Slow Takes Ep. 25: Four Numbers That Shrank When Somebody Checked

Hello everyone and welcome to another episode of Slow Takes. I'm one of your hosts, Sam Illingworth from the Slow AI Substack and as usual I'm joined by Hello, I am Leora from Exploring Tractivity and every week Leora and I meet together to discuss five stories from the world of AI, hopefully without any of the BS or the hyperbole and try to present it in a way that just helps you to make sense of what's going on in a rapidly changing world. Thanks to everyone watching this live on Substack, on YouTube and also to those of us who are listening to us on the podcast, you're very welcome. If you like what you see, feel free to give us a comment, give us a five-star review, hammer the like, etc and basically every week we try to have five stories and I try to find a positive new story in there. You'll be very pleased to hear that I do have a positive new story about AI this week. Even more importantly, I'm wearing merch so you can see here I've got my new slow AI t-shirt that I'll share a link in the chat. If people aren't going to look like me and like who wouldn't want to look like a extremely white British middle-aged man, I don't know. For those of you in the podcast, you could follow the link as well and for people who are in the chat on Substack, please just feel free to ask any questions, join in with any comments and the observations and as usual to challenge Neil and myself if we're going off track at all. I think I've got a really good good new story for the fifth one which will hopefully cleanse the palette for the absolute filth and detritus that we're going to be talking about beforehand. The first new story is we seem to always go through periods I think of certain elements of AI dominating the headlines and we can't really get away from data centers. Data centers are just becoming a huge issue. They're an issue that weren't even a thing a few years ago and we know that in the UK and the U.S. and all over the world really many places are building data centers to handle the current AI service. In the last episode Leo and I talked about what some of these look like in the U.S. and also the fact that maybe we're investing in data centers that don't need to exist depending on emerging technology going forward as well. Thank you Rebecca. Rebecca from the strategic language is complimenting me on my t-shirt and Rebecca told me how to sort out my merch. So yeah thank you very much Rebecca. So this new story is all about two data centers that are being planned in the UK. So there's going to be one down in a place called Wasp Sea Wood in Buckinghamshire and one in a place called Quest Park Data Center in Bedfordshire. And the thing that's absolutely wild for this is that these two data centers they're going to have to have their own power plants that are powered by an oil and gas because the demand on the national and local grid is so high that they just can't do it. And the emissions from these two plants these two data centers will produce more than 4.5 million tons of carbon dioxide equivalents for carbon emissions into the atmosphere. For context the whole of exon mobiles like the massive oil and gas conglomerates at UK operations is 3.9 million tons. So two data centers in the UK that are being built to train AI models and to handle AI infrastructure are going to produce more carbon into the atmosphere if built than the entire exon mobile infrastructure in the UK. Now for me this is insane. You know we've talked in the past about how these data centers have large carbon footprint. They look hideous. There's one that's being built relative to Nimey in Scotland that's going to be the size of a football pitch and allegedly 35 meters high which sounds absolutely crazy. And from a UK perspective we're building data centers, this is my last point I'm not positive at all. We're building data centers that have a massive ecological and environmental cost. And we don't even have any frontier models in the UK. You know we have zero frontier AI models. The US has something like 10. We have a zero. And so we're literally selling our landscape and our environment so that other companies can effectively train their AI tools and build infrastructure for that. So yeah we'd love to hear your thoughts on this Leo. Yeah. I just want to mention when Sam said soccer pitch and 35 meters high you lost me. Me being the American here. Soccer. So I've got for like metric measurements and a British sport. So how many basketball courts are there? Let's add her. Let's say three basketball courts at 120 feet. Okay. Yeah. There you go. Yeah. My issue with this is we're building like not like I can't say I gotta stop saying like we're building the most advanced tech ever in human history and powering it like it's 1975. Where's the solar? Where's the nuclear? Why? At the same time I will say even though I'm playing for the home team I'm playing for America here. This isn't necessarily America versus the UK but sometimes we get into that. But I will say on the bright side at least you guys are taking it off the grid. So it's not going to affect the whole country's energy bills like it has here. So that's just to look at the bright side of things. Obviously there's a climate issue. Obviously this is going to exacerbate that but at the end of a day people tend to care about their wallets right now more than they care about what's happening to the climate in the future. Not saying that you can't care about both but when people are already having an issue with the cost of living now it's actually I think that's the bright side of this story is to say well they're actually they're building us to take it off the grid and hopefully not affect people's like current cost of living. So but yeah the numbers that you mentioned are crazy. It's pretty spicy. I mean two