What I really think about: Is Wibberley Actually AI? (live)

Hey, you were live. Thanks for uh coming to watch this. In this video, I'm just going to be reacting to sorry, in this stream, I'm going to be reacting to this video that I made three months ago, March 6 or published March 6. And I want to keep it fun. So I also want to get through this within two hours.

So if I spend more than five minutes on one minute of the video, then I have to do a push-up. And uh if I forget to actually uh look or if I forget to start the timer. So the five minute timer then I have to do two push-ups. Uh here let's start a timer. Uh five minute timer. Okay, that's ready now. Cool.

So let's begin. Uh, there we go. Start. Okay. One minute. They think they're having a heart attack. So, this is YouTube's fastest growing doctor. But does he actually exist? My name is Naz. I'm well known in the health research space and I'm currently engineering in AI companies and I know how AI works.

Let me show you an example. Keep watching this AI generated slop for the next 20 minutes so that one day I'll be able to sell you a whole lot of books. I believe it's only a matter of time before YouTube takes these videos down, but I want you to be able to identify AI on your own. I was pretty proud of that hook and I'm still pretty proud of it.

I think it's probably one of my best hooks. Um maybe I could have put more proof early on. Um just like the the later parts uh I sort of like buried those deep in and as a result people will drop off probably around this point which if I actually show you the data they do they drop off a lot here and we'll get to that eventually.

Uh one thing that I want to say now is okay how much time do I have? Okay. Minute three and a half. Um I'm not sure if YouTube would take them down. Uh even if they did believe that it is AI. Hey Kei, thanks for joining. Um yeah, sorry about the late one. I want to try to do more uh early um early ones.

So uh yeah, if that works for people that are in the UK as well. I think this time is probably good, but it'll end up probably being on the weekends. Today just happened to work because it's Junth. Oh no, I'm running out of time. Wait, is this one? Yeah. Okay. I don't think YouTube will take these down.

And to be able to identify AI on your own. So, in this video, I'm going I think I have something else to say. Okay, we've got like three minutes now and I got to get through the one minute. Just as a reminder, if I I'll tell you when I get I'm going to tell you about the limits of AI factories and how they fit within this YouTube channel.

Let's go. That was pretty I was pretty proud of that one as well. Uh that transition I thought it was really cool. Uh I made that one using VO3. 1 if anybody wanted to try doing something like that. I'm uh I thought I also I don't know if you noticed, but I tried to make the music sync up uh with the animation.

That's sort of why I did this. Um but I'm not using this one anymore. I also changed sort of my branding. But yeah, I also I also made or generated the music. I should probably tell you that. So when you're engineering a factory, the first thing you want to know is make sure you understand the problem.

And we can do that by looking at a very simple example of what the problem is and describing our goal. So let's look at an example. keep watching this AI generated slop for the Okay, so I immediately got into the lecture part of it hoping that people were just like uh hooked, but I think if you look at the data, a lot of people do do drop off uh like or or it tapers off until probably probably until like when I actually start talking about the channel, people didn't didn't really want to understand how the AI factories or not everybody did.

That's okay. Um but if if you are interested in that that's I think that's worthwhile. Okay, we got one more minute for our timer. So then okay, so this timer is if I can't get through one minute in five minutes, then I have to do a a push-up. It was just a push-up, but that kind of seems easy. Yeah.

All right. I think the first minute was a success. I wasn't nervous about this one because there was a lot going on the first minute. Okay, let's let's start again. And yeah, welcome again, Kei. Okay. Oh, that was the push-up guide. Next 20 minutes so that one day I'll be able to sell you a whole lot of books.

So, our goal is to determine whether what I just showed you is AI. Now, in this example, it's easy. I can hear why some people think that my voice sounds AI here. I'm not sure what happened. I think maybe there was too much resonance in the post-processing and it got compressed. Um, which just makes it sound like you're talking into a bottle.

And I guess that sounds like sci-fi AI. It's not technically modern day AI, but if you if that's your impression of AI, that's probably what you'll get. Because I just made that, so I can tell you for sure that it's AI. But what if I didn't tell you? Yeah, I did make that one. I needed to notice and that's I didn't make that.

And I that's why I I tagged it as this now, which I hopefully I'll show you guys eventually that the video was simple to show you what I mean. I'm going to play the original video on the left side of the AI generated one. Keep watching for the next 20 minutes. One day I'll be able to sell you a whole lot of books.

Yeah. So, we got the audio in unison at the same time here. whatever if you if you're familiar with the tools, what I use is character replacement. And so that's when you just replace the movements of something uh in a video with another character. And this is just a synthetic doctor that I generated as well.

You don't want to do real doctors. And you can see they're in my background or my home. Yeah. So, there was a lot there. We'll go over it one by one, but first let's pay attention to the audio. Keep watching this AI generated slop for the next 20 minutes so that one day I'll be able to sell you a whole lot of books.

I did I do notice a little bit of u abnormal blinking here. Keep watching this AI. I'll be able to But first, let's pay attention to the audio. keep watching this AI generated slot [clears throat] for the next 20 minutes there. That was like a really fast uh blank, but there are lots of AI tools uh that do a better job at that.

Um whatever I believe um that channel is using is doing a really good job. I'll be able to sell you a whole lot of books. keep watching this AI generated slop or the next 20 minutes so that one day I'll be able to sell you a whole lot of books. So the the voice I was I generated with 11 Labs which is the leader in [clears throat] um voice generation and it I do think it sounded pretty good in that one.

I I think I also sounded normal there. That is a real video. Um, but I would say that they're on par. So, it's it's pretty good. Okay, we're almost at the second minute. The first thing with the audio is that Okay, we're over with the second minute. That's going pretty well. Maybe it's too easy. Let me know if you think I should reduce the timer, if you actually want to see me do some push-ups.

