like if you forget your password, we can't get you back in so don't forget your password. And they have like recovery mechanisms. You have to print stuff out on paper. Same thing for Apple. If you, if you turn on advanced data protection mode you can print out your private key. It's a bunch of random characters on the page and I, I reckon if you do that you should print out multiple copies and don't label them, don't say this is Jeremy's icloud account. Don't label them. You know what they are. Place one in your safe at home and place one in a different bill, you know, at your parents house in a safe or in a bank safe deposit so that you never are without a copy.
- [73:54]
B
Right, interesting.
- [73:57]
A
Yeah. And so that stuff is really good. And yeah, I mean the face detection stuff is amazing. And I remember the technology to figure out what's going on in photos has been really good for a really long time. I had a friend of mine in 2014, we were at a wedding together and he was one of the software engineers for Google Photos, he worked for Google and he was telling me, he's like, he's like, we don't even need, you know, you can put GPS data inside of a photo, right, and figure out where it was taken. He's like, we don't even need the GPS data to figure out where the photos were taken. With like a high level of certainty. We can, we can use all the photos that we've ever seen. And, and if we can see out the window or we can see an object that we've seen before, then we can figure out that it's in the same place. And then also since we're following your Android phone everywhere you go, we can then extrapolate that if you took a photo here at this time and then you were over here at this time like that we, that we know what the breadcrumb trail is. And that was, that was 1112 years ago. And they, they knew everything that's going on in the photo, like who and what and emotions and places and all because they're an advertising company and that helps with their targeting. You know, what you have going on in life, what kind of bag you're holding. And so in 2026, I mean this stuff is, I figure if, if you, if you let a photo of yourself or you let a, in this context, you let a photo of your kid touch the Internet. Those computers that are run by these companies know everything in that photo. You know, like every. Every little detail that you'd never imagine.
- [75:41]
B
So definitely.
- [75:45]
A
So, yeah. Jeremy, where can people follow you if they're interested to follow your work? Where should they find you?
- [75:52]
B
Well, right now I'd say just look up Jeremy Carrasco on your preferred platform. I'm going to be relaunching a platform. Maybe by the time this is out. I'll say it in case that it is out by now, but riddance AI and we're going to be all focused on, figure on helping people figure out what's real in the age of AI and give them tools to do it, because this problem's not going away. So riddance. Like, good riddance. But right now, my. My accounts are at ShowTools AI or ShowTools AI on all platforms. That's probably going to change because I don't want AI in my name anymore. It's a whole. It's a whole thing. I started this seven months ago, and a lot has changed in those seven months of, like, how I'm, how I view my role in the world. And back when I started my pages, I thought I was going to be showing people how to ethically use AI tools, and then I ran out of ways to do that myself. So I stopped doing that because I think most uses are not ethical. But, you know, it was just so early on in my journey there. So it's all to say that if you look up Jeremy Carrasco on the social media platforms, you'll probably find Find me.
- [77:12]
A
Jeremy, loved the conversation today. Thank you so much for your time and thanks for coming on the Family IT Guy podcast.