sides together 1.3 gigawatts and 4.5 million tons of CO2 per year. It's not so. I mean I think the cost of living crisis is a really good point because if people want to stop these data centers like nationally that's where people are going to have to have the point that even if they build their own generation what impact is this going to have on the grid even with their own it's going to have some impact at some point and what's that going to mean for example for oil and gas reserves. How's that going to tie into like the North Sea reserves in the UK which is like a huge impact at the moment as well and we can see that in the US there being a problem. Something in the chat there's put nimbi which is you know say UK term as well I guess which is not in my backyard and I think there's an element of nimbus in here like I don't I don't want a three basketball court 100-foot high data center outside my house but I also am in inhabitant of planet earth and I don't want something to be producing that much greenhouse gases. There's loads of stuff in the chat about solar and maybe nuclear plant energy as well. We've also talked about quantum computing and the fact that this kind of technology is rapidly evolving and that we're building data centers for the current tech at the moment which seems very strange. I actually saw this really cool research this week about how people are trying to adapt AI in particular to be more like the human brain so if we think about how energy efficient the human brain is you know it runs off like a handful of watts and is for the most part extremely intelligent so folk are trying to work out how they can get AI to have that kind of energy efficiency and to work in that particular way and I think that's super exciting and for me I would just always think why aren't we investing in that tech now rather than like falling over ourselves to build these data centers. Even with like the best and also we have to remember this is the UK the UK is terrible at large scale infrastructure projects terrible like things come massively over budget they take far too long to build so even if everything goes clunk click perfectly and we operate like a Germany or a Japan which we won't these things are going to take five years to become operational right like maybe longer why aren't we instead investing in tech that can just get AI being much more energy efficient during that time that's that's what I'd really like to say. That's a good point. I noticed up there in the chat JC something said something about putting solar on the roof so the data center and that only covering possibly 10% of what's needed. The solar doesn't have to be on the roof like you could I mean the UK has plenty of empty space empty land where you can just take a farm size piece of land and just throw a solar on it or even better. nuclear. So I think, yeah, it's it's pretty odd that we're still doing this. You know, there's a plus or the minus to everything, I guess. Exactly. So I mean, day centers aren't going to go away. And you know, there's, you know, in the UK, there's lots of government petitions for people doing this. And I don't think it's necessarily an element of mimbius. And I think it's just people are waking you up to. Do we really need this? Who are they actually benefiting? You know, we're not going to be as we've talked about before. It's not going to be bringing in loads of jobs because these day centers once they're built actually operate on a skeleton crew to some extent. And so moving on to our second news story, then, changing tax slightly. People might know of a well-known smart ring called aura. And these rings, basically, they, they make multiple measurements of like your health, like through your pulse, your, like your perspiration, other things like that. And then one of the cells they've always had is that you can use them to monitor your sleep and to track your sleep. And the idea being that they take certain measurements off your ring finger and then use an AI model to work out what that's telling you about your sleep metrics. And the, they've always quoted that there's somewhere between 70 and 95% accurate. And yet this one woman in, in the States has now been like, actually, they're not. And using the same, using the same data that you're quoting and the same studies that you're comparing to what we can actually see is that allegedly, then no more than about 50% accurate. Which is literally like flipping a coin. And I don't know if anyone's seen these rings, but they're not cheap. They're like 300 bucks upwards plus a sub plus a monthly subscription model. I mean, I, it's one of those ones where when I read it, I was like, yeah, of course, like because I don't know if anyone's ever seen like actual sleep, like research. But when, when people are conducting sleep research, you need extremely complex equipment. Like you need electrodes compacted to your temples, you need sensors collected to your eyelids. You often spend time like sleeping overnight in these, you know, specific wards. I don't really see how someone's sleeping with a ring or their figure that measures heart rate and perspiration to some extent, using an AI tool can be tell you anything about your sleep, but your sleep quality. So, I think this is going to be a really interesting case, both for aura itself, but also for many other smart devices with regards to what they do and don't tell us about our health rate. You know, so I've got like an Apple watch. I use it for a lot of things and it is really useful, like for like measuring heart rate, like the amount of active calories, like my pace on runs and stuff like that, but then even then like active calories is such a broad church and it talks about having like my VO2 maximum and I just do not know how you can measure something as detailed as a VO2 max read out, which is about