But, uh, okay, let's go through that one. the second minute or the third minute, [clears throat and cough] the AI had very predictable audio. Like it was smooth. It's exactly how you'd picture somebody saying it. Whereas when I was saying it, it had these natural slash awkward pauses. And yeah, it's really hard to get it to It's not really hard, but it's a lot of work to get it to do those pauses and splits.

And we don't have, you know, we don't have that in this in that sample. Uh, pauses and splits come naturally and free for humans. So, you'll see a lot of them unless they super highly edit their videos and do a lot of takes. Um, so that is it's not a exclus exclud or excludory factor, but it is it is something to factor in because if you've got very long videos and it all looks like perfect takes, then you've how many you you should be able to do the math on the bad takes and just calculate or estimate how much time um the creator used to create those videos or speak speak those lines if it's a human the I1 was simpler in movement each Dr.

Alex is too perfect no hesitations no uh no pauses yeah exactly I agree with that uh Mr. Shan G check his video from yesterday at 6. So yesterday uh about until time 8:52. I I do want to do that maybe. Okay, I could I mean I could do that. Um I do want to focus on this one. So maybe we'll do it after this fifth minute.

Uh sorry the third the third minute of this video because I am running out of time here. Movement is very well planned and executed. So, for example, let's compare the blinking. The AI does a full smooth blink. Yeah, I agree with that. And I didn't even blink. So, yeah, I thought that was pretty pretty cool that they just added a blink in for free.

It didn't look super natural to me, but uh it was it was interesting to see. Yeah, I was adding things just so that it looks more natural or like a simple human. And then let's compare the hand raises. Okay. Uh you might notice that I look super different from here. It looks more natural or next cut like a simple human.

And then let's I actually had to take those on separate days. Uh this was original. I had really good lighting. It makes me look red. And uh this other day wasn't um as good. So I look different. I don't know. Maybe you could say this looks better. compare the hand raises. I was also surprised that they added that for free.

So again, I didn't raise my hand and I was just adding that because it was more in pattern with the rest of the video. So it's simpler. Just keep doing the same thing. So simplicity is the ultimate telltale sign of AI. Okay, we're done with the third minute. So, we can stop that timer and we can look at that if you'd like us to.

Mr. Sean G, are there other comments on his videos bots though or should a load of folk that just don't see? Yeah. So, we're having this discussion that I don't know how people get that. Alex is I believe there are a lot of people uh that don't get that and they are the ones commenting for real. Um I just mentioned that in a in a random comment that I responded to the other day.

not a random comment, but like I randomly mentioned in a rand uh in a comment that it's unlikely that YouTube is so incompetent that they miss swarm bots on videos because swarm behavior is easier to trace than individual behavior. um if you from a networking perspective and they'd have to also generate the swarm to be uh like the demographic that they're targeting. So that would just be too easy to catch especially at the scale that we're supposeding uh bots are on the channel.

Okay, five. That's a timer. And Mr. Shan G said, "Check out the insulin video." Okay, I'm gonna go to that channel. I'll try to do this quickly so we can stay on time. This one, that's from yesterday. So, for [ __ ] sake, [screaming] Just imagine uh in which video you discussed your AI stack. I remember seeing it somewhere.

That was in the uh the latest video that I had the robot council. Um oh wait, my AI stack like the for the defake video. Um or like my work AI stack. Okay, let's go to that A52. right here. It's 852. Okay. Around it. There is a way to do exactly that. You take two groups. You give one of them a real magnesium.

What are we looking for? Okay. 852. So, I'm going to start 10 seconds earlier in case I missed something. So what actually happens when you test a magnesium supplement directly stripped away from the healthy diet around it? There is a way to do exactly that. You take two groups. You give one of them a real magnesium supplement and you give the other an identical uh deep fake stack.

I don't publish the one that I used uh yesterday. I don't want it to be easily replic replicable for the last video. Uh but there are many tools. So that one is in the last video uh like where I enumerate all the tools that exist out there. Dummy pill like a smarty or just a sugar tablet and you wait.

What am I looking for? Um Mr. Shanji that is what we call a randomized control trial and it is the closest thing an ident around it. There is a way to do exact did I get the right time stamp? time 8:52 um yeah yesterday June 18 there is a way to do exactly that oh is what the is in the audio okay exactly that you take two groups you give so what actually happens when you test a magnesium supplement directly stripped away from the healthy diet around it there is a way to do exactly that.

You take two. Oh, the Okay. Yeah, the exactly that is like smashed together or compressed on an audio timeline. Yeah, that's a good tell. Nice find. Um, thank you for sharing. We We could use this as like u more evidence if people need it. Thank you. Yeah, thank you for sharing. All right. Trying to I'm pretty bad at staying on on task here.

All right. So again, so every five every I'm trying to get through this video uh one minute at a time within five minutes for each uh bit uh each minute. And if I don't, then I have to do um I said it was one push out, but so far this is looking pretty good. I'm feeling confident. So let's up it to five push-ups.

All right, let's try to get to this one. Or AI generated content or AI factories. Let me tell you why. If you make a factory, you want to minimize the cost that it takes to produce something, right? You want to have a lot of output, but you want to have really low costs. You also want to ensure that all of it has really good quality for AI generated content.

That means you want it to look realistic. And to do that you have to maximize simplicity technically or specifically AI engine. Yeah. So that is like one of the things that I consider to be the easy way to tell whether something is AI. I say simplicity here, but if you try to add complications or variance to AI um to your to your AI system, then the error rate increases.

That's why it's it would be super hard uh to um for these channels to add a lot of variants all at once. And that's pro the main guiding principle that I use when I'm evaluating whether a channel is AI or not. Uh if you if you sort of force them to add variance kind of like I am with these videos, they make more mistakes um and eventually just foil themselves.

Okay. 320 engineers minimize what we call variance and that's because you want to ensure that everything follows a predictable curve. Show you what I mean. Look at this plot of blink rate and emotion. If you were building AI, you'd have talking about you've got the points that are human output and the line of fit as AI generation.