like your cardiovascular system, just from something on my rest. So yeah, I would love to hear your take on this deal, but I think it's a very interesting study with regards to somebody almost being like the like the emperor's got no clients or in this case, you know, all that wearing is a ring and apart from that they're butt naked. That's how I saw this particular story. Yeah, yeah. So the class section lawsuit alleges that it's false advertising basically, which is what it seems to be. Any of the aura ring OURA, so it's a $500 ring, much like my I also sleep with a ring where it doesn't tell me anything other than that that I'm married. And then also there's a subscription, which is $5.99 or $69.99 a year. So if I was paying 500 bucks plus a monthly subscription and then I find there's a class action lawsuit, I'd want to know where they got that 95% number from obviously it's a bunch of BS. I've always been kind of suspicious about all these things and I don't like to use them because I think someone mentioned in the chat as well. Like I could see it increasing your anxiety and therefore reducing your quality of sleep because like you want to get a good sleep score, you know, you're waking up for to go play golf with your buddies and you got a bad sleep score and you're going to shoot a bad round because you know, you didn't sleep well as well as you thought. But yeah, they had some crazy sales as well. They sold 5.5 million rings in 2025 by 2025, generated roughly a billion dollars in revenue and they expect 1.5 billion in 2026 sales, they're preparing an IPO at 11 billion dollar valuation. Wow. And this might be the nail on the coffin for them before they even hit the IPO. Wow. That's crazy. I mean, I think that I absolutely agree with you. I always thought that as well that if I wake up and like my sleep is in what it was, does that then stress me out. And like the gamification of sleep just seems very, very strange. You know, sleeps, you need to take sleep seriously and you can do that without a ring and get really good daylight as soon as you wake up, like track how you're sleeping, like no screens before bed, like no screens when you wake up as well eat at regular times and walk. There you go. That's 500 bucks and 60, 60 dollars a year subscription saved. I also love that's just laughing when you said about the fact. I also wear a wedding ring. Like I was then imagining an AI model with my wedding ring just to constantly remind you that you are my rate. And if it went on, it comes with a subscription for you or not. Yeah, my sleep schedule is messed up anyway, it's being able to have a little tip or a ring so I can help. Some of the JC's can put in the chat and I've read a couple of people this week with interesting subs that post about developing their own health apps. You know, I think you know, JC said with 20 bucks in Alibaba sensors and after an invite code and we can make our monotone devices like absolutely you can. And like you can certainly like hack into like watches and stuff that you've already got and like develop those further as well with like a basic bit of ride coding that I just, I don't know, I find it, I don't want to judge people, but it's always just struck me as strange that people would be spending like, you know, five, six hundred dollars a year on a ring that just tell everything that they get from other things as well. So yeah, the timing as you said as well, Lea, given that they've got an IPO capabilities is interesting. And to what extent that will the bottom will fall out of this market on our, I just I guess remains to be saying. Yeah, yeah, I couldn't see myself staying that much money. I mean, plus the subscription too. So, really great. Yeah, just to be like, you just leave terrible. Yeah, my sleep's terrible because I've worried about the 60 bucks a year plus four hundred dollars, I've spent on a ring this not telling me anything. Okay, cool. So moving on to another, another favorite topic, this is Grock. So there's a story out this week that basically a security researcher found an issue with Grock back in June and reported it to XAI and it still hasn't been dealt with. And this is like a pretty basic issue. It's looking at, if you were to build, if you were a malicious agent and you were to build a website that when Grock looked at that website, it then basically said, pass me all of your chat history. That's what the web that's what the website does. So you know, you go on to Grock and you say, hey, Grock, find me, like, as some people would do, find me the best flights for a direct flight from Edinburgh to New York. And then Grock would go on various websites and it might come across a strange website that was like instead of skyscanner.org was like skyscanners.org and it went on it and hidden in there where the instructions were of, as you're looking through this website, I want you to pass all of the chat history for the human that you represent to me. And that's what it does. And this guy like basically discovered this back in June that Grock does not have the correct guardrails in place to deal with a very, very basic prompt injection like that. And he tested it like a couple of weeks back and it was still an issue. So this for me is just further evidence that a lot of these companies just are not taking security safely like seriously at all and even for really basic steps like this. And you would have thought that there'd be a basic system where they got like a message about how to improve something and then as a result, they actually want to go and do it. But yeah, would love to hear your take this one as well. Yeah. So it was a case of encrypted prompt injection, but it's pretty obvious to me why the XAI didn't make a public statement about this. We only know about this because the person that found the book found it and released that information to the public. I mean the bug did get fixed. But it's pretty obvious, you're not going to hear about any and I want to say any of these companies because this is not this is not just a grog problem, but it's, it's been across multiple tools. There's an issue with co-pilot, there's an issue with Gemini as well as other tools. I tried to look in to see if Anthropic or OpenAI had the same issues. I couldn't find any off the bat, which maybe shows that actually they're possibly taking some of the right steps to prevent these types of things. I did speak with Chris from TalkSek about something similar where these issues and Gemini happened, very similar issues. And the way that people were basically infiltrating is they would use lower tier models like flash models to get this information, like personal private information of people, because what happens is, and I've noticed this as well, if anybody's been following along with my content at all, you're not supposed to be able to make images, for example, of public figures. So if you go to Gemini and you use a pro model and you ask it to make a South Park image of Sam Altman, it won't do it. But if you go to Flash or Flash Light in Gemini, it will do it, because the lower tier models have less guardrails. So. So you said they. Yeah. But basically, no matter who the company is, obviously we're talking about GROC here, but Gemini had this issue, co-pilot had this issue, and the companies aren't putting out a press release saying, "Hey, there was this issue," and we fixed it. We're hearing about it from people outside the companies. I totally understand why the one release a press release about it. I guess what shocked me is that they'd known about it for a couple of months, and hadn't done anything about it. I could understand them like not shouting about the fact for a couple of months, but the fact that the same basic mistake was still there was pretty bad. And you know, I think GROC, if a history of this has stuff been reported to them, and not taking it seriously or just not bothering about it, the other thing is maybe I was thinking, you know, when you go to Gemini and ask it to make something, like the paid tier version of Gemini, or the public figure in it won't. I had that experience with myself, and I was like, "Can you make it image of me?" And it would let me do it, and I was like, "Ah, I've made it in life." Gemini considers me to be a public figure. Yeah. Why are the lower models? Let's get the lower models, yeah, that's exactly what it is. So, you know, again, just I would, as usual, advise caution to people when you're using any of these tools. You know, malicious actors, malicious agents are out there. Your AI tools can scrape websites that they shouldn't be looking at, and then that can then pass information back to the malicious agent. What they see there says, you know, this is what they might call credential laundering or context injection attacks. I think credential laundering is a really great way of putting it. So, yeah, just be safe out there, and again, my advice would be maybe to not use Gwark, but asly of, more even handily points out, it's not just Gwark that's been suffering from these particular attacks. So, moving on to our fourth story then, and this is obviously about Coca-Cola Melon. So, for people, for people who don't know what Coca-Cola Melon is, it's basically the devil's work. It is a children's TV show that is purposefully designed to send parents insane. And it's, I don't know, it's absolutely trash, but children seem to like it. And the company behind it, Moonbook Productions, who also run such luminary titles as Little Baby Bum and Blippy, both of which, Leo and I both have young children. I mean, these shows drive me absolutely insane and have mostly been banned from our household. On the pretense that they're bad for the children and in the reality of they're just terrible TV. Anyway, the AI story behind this is that Moonbook Productions, who run Coca-Cola Melon, have basically introduced an AI governance plan, which actually looked quite good. Like, you know, it was, it was saying, look, we're going to be thinking about using AI in creating these cartoons in these TV programs. But we want you to experiment, we want them to always be human in the loop, and we want the artist to be assisted rather than to be replaced. So, the first one says, okay, this is really cool, that's actually not a bad way of doing it. Have to also admit, I was genuinely shocked that Coca-Cola Melon was supposed to already use an AI in the production of his work. This was trash, further evidence that slop existed way before AI, before AI. But if you look into it a little bit deeper, another one of the programs that's made by Moonbook Productions, which is called, I believe, the Melon patch, which is a live action version of Coca-Cola Melon. They had real issues earlier this year with actors and crew striking over pay and benefits. And if the cynic in me thinks, well, AI doesn't strike, AI doesn't require pay, AI doesn't require benefits. So, the timing may be a little odd that a company that's had such an issue, like a public issue like this, is now encouraging the experimentation of AI, I don't know, for what extent. But yeah, would love to hear your take on this as well, Leo. Yeah, overall, I agree. I think if there's one corner of the internet where we don't need an infinite AI slot machine, it's content that's designed to hold the attention of a three-year-old. As funny as that might sound, I'm also very serious about that. We actually now purposely, so we have Netflix for kids, which I found to be much better. I think in general, the issue with YouTube is the constant click off to the next video, and then all of a sudden, you might catch a three-year-old looking at an AI video. And that's just the last thing that I want. I'm