This is just a simple two-dimensional curve. Um, but you can imagine that you can measure anything as a dimension. And in a video, there are tons of dimensions um going on at the same time. And it's not just here I have for example blink rate but it you know if you if you think back to calculus the the first thing that you're going to look at is like how many blinks and then you look at blink rate and then you look at the acceleration of blinks and that's just one thing split into three dimensions.

you've got many variables or dimensions um that are all going on at the same time and it's really I think it's to for AI to be able to learn and replicate those in a concerted way. Uh it would require a very large model and a lot of compute um power computing power. So just energy being wasted. Uh that's why that's part of the reason why it fails to replicate a really high variance.

Here's the important part. By minimizing variance, you ensure predictability, which means Okay, that was that was I'm still a minute 30 ahead. So that was good. Maybe let's reduce the timer to four minutes. Some make myself sweat a little bit nervous. I don't know how use how to use this interface. Okay, there we go.

Let's start. You only have to test a few of the outputs to ensure that all of them are high quality, right? Because you can have a human evaluate a couple ones in a lot of detail at a low cost. So this is what AI engineers actually do in production. They minimize variance so that they ensure some degree of quality.

And so in the video context once you have 10 or 30 the light keeps changing on the video. Um I should have paid more attention to that while editing. So I actually edited this video myself. A lot of the newer ones are edited by somebody else. Seconds that are good. you know that the rest are the same and you could stitch them together and you could even change what's being said and their body language for that.

Okay, so that's like scary, right? But what can we say? What can we notice? And how do we know whether something is AI? We can anybody notice how I changed the music here to make it a bit eerie? I also generated this music. Uh maybe this was a bit too much. I don't know. actually do it. Look for simplicity.

If it's too simple, it's more likely to be AI. Humans are complex. And that's not to say that you can't have simplicity or consistency. A lot of people do that. But people like want to Yeah. two things there. People are complex. Like in the real world, people are multi-dimensional. um maybe you'll think of a of a doctor as just like basically their role, somebody that heals you, but they have lives outside of that.

And so, um those lives could differ a lot. And that's something that I just like really believe in when I'm looking for a person and trying to understand them and evaluate whether whether they're being a humanlike with me, I guess. Uh I I kind of try to like look at how many dimensions they ex they expose people.

Okay. So Oh, and then what what else did I say here? There's something else that I have to people do that. But Oh yeah. Um people like want to create I'm saying that a lot of people keep things simple for consistency. And that's also a thing like the when when you have a process or a system I'm running [clears throat] out of time here.

When you have a processor system uh that works why would you change it? So that that a lot of real humans actually do to avoid changing their processes or systems and they they don't um they don't add variance but you'll still see variance in some of them. Yeah. Eventually their their old content is just going to stop working and they have to make some sort of massive pivot.

Yeah. You'll probably notice that if you know any famous YouTubers that have been all around for a long time. Okay, I was starting to sweat there because we only got 40 seconds left. I guess even the stakes of like having to do five push-ups is not are not even really that bad. But it is different. So, it's exciting.

Okay. Uh Sandep Sueda Suia is another fake channel with a madeup story and a fake LinkedIn profile. That's interesting that it has a fake LinkedIn profile. I thought that LinkedIn was really good at detecting um bots or other other characters characteristics. Um we could look at that one some other time.

And then okay, talking about simplicity. Yeah, one of the things I've noticed in his videos that plant shake when he leans on the table that doesn't seem simple. How does it do uh how does it they do that? So I don't have the the vision on what models I believe um are being used on that channel, but you could prompt models like video generation models to make to say something like make the plant shake, move the arm this way in this trajectory.

Um what else can you do? You could even have a dedicated model or or a small AI piece that just shakes the plant and then you could combine the videos. There was one one part that somebody sent me where they uh the character or the avatar supposedly interacts with the plant, but I could notice that like it looked like I don't know if you can see.

Oh, no. Okay. It looked like they they had about a finger left behind the plant, but they still touched it. So, that didn't make sense uh to me. It's they pro which I guess just based on that actually. No, I I don't think that gives us information to conclude whether it's separate model or the same video generation model.

Um yeah, it's one of those two. Okay, let's get back into it. Okay, different things. You want to test different things. That's just like part of the creative human mind. If you're not testing different things, you're probably limited in some way like an AI factory. But you can imagine that this is on a spectrum.

It's from 0ero to one. And on one hand you have maybe AI probably AI and on the other hand you have maybe yeah so I said I have spectrums in the mind uh because I have some like too much maybe physics background and they they tend to have spectra instead of like the electromagnetic spectra or like the light spectra instead of dimensions. Now I I prefer to think in dimensions or parameters.

It's uh it's the same thing but it means more to more people. Maybe human probably human. And I encourage you to set your own threshold for determining whether something is AI or not. But whatever it is know that you should be moving. Yeah. I I think that will be subjective for people, right? Like you can have somebody that like may be considered extremely paranoid and they want to see really high variance.

Uh hopefully I don't have hopefully people don't consider me to be that. I um I don't want to I don't want to distrust everything. Uh, but you could have somebody that, you know, doesn't doesn't care at all and they'll have their threshold for judging whether something is or not at a very low level or they might not even have a threshold, right?

Because they just don't care about finding out or knowing the answer to that. ing this threshold. As AI gets more advanced, it's going to be harder to tell whether something is AI or human. As AI is able to make more and more complex things. Yeah. Right. Like AI is getting better and you could also like pay more compute to these video diffusion models to just make better videos.

Um, but it is it's more expensive to to add more diffusion. It's more expensive. It does get cheaper with time as GPUs get cheaper and energy gets cheaper um for data centers, but to add enough variance such that it actually looks human to me probably um it would be very expensive. It would just not be a profitable operation.

I personally think we're a long way from that. I can recognize AI from a mile, but a lot of people already can't. That's because they've never had this training that I'm sharing with you right now. Yeah, people totally skipped this um on the back end. or not totally, but like for this part, I think it was like 35% of people that actually watched um this part, they just skipped to this one, which is fair uh if you don't have time, but it I think having that helps you recognize other channels.