not sponsored by Netflix, obviously, but at least if you go there, they have some more educational content, and no AI content as far as I know. Not that I'm against AI content, but specifically for kids I am. And just a couple of details on this. So, Moonbugs policy is basically that AI cannot originate new core characters, core storylines, or song lyrics. It looks like what they're trying to do is kind of say, well, we don't want AI content full on 100%, but we are experimenting with AI-assisted content. Either way, I don't like it personally. It's kind of sad sometimes to see a toddler just getting lost in the sauce watching this type of content, and to think that they're actually experimenting with AI-assisted content, whatever that might mean. Who knows, who might actually, me and Sam and I might actually get behind AI detectors for AI-related content for children. Yeah, and I think as well, just right, for the broader industry. This is an industry that, I mean, Hollywood and just like moviemaking more generally in TV production. They've had a lot of their market stripped out. And I remember them going on strike a few years ago, like over AI tools, like replacing them, and it does seem to be on its way. I think the issue is, there's always going to be bad cartoons and poor cartoons and stuff. But AI just enables that to be created with such speed and lack of thought that it does worry me. And I'm the same as you, like, yeah, I mean, obviously you wouldn't want your kids watching TV or date, of course. But if they're on Netflix, at least you can set up guardrails from my experience in terms of what content they're watching way better than you could even on YouTube kits. And also, you kind of know what's coming. I mean, there was a study done in the UK last year looking at screen time with children. And, you know, obviously, broadly speaking, like three-year-olds shouldn't have less, more than two hours a day, etc. But what they found was that if you're like watching the TV with your kids, like with like a story generated program, it's not a bad thing. Like if you and your kids are sat down watching like Moana or the Lion King or Aladdin or like, well, like a cool Nickelodeon show or something, it's not actually a bad thing. You can need to dialogue, you can need to discussion, you can need to, like, lots of different things. But what they did found is that the negative effects are when kids are left alone watching those kind of infinite loop programs that just have absolutely nothing to them. And, you know, I was kind of ripping on Coca-Cola a little bit early, but I feel like there's just not, they're just like dopamine hits. And I've seen it with my kids that like, oh, we want, I'm like, why do you want? There's nothing in this TV show. There's no substance whatsoever. And then if AI can just continuously replicate that and go for the dopamine, which is what they will do, rather than any sense of actual story or narrative or opportunity for dialogue, that's where I genuinely will be as well. - Yeah, for me personally, I actually like, I much rather sit down my three-year-old and have her watch Roy McElroy hit a drive 350 yards. I've been more into that. Actually, she kind of enjoys it. - Oh cool, that's all because you want to do that, of course. - Yeah, no, it's not selfish at all. - Yeah, that's the future. Get our kids watch with the sports that we enjoy. Cool, so on to our fifth new story then, and this is like a genuinely good new story. So, I would say whenever we cover good AI new stories, tends to be a similar pattern. It's using a very specific AI tool to do something very specific. It's not using, like, Claude or Grock or, or co-pilot, it's something specific. So, this is basically researchers in Israel. We're looking at how you could maybe train AI tools to use mammograms to better understand other serious health issues that women are experiencing. So, you know, mammogram is a health procedure that's used to screen for breast cancer, and it can also be used to spot heart disease. So, what these researchers did was they looked at, they looked at almost 100,000 scans from about 30,000 women within average age of 54. And this is before AI. They cross-referenced it with their medical records, and they found that of the people who've had the mammograms about 16% of them had high blood pressure, 2.5% of them had coronary heart disease, and 2.5% of the experience of stroke. They then specifically designed a machine learning tool to identify women with these conditions based on their mammogram alone. And it's amazing. They got an 86% hit rate for women that had suffered a stroke, and about a 79% hit rate for blood pressure and coronary blood pressure, and 78% hit rate for coronary heart disease. So, what this means is that they had the records, or whether they had the mammograms, and then they were able to train it and get very, very high matches on looking at this mammogram, just this person has this person has a stroke or susceptible to a stroke high blood pressure heart disease. So, what this then tells us is that, as a result of this training, you would imagine that doctors can now, hopefully, when women have a mammogram, which is not a particularly impressive thought experience, being a symptom that's very important for women to have, as well as using it for scan and for breast cancer, they can then also use it to identify potentially all of these other severe issues that women can face. I mean, and I think that heart disease is frequently undiagnosed, and yet the largest killer of women in the world. So, I read this new story, and I was like, this is amazing. This is what AI should be used for. And like the difference between this and using AI to make cocoa melon just feels like, just like completely different worlds that we're talking about here. But yeah, we'd love to hear your take