All right. So, now that we know that about AI factories that they make simple things, I got to start. I got to restart the timer. Almost got myself there. What about this YouTube channel? Does this YouTube channel make things that are very simplistic and therefore likely higher? So, I decided I'm going to score it in six.

You can tell I have a problem with like shaving here. I always miss a little bit. Maybe you can tell. I don't know. Just like seeing it on this video. It just reminds me categories of simplicity. The first category is facial progression for which this channel scores a very which is like yeah part of I don't I'm not saying that people should like shave in a shitty way or like be bad at shaving.

Oh, I I thought I said I wouldn't swear for this one. I don't want to. Uh, I'll think of that back. Sorry if you if you don't like swearing. Uh, yeah, you don't have to like be bad at shaving or anything, but uh w with a a beard that's as groomed as like the channel has you'd expect that to change a little bit from video to video.

Maybe not from video video from video to like the next week because supposedly um their their claim is that this um the real doctor records once a week and then puts them all out once. And so you'd expect to see seven videos that are like of with the same beard and then seven videos that have a slightly different beard. Unless you have some amazing precision when grooming.

Uh, okay. I got like two minutes. I just want to get through this so I don't have to do the push-ups. But very simple red X. Look at the videos. His face never changes over three months. No growth. I I didn't really like intend to capture like this this AI artifact with uh with a teeth retracted. I guess people also probably look like this at some point.

Um yeah, but when I when I saw this picture later, I was like, "Oh, that looks kind of AI." The point that I'm making here is about the beard, and it just looks the same. And hair, no shaving. No blemishes. I don't know any person like that in real life. Do you know anybody like that? Like I I don't know anybody.

If if you meet somebody like that, introduce me to them because that is awesome. Like to be able to maintain your hair at like, you know, I really want to meet this guy. The second category. Yeah. I I just like to me that was like what really was like what the face doesn't change at all. And you know the uncanny feeling that a lot of people mention a lot of people talk about uh the chess hair as well.

We can restart the timer because just got through that minute. Uh Robbo Council respect I found the word useful in describing people and society related phenomena. Exactly. Because categories are usually unhelpful. Yeah. Big five is a better model than 16 wire breaks types. Yeah. Uh but just like because you have I guess it tells you about the dimensions as opposed to like having to memorize 16 buckets.

Clubby uh this is what I expressed in his videos. I said this is AI as he never changes. Not on single hair, not on the scrubs, nothing. Zero. Yeah, you'd expect the maybe not human hair on the scrubs, but like supposedly the the real doctor has a dog. I don't know if that's true or not. Probably is.

Uh if you know the real doctor, maybe you can verify or deny. But you'd expect some dog hair. Like I get dog hair on myself all the time. And that's actually why I wear white. Uh because she's white and she sheds like super hard. Uh I wonder if I got some on me right now. We just uh she's out with her mommy right now.

Or not mommy, my wife. Okay, I don't have any hair on me right now. Ah, my point is foiled. But there is one here. Oh, I know why I'm doing this. Hold it up to the camera. Look at that hair. It's a white hair. It's not mine. It's my dog. I'm not green yet. Okay, cool. Let's do the next minute. Uh oh, wait.

Before that, there's a comment if I can also add AI stuff has a cartoon smoothness but real video um with real people like. Yeah, that is so that is something that a lot of people um have noticed maybe here maybe in this early on. Where was it? Oh no, wrong video. Maybe at this point the December 2nd, 2025.

Okay, I gotta I gotta press there. It was noticeable than at March 4, it was less noticeable. uh like they they began to add texture which if you see my deep fake video, the latest one that I had, you can add you can have texture in them, but they got they got better at at adding that I guess to like hide this and eventually now you can see that it looks uh super highly textured.

Um so it is possible to add that texture. Uh, it took him a while to do it, but um, it's definitely throwing off a lot of people. Like I think I had a comment say, "Oh, I think he just used a filter or like a smoothing filter on the videos, which is Yeah, I'll try to stay on topic here." Uh, okay. Got to restrain myself.

All right, back here. Four minutes. Let's go. category is video length for which this channel actually scores a green check mark. Very complicated. A lot of the videos on the channel have a length of 15 to 25 minutes which yeah one of the earlier comments did mention uh that it's possible to generate something that long.

Um and I agree with that. I think at the time I was yeah I would I was skeptical on on the amount of work that it would take to make a video that's long by AI but I generated one the 40 14-minute one. It wasn't great, but it, you know, you just got to stitch some things. And [clears throat] I looked at the deep faking tools or tools that can be used for deep faking out there.

And there is no limit on how long you have to go anymore. You can you can probably do it all in one shot. So, at this point probably now would turn into it would probably turn into a yellow I think it was a circle or something. And yeah. Oh, we are not doing well on time here, which is pretty long. Like that's like a TV show episode.

So I would score that as a a complicated grain. If it was AI generated, the operator or the owner would have to engineer a really good quality control process. I think it's possible, but I think it's hard, honestly. So, I'm still going to give it a green complicated yes. All right, the third category we already talked about, movement simplicity.

The channel scores a neutral yellow circle. So, in my opinion, the channel's movements do seem a bit too elastic. They bounce perfectly like a bouncy ball, arm ball or rubber band or whatever. Like for example, the scrubs fold so perfectly every time they crinkle. Okay, great. We got through that. But I still have to comment on it.

Movement simplicity. Um, here's the thing that it AI doesn't seem to be capturing yet [snorts] or these diffusion models is like basically the the the elastic curve of objects. I want to try to Okay, I'm gonna I'm gonna have to just like show up. Oh, did that have audio? No, sorry. I didn't I didn't realize that that other one didn't have audio.

I was just looking for the elastic curve, by the way. Yes. Yeah. Sorry about that one. Thanks for pointing that out, Kefi. Uh, I was just looking for the elastic curve, but yeah, I'm definitely going to have to do push-ups for this one. But the the uh like objects in general have an elasticity curve, so they're super multi-dimensional.

Sorry. uh the dimension of how they change actually changes with their strain or like how much their position or how much how displaced they are. AI right now seems to be not aware of this curve or the diffusion models and so a lot of um I guess it's too complicated maybe to capture that for different objects and so yeah okay we lost that we lost the formulas so I'm going to have to do push-ups but because of this um a lot of things seem to have like a very perfect elasticity they they like I say in the video they just like compress and contract really easily.

That's something to look for. It's possible that, you know, they might get it or AI might be able to get around this quickly. You just need a bigger model, but right now or back then I'm not um yeah, it wasn't like this was a problem. Okay, so I have to do push-ups. Sorry about the audio again. Uh, this is a random stress strain curve here.

All right, so let's see what a push-up looks like. Uh, so I can know how to do one. So 30 push-ups a day to some people might sound almost too small to make any real difference. But there's a reason this one humble movement. Okay, so like this. Maybe like this. Is this on camera? I have a bit of a latency thing.

Sorry. I don't know if Okay. One. Okay. No, it was like It was like This is a push-up. So, one, two, three, four, five. There we go. We got it. Okay. Uh Robin comment yes it is the sixth day after his last haircut and every day for the last six months and that a natural can level of smoothness. Yeah.

And everything about his appearance and delivery. Yeah. Sorry you guys are probably just like I'm getting a bit of a latency thing. So here, let me try to refresh this. Oh, am I still alive? Yeah, I'm still alive. Okay. Yeah. So, that's five push-ups there. Oh, man. All right. Okay, cool. We need to go to the next one.

All right, let's start the timer or whatever. They're just like and then it's it's awesome. I It's very elastic, very springy, very bouncy ball. It's I've never seen that in real life. On the other hand, it's not as I I tried to be like concise when I was describing on video because, you know, you've only got so much time.

Uh, but the the explanation that I that I did just now I think was better. Um, yeah. Bad as the videos I generated. I think the videos I generated had movement that was too simple. So, instead of a red X, I'm going to give it a neutral yellow circle. All right. So, the fourth category is the presence.

Yeah. Yeah. Uh, I was saying that like they do have some like their their movement the movement in those videos isn't that bad. So that's why I gave it a yellow circle. I'm not really sure. So like maybe the push-ups aren't super realistic, but I don't know if they've worked a lot on the elasticity. Quietly become one of the most studied predictors of how long and how well a person is going to live.

And it still looks it still looks uh static to me. How? But that could I guess that you you might argue that that's the fix scrubs. It costs you nothing. It take Yeah. Like Yeah, you can still say. So like if I'm watching this part of the scrubs, it's just like it's jiggling like a jello. It's It costs you nothing.

Yeah. Like scrubs don't do that in real life. They're not jello- or figs scrubs jello I'm not sure or absence of other people and the channel scores a neutral yellow circle so there are no other people so it's very simple but there could be other reasons for that if it's a real person you might be really busy I very high pitch there like an emergency doctor and it is a young channel it has only been three months so let's give it some more time but I would certainly love to see this channel on a podcast or something or even like over a Zoom call.

Anyways, because Yeah, that's a good one. Like podcast channel and because it could be a really busy person. I'm going to give it a neutral yellow circle. Yeah. Yeah. So that is like one one great way that most people probably can use to check whether a channel is real. But you have to be careful about it because you can basically synthesize a podcast if you just have two talking heads and you you make them away and then you you you pause and you resume or you tell the AI generation model to wait five seconds before saying the next line kind of thing.

So ideally in a podcast it would be something more like a diary of a CEO where you can see both people in the same time and also you trust Diary of a CEO to um it's like a it's a popular podcast so you trust them to not you know corroborate an AI tell or an AI story. Okay, we're good. We're good on time here.

All right, let's start the next minute. But let's see what happens. The fifth category is whether the output is supposed to model a real person. This channel should receive a green check mark for complicated, but there is a catch. So although the character can be traced to a real person, I don't think the voice is the same.

Let me show you what I mean. So the real person was on a BBC show. This is the real person saying every single day in his beautiful British accent. Every single day we don't. Okay. People. And then this is the YouTube channel saying every single day. So the exact same words. Every single day. Yeah. Um, I still think the voice is different.

I don't know how like how how I could describe what it sounds like to me, but I would say that like this is a shorter voice like that a voice that comes with a smaller lung and [clears throat] a shorter stature. Whereas like the the one that the actual doctor has is super super deep massive lungs or big lungs. Sorry, I don't mean to like offend the real doctor.

Um so because of that it it sounded different to me. And I would say that this is actually more consistent with like his his stature. Supposedly, the real doctor is like, I don't know, six foot three or something, six foot two or six foot four. So, that is the voice that you'd expect from him. I wonder if the real person Oh, wait.

Sorry. I still got to go through like 30 more seconds. We got people. And then this is the YouTube channel saying every single day. So, the exact same words. Every single day. Every single day. The every single day. Every single day. Every single day. Every single day. And I turned Every single day your body.

Every single day. Every single day. Every single day. Every single day. Every single. This still makes me crack. Every single day. Oh, so funny. That's probably a bit much, right? Let's just hear the real person. Oh, yeah. I don't I can't believe they didn't check the scripts for that. Um, just like for repetitive words.

All you have to do is just like parse the phrases and look at a histogram. Um, but I guess they didn't think any [clears throat] anybody any any technical would be analyzing the videos. Yeah, I wonder if the real person owns that channel or if his image is used by somebody else. Uh, I believe that it's being used by somebody else.

I don't think the real doctor is involved just based on like the doctor's oath really and like potentially just like letting AI control your voice, your medical authority without like review. I feel like that's against the doctor's oath. Uh so that's why I don't think that the real doctor is involved.

But you know I could be I could be naive here. Yeah. Oh dear. He isn't repetitive at all. Is he? Yeah. Well, that that channel or like that script too. There are I think there are a bunch of phrases that are just like repeated over and over again. It's like yeah, I will do that unless you like tell it not to.