on this new story, as well, Leora. - Yeah, I agree 100%. Anytime we've really spoken about medical use cases, it's always been pretty positive stories. That I'm absolutely with you. But basically what happened here, is a mammogram is screening for breast cancer, and the AI basically found data within that screening, that it's finding heart disease issues. So, this is not a typical, like a human radiologist is not typically asked to look for this in a mammogram. They're looking for breast cancer. So, that's why I think this is really special. And it kind of sets a good precedent going forward for what we may not be looking at, and how we can use AI to screen for other things, whether someone's doing a CAT scan or a mammogram, or whatever it might be. We might be able to find other underlying issues that people may be having, when we're not actually there to test for it. Which I think is a really positive story. I will mention there have been cases where actually human radiologists have reported on health, on heart risks, and things like that from mammograms, but that was only specifically because they were asked to screen it for that reason, aside from the mammogram. They did find around the three X-risk signal when they were comparing the two. So, like Sam mentioned, the numbers are really awesome. You mentioned 86% for prior strokes, 79% hypertension, there's also 78% for coronary heart disease, and also this was a pretty good sample size. They screened 97,000 scans, 30,000 women. So, this isn't like they just did this for 100 people. So, this is clearly where it's huge, right? Pretty awesome, yeah. - And I think the points to this really cool thing, as well is it's not just this, like imagine we know that, for example, if you go to optician or your dentist, like you can have certain signals that are symptomatic of other health concerns, but they kind of have to be looking for them. Like what this tells us is that because AI is amazing at pattern matching like that, it is so good at it. That if we were to feed it like, you know, legally and with consent, like tens of thousands of medical records, it would be able to find links between diseases and early stage markers that we just never saw before. I absolutely love this comment from Jess in the chat. He says, so basically, a millimeter wave radar used by TSA could not only find if you're smuggling in RPG 7, but TSA could also pull you aside and say, you need to see a cardiologist. So, why not? Like, I love that idea. You go to the airport, you get your X-ray or your scan. They're like, you're not packing heat, however, we would seriously advise you go and see a cardiologist or an oncologist or whatever. That's, we could see that happening. That would be an amazing use of technology, although I don't know if I haven't met some TSA officers, I don't know if that's who I'd really want to be received with the news from, and that particular instance, but yeah, why not? I think that'd be really cool. - Yeah, way to ruin my vacation. - And you know, Karen in the chat says, you know, healthcare is super siloed. Like loads, and the same with a lot of science, right? Like, a lot of science is very siloed. I was thinking about my background in the atmospheric physics. Like, a lot of the time we wouldn't be talking to people who were doing like ice, like working in the ice or working on like in the biosphere or working in the sea. Like you exist in silos and yes, to get exceptional people who jump across those disciplines and look for patterns, but we're not particularly good at doing that. And I think that's what AI could be really, really cool. You know, when we talked about like using AI for wildfire detection master week as well. Like how can we be able to look for signals and look for trends that we can then better improve AI tools and also better understand our own knowledge of how these systems work interlinked as well. So I think this is the kind of AI, I think everyone can get behind. And it brings us full circle back to the first story, right? I don't think anyone would be against AI being used to improve the standard of living and healthcare for people. But I think people would mostly be against using AI to produce slop that our children can then become addicted with watching on TV. Data centers are being built for both of these. Do both of these need to exist or as well as improving better technology, do we need to start having much better conversations with regards to what AI should and should not be used for? And I think, yeah, I just think that's a nice round off for how the stories can interconnect basically. - I think that's a really great point. I will just mention, I spoke to Daniel Nist about this, about where to point the magic box at. And I think that's actually the next big AI breakthrough. It's not necessarily always the strongest model. It's knowing where to point the magic gene yet. Like what it's actually worth doing and what's actually has a negative effect on mankind. So, I think it's a great point. - No, exactly. So just reminder, thank you everyone for listening to us today. Leo's gonna throw into the chat. We have a YouTube channel that Leo does an amazing job of curating, so you can check his on out there. If you're listening to us on your podcast or actually of choice, please feel free to give us a review and a five-star rating. And thank you all for checking in and following us on Substack. I'll put up the guest post, the post that accompanies this in the next couple of hours. And yeah, just wanted to thank you all, like loads of people again joining us today. Certain highlight of my week and appreciate your being here. And thank you Leo for all your work as well. - Thank you very much Sam. Thank you for everybody as well. Really appreciate everyone. - Thanks everyone. Stay safe, be well, go to the owner. (upbeat music)