And if you tell it not to, then you lose you lose u you lose other parts about the the prompt. So, if you're telling it to like have a certain character, that might not show up in the script. Yeah, I don't want to get too technical on that one. Okay, let's go for the next minute. This is going pretty well.

Every single day we get people. All right, now let's go back to the YouTube channel. Every single day. Every single day. Every single day. Every single day. Every single day. Every single day. Every single day. Every single day. Every single day. Well, every single day. Every single day. Every single day.

I don't know why they get to, but like every single day. Every single day. So, that was a lot of those sound the same, but the real doesn't sound the same to me. So, I have to give it a red simple X. The sixth and last category is emotional simplicity or and one thing that I was doing here with the real person model is I was trying to like I I'll just be honest I was trying to make them say more personal stories.

Uh and when you do that you force them to increase their variance which means they'll make more errors. Um, yeah, complexity. The channel scores a cold hard red X. Not because it's a doctor making a lot of people think that they might have deadly diseases that they did not know about. Not because of little things like a British person describing their weight in pounds instead of kilograms.

No, that was okay. A lot of people commented on that one. Did I start the timer? Okay, I didn't start the timer. So, I have to do five push-ups. Um let's do them now. So that was this is one two three four five. There you go. I'm just doing the uh the ones that are in this video. If you think that's weird, I don't think those are real bishops.

Uh okay. So like I am being now with the petty thing. I was fetty here with the uh with the weights versus or the pounds versus kilograms or I was I was wrong about that one because it turns out that there's a lot of regional variability with uh pounds versus kilograms and then there's stones as well in the UK.

So, that one I wouldn't consider evidence, but I didn't even put it as a thing um on the list that they did not know about. Not because of little things like a British person describing their weight in pounds instead of kilograms. No, there are some very subtle inconsistencies. So, let me show you some of the early videos on Facebook.

Uh but before that I must tell you that this might have some emotionally triggering stuff. So [snorts] if you do have any thoughts of self harm, please seek help. So I want you to notice the smiles over here after talking about mental health. That's what I'll probably uh skip this part just for the stream.

But yeah, has anyone tried to contact the real doctor to at least make him aware of the channel? I think there has been I saw a comment on my channel where a a person claims haven't validated it and so it's just text that they went [snorts] recently to the reception and uh basically tried to alert the doctor to the channel. that we haven't heard since.

And if that operator is watching this stream right now, they're probably going to mention that on the YouTube account. Now, that probably skip actually found to be a good point with with him. Yeah, skipped it yet. But yeah, I I do think that is a good point, right? Uh it's just like the emotional the emotion is not consistent.

Um, and here the thing is like if you try to generate AI yourself, you'll probably find that a lot of platforms at the end have like they'll they'll say they're lying like, "Hi, nice to meet you and then they'll smack a weird smile or something at the end like their their facial expression would change." And that's that's what I noticed here is mostly a smile in the eyes. So a lot of people don't like miss it.

The quality control there miss it. That's what I believe. But yeah, I will I will have to skip that one just because I don't know who is watching the stream and it might be triggering. Okay. Come. Let's restart the timer. Look, I get it. If you're Oh, okay. nervous laugher. You might laugh at inappropriate times, but this is supposedly an emergency doctor who's had a decade of experience.

Yeah, that would probably be the first thing that they train out of him if he was a nervous laugher. So, I would say Yeah, exactly. Like you wouldn't want your emergency doctor accidentally laughing at because they're nervous and real real people's pain. That would be that would be bad, unprofessional.

It's you know there are nervous laughers exist. A lot of people do nervous laugh. Um, I think I was a nervous laugher when I was a kid. Um, but it's easy to train that out. So, or not easy, but it's possible for me. It was easy that this was an attempt at complexity that failed, but now it might be able to do that.

So, we're done with simplicity there. And we're going to go [snorts] through the next category. not simplicity that is characteristic of AI factories versus humans. And so the thing about AI factories is that they don't get tired. Yeah, I think that's that's still true. Humans get tired. I'm pretty tired.

Uh factories, no, they don't. they just you know just uh put more energy in more more electricity in and you're good or not not material I should say not not physical or material factories obviously they need a supply chain for things but for data centers they just need they need power or electricity and to cool that off you could say there there are technically more things but in terms of like in out production. As long as you keep giving them power, they can run all night.

You also have to give them supplies for complicated factories because I'm glad I said night there because uh it's cooler at night. So, you could actually probably run them all night as opposed to all day. That was a fun accident. Um [snorts] so, this man actually started at 12:47, I believe. So, I'm going to stop that timer and then address this last comment.

He generates far too much. Oh, yeah. His eyes are always electric blue, which I found very odd or found find very odd. Are they still like that? Yeah, they're still electric blue. Uh, people are noticing that. I wonder if they're going to try to change the eye color. That would be that would be nuts.

Uh, cuz then people like, why did the eye color change? Um, they're too deep in on that one. But, you know, I'm maybe some people are uh do have that eye collar. Um I I wouldn't say it's evidence, but it is it is uh it does stick out. He generates far too much or like the channel generates far too much content for it to be done by one one person.

Yeah. Like one video every day. It's very it's very gutty. Um and you know, claiming no editor as well. we'd have to cut so many. It's very gutty on their part because otherwise the AI flags YouTube would have gone off if you did u more and like they I think they do have other channels that do less or at a lower frequency.

Um but yeah, that should be in other videos. Um they're I think maybe yeah I'm not Yeah, I don't know why they did that. This might be like a test channel for them in terms of the frequency. They're testing a frequency here, but we'll see. It's also possible to keep up that that uh uh um how am I how do I say this?

Like it's it's kind of like charming. Like somebody said that, you know, what an idea of a doctor just like recording a batch of videos at the week and sending it to everybody. Uh just just like the idea I think that heroism is is appealing to a lot of the viewers. Okay. Uh his blue tunic must be Mankey.

I don't know what that means. What is Mankey? Is that I'm going to look that up on my other screen. Informal British slang adject to mean dirty, filthy, or unpleasant. Okay. Uh what's the tunic then? I think I [snorts] should know tunic. Uh oh. Like his his uniform will get dirty. Yeah. Yeah. Well, the clothes.

I don't think it's supposed to be uniform. I think it's supposed to be a costume. But yeah, thank you for explaining that Ke I did I did look it up as well. You were right. Okay, let's resume this one with a 4m minute thing. We're almost done with the video here. pretty proud of myself. This is software.

You don't need much human labor. Probably mostly in the quality control step. That's what makes AI factories really amazing. All right, so let's score the YouTube channel in terms of three categories for fatigue or tiredness. The first category is output. The amount of output that this channel produces is it's a lot, but I wouldn't say it's impossible for a human.

I would say it's super human but it's possible. Yeah, I still stand by that to be honest. Like it's super human but it's possible. There are people in historically that have been super productive uh in in in their ways on YouTube. I think there was like one other person, but like he had a whole team.

Uh, and it was like vlog style vlogging. So, he would just basically record himself on the camera and they would make the videos. Um, but like what this this type of content scripted medical hasn't I don't think that's been on record for 13 weeks in a row. It's less possible but it's still possible and while working.

Yeah. So now it's I think it's at technically six months uh consecutive days. So I I wasn't able to discourage them or convince them to reduce the frequency here. [snorts] Oh, I love that my title I didn't even know that it says AI like right there tag. Like I knew I tagged it but I didn't know they added this thing here.

That's cool. I have been getting a lot of comments about that. Um, what was I saying? Yeah, that didn't discourage them from or encourage them to reduce the frequency of the videos 30 hours a week as like there has been no like fake burnout thingy on the channel yet. an emergency doctor. Still possible, but less possible while supposedly also having a wife and a newborn.

I don't know. The only reason I'm Yeah. To me that was like I if I did that if I like had that work life balance where like I'm working and I'm just like recording every day. my like I love my wife but she'd be a little upset and so like imagine that like I would be leaving her with our newborn. I would just get it though.

Like I'm not going to give it a red X is because AI factories can actually produce a lot more than that. Okay, finish it up for a minute. Uh so Nelly Free says Dr. Dr. Dre is a dermatologist that is on YouTube for more than eight years. Um, let's look let's look at that one. Uh, Dr. Dre here. Dr. Dre, is that the one?

Vlogs only on weekends gets a video every day. So 5,000 videos. So that's pretty much every day. Let's see how far back it goes. This is pretty interesting. Okay. Over one year ago. So that's pretty that's pretty good. So when I'm checking if something is AI, the first thing I check is like, well, when did the videos start?

Did they start after diffusion models were published? Um, but then I look within the videos to see the distribution of variance. I think we're going to get into the two-year range. I'm trying to look for 2023, but this is a ton of videos. We can probably just stop there. Vlogs only on weekends. So, does it tell you which day it was published?

That one looks like a vlog. Um, this one looks like a a vlog as well. Skunk care need to discuss. This see this seems like a real real person as well. Like they're moving their camera around. This is pretty complicated video. Um the hands going in and out of the frame. How long are these videos? 12 minutes.

Wow. This is amazing. Okay. So, oh yeah, there there's the vlog. So, okay. So, this is a weekend and then this is like a regular week. This this looks like definitely looks like a real human to me. So, this is like super productive. very amazing. Um, yeah, this person definitely is an outlier. Yeah. Yeah.

Yeah. Nelly. Um, def. Yeah. No, I could I just like I I wanted to like run my tests on it. Um, definitely. Yeah, definitely real. looks like real. But so she do you know uh if the if the scripts are the same like what can you tell us about the uh the scripts I guess? Do they look scripted? Do they look uh they sound scripted or or is there a lot of buts and ums?

Are they pretty focused? This is very interesting. I will I will cover I'll probably make a video about this because this is just so amazing. Okay, I'm going to subscribe. Um, but we'll get to that and then we'll compare. This is really cool. Thank you for sharing. I don't know why that sounded sarcastic, but thank you for sharing.

Okay. Uh, okay. Let's go. 4 minutes. Next minute. That said, it could be a way to avoid YouTube's AI detectors. Anyways, I'm going to give it a neutral yellow circle nonetheless. All right, so the next category in fatigue or tiredness is team. The channel claims that there is no editor because it would eat up all the assense money.

If I got a video editor for this channel, it would honestly cost me hundreds of pounds per video. [clears throat] Hundreds of pounds per video. Uh, let's do the math on that one. 365 times 100 I don't know. I don't know if it's hundreds of pounds per video is realistic. Uh, especially for just like a talking head video.

I'll say 100 pounds. Um, but you can correct me if I'm wrong. If you I my editor doesn't charge that much. Uh, and then times 365 that would be 36,500. It would it wouldn't be close to how much this channel is making like this just based if you like according to Vid IQ estimates. So, if I go to a random channel or any channel, I guess it's not a random channel.

Um, this channel, Dr. Dre, is making 21,000 USD estimated to 21,000 for the seven for the 8 million views per month is what they're estimating. And then this channel, um, they estimate, okay, 1 million view, 31. 2. I'm not sure why it's a bit higher per per view. That's very interesting. Maybe it's like the demographic or their intention.

These things aren't super reliable, but are like the these estimates aren't like they're not the truth. They're just estimates. But this channel is supposed to make um million by end of this year. Like if it's if you look at how fast is growing. Um, so 36,500 that you could pay to an editor uh wouldn't would indeed much uh relatively.

I don't know why I'm spending so much time on that claim. He's changed his mind as he's advertising for one. Oh yeah, the the editor uh at that stage that would completely wipe out my AdSense revenue and then some. I think that is plausible because the videos aren't all that much edited and I do feel like it's true actually because if other people were involved in this kind of thing and if it wasn't a factory they would be easily able to destroy the channel uh by sharing receipts.

So because that's too risky I'm going to give it a green complicated check mark. The third and last category for tiredness is supply chain. If this was a factory, it would need AI mod. Okay, we got 15 seconds to discuss that one or I have to do push-ups. Um, the there with the team thing. I uh I was I'll be honest, I was kind of trying to like uh see if the channel would react to that cuz like whenever you have a team and if you're doing something that's like, you know, ethically questionable, there's always going to be somebody that's not really on board.

And by saying uh by saying this it's like well who is not on board from that channel uh and will they ever come forward models or data centers. I think it's unlikely that a major provider would provide those. Oh yeah, let's restart that one. I made it with just 11 seconds. By the way, I was looking at the tab.

This is really cool. By the way, Dr. Dry AI models, let's just start this again or data centers for a factory that produces non-genuine or nonauthentic content, especially for producing content about a doctor giving out medical advice. It would compromise their legal standing. So that's it. It has only been three months.

So it's possible that the data has just not been audited yet. So that's why I give it a yellow. So that that's that is true. I think like data centers do actually audit they do actually audit what is being uh produced. So it is a risk that they're taking if they are using like AWS or Google or Azure.

Um, especially if it's like supposed to be health data, it might I guess they could try to pay extra for privacy, but the data centers are technically responsible or partially responsible for that circle. So like to say that it is possible that the channel experiences fatigue like a human, but for my simplicity threshold. I personally scored this channel as probably AI.

I think you got to make up your own mind based on your own threshold. And then once you've done that, even if you decide that it is AI, you could continue to watch it as long as Yeah. Uh so I scored that. Uh I said my opinion is that this channel is probably AI. I'm I'm still uh with that stance especially now that like it's been three more months and like the variance has increased in a very predictable way or the quality has increased in a very predictable way.

Uh but at the same time the uh like the the volume of videos or the frequency that they're being released has just like been uh super impressive or super high. You know that the advice might not be medically reviewed by a medical person. But the unfortunate reality is most people won't know. Most people won't notice.

They won't see this video. The channel will do everything. This is true. most people won't notice and won't see his video. And for like three months, this video sat uh or two months, it sat pretty much unwatched in its power to avoid letting this video get attention. They might even replace. So that wasn't um I don't know if they tried to do anything, but like they could have they could have just like confused algorithm with bots that have nothing to do with the audience and then nobody would have watched it.

It's it would be a risky move, but it it could have like completely killed the video for this video in the host of it getting shut down instead of actually having a response. And on top of that, the threshold. All right, finished that move or that minute. Got like two more minutes between human and AI is going to change with time.

It's going to be Oh, wait. I got to start the timer. Okay. harder to recognize whether it's humans or AI. For example, in the current videos, it's pretty hard to tell. In the earlier videos, [clears throat] to me, it was still visible even within a video. Right now, it's still visible for me because of the overall pattern.

Okay, so that said, you probably noticed that I can also make this stuff. My friend said, "Why not just do the same thing and get a medical reviewer if you feel bad about it?" It's true. Like I think a lot of people uh like I don't know if it's all like one operator that is controlling all these channels, but there are a lot of channels out there and I don't know if it's all like the same operator or just people just running to this like a gold rush.

Honestly, if you think about it like from like a economical perspective, this is this would unfortunately be probably one of the better use cases or more profitable ones of deep faking deeping a doctor um deeping yourself people will just get that uncanny feeling if you're already creator and they'll be able to tell. Um, so I don't think that would be as profitable. But then, you know, declare that it is an AI generated and it is reviewed by a medical staff member, right?

You could just hire a doctor. I could, but for me it feels Yeah. So, I'm I mentioned that you could just hire a doctor and definitely like I I don't think you could find one on on Upwork or something or Fiverr, but you could find somebody in person or in real life and have them just like review the basically just the facts uh that the channel is checking or disseminating.

And I don't think that that they did that because I found a lot of hallucinations. Uh, I have another video about that one. It's like cheating other hardworking people. Like I'd rather put all my soul, my face, my efforts, fail, and then grow out of it. Cringy. I think that's the best way to actually make something meaningful.

And I strongly believe in some people did drop off at this point. They're like connection. And I also believe that you should authentically empower I do to age healthily. That's like that's the whole reason why uh yeah impro age healthfully. I I did my research. Okay. All I all I can say about that is like it was pretty tough to watch uh just like that that last minute cuz I was trying to be vulnerable.

I guess it kind of came off not super vulnerable but whatever. [snorts] on stroke and Alzheimer's disease that affects people more and more as they age. I really care about soren Alzheimer's or dementia in general because I'm afraid of like being of like losing control of my mind really and that tends to happen with dementia though not with healthy aging arguably and so be honest to yourself that's the hardest thing to do at least for me what your goals are write down your goals write down the actions that you took towards those goals and the results and do that.

This is the this is the best thing you could do for your health. I think uh ever since I actually started doing that, my health has like been up tremendously. I have a different channel that's just like about my fitness goals. Um it's hidden on my description or channel description if you if you look for it.

Uh but it's helped me a lot because like I used to go to the gym for probably like 14 years now and until I actually had goals that were like okay let's actually try to achieve this. I wasn't making progress. Uh and I suspected that works so that was in the gym. I also like I'm doing that for a half marathon.

I suspect it works for all health goals. So over and over again until you're actually achieving. But it is really hard. Like you have to kind of be mean to yourself. I mean I don't know. I kind of feel like I am being mean to myself sometimes about that. But like that is kind of what I need to achieve my goals.

So else you want. All right. One last thing. If you want to connect with me, uh, leave a comment below or you could sign up for my newspaper. I'm going to also leave it a link in the description. So, I said uh I said newspaper here and I just like I didn't know what to say. Uh it's like an email list or whatever.

Um I could have just said that but I was like I don't want to do that take again and it's a lot of work. Anyways, there we go. We got through that. Uh thank you for hanging around and uh that's all we're doing we're doing today. Thank you for coming and uh I hope um I hope to see you again soon. So,