The Best of 2025 – DTNS 5174

0:03This is the Daily Tech News Show for Monday, December 29th, 2025 in Los Angeles. I'm Tom Merritt. Why am I here alone? Because it's time to wrap up the year. 2025 is almost out the door. We saw so many big tech stories this year taking the headlines for DTNS from broad public skepticism over artificial intelligence to the increase in costs in computer components and devices as companies invest huge sums into data centers, the biggest data centers to ever roam the earth. So Joe put together our best conversations that we've had all year long, plus some of the more pithy comments from this year's Live With It and put them all together here for you to enjoy.

0:48So please enjoy our look at the best of 2025. Hollywood, have you played around with operator at all? You're like, where do we even start with this. I know. I was like... I mean, there's so much. I have not played around with Operator because I am not interested in paying$200 a month to try it out, but also because I have migrated almost completely to Claude. I just don't have enough faith, frankly, in ChatGPT's privacy and ethical practices to keep using it. I do think that Operator, I mean, for those who do use it, it does sound pretty promising. Oh, for sure. And I think that the, like, agentic world, apparently we call it that, which is just fun to say, agentic.

1:37We're inventing a new language around this. I think the agentic world is going to be freaking awesome, right? Like, as a person who read The Diamond Age by Neil Stevenson early on, I want the young lady's primer immediately. I want the thing that is just out there on the internet acting as my behalf and And frankly, that is how I use Claude right now. Because these agents ultimately have to interact with websites on the other end, and we all know how broken the web is. I'm just so curious to see, like, is it really going to, is it going to kind of be like when you try to log into Wi-Fi on your Kindle?

2:10Or like you try to use that Rocket Money app to cancel things for you? And it's like, cool, here's the page to log in and cancel it. I found these results on the web for you. Yeah. That nighed me literally this morning with Amazon's assistant. currently I pay an assistant to do for me a human to do and then I still have to like come in there on the back end and say I want this one and I want this one and oh here's my hopes I forgot to put my credit card in last pass like I think until until the whole world I'm excited about this potential I am in no way trying to diss this like I think the agentic web is going to be freaking amazing it's going to upend the entire advertising ecosystem and I can't wait because that's the best thing that can happen to the web.

2:49But it will it until we have agents working with agents, if the the if the barrier is the web, as we know it today, I think these agents are only ever going to be able to do so much. The whole capture thing is a really interesting situation. Because up until now, if it was an automated system, it wasn't legitimate by default. Now it will be how do you how do you deal with that? How do you make it so that legitimate, like I have given this agent, my authorization, please let it log into my account can happen. One thing that I thought about while you were talking, Molly, which I'd never thought of before is sites optimizing for agents.

3:28Like we optimize for SEO right now on websites. You're going to want to optimize so that agents have an easier time if that boosts your sales usage, whatever. Yeah. You know, you remember when there were digital transformation strategies, like everybody go update your LinkedIn and start calling yourself an agentic transformation strategist. Trust me on this one. You will make all the money. Wait, why am I not doing this? Why did I just come up with this idea? Let's all four of us do that. Yeah, let's do that. My question to you is if you had a roboboss, a robot supervisor, or perhaps a robot middle manager, how would you think you would operate?

4:07Would you think you would operate just as well, poorly, or better? and would it matter if the robot actually looked or acted more human or it wouldn't matter if it looked like a giant toaster what if this was just a supremely intelligent robot for which you could say hey look i'm not going to get this done today what are the repercussions and it's like uh you know you're kind of at a 60 percent uh production rate so you can probably survive this one but next time there's really going to be some problems i will say that the only thing that i can compare this to is the robot boss that I have now.

4:41And it's on my wrist. And it tells me when I am moving and when I'm not moving. And I got to say, sometimes it's good to see that I have moved. I do know that sometimes like when I'm a solo parent, as I am right now, because my wife is out of town, and I've spent all time with the baby, and then my watch reminds me that I'm behind on my metrics. I look at it with a kind of shock bemusement that Walton Goggins looked at Sam Walkrell last night on White Lotus. When it's a robot, I don't think I'm going to have to work as fast because the robot's not going to be actually disappointed with me. But I think there's a lot of subconscious things that play in is like, well, I don't want my boss to be mad at me.

5:19I don't want to have a conflict. I'm going to have to see them, you know, around the office. And then I don't want it to be uncomfortable. Like there's so many other things that play into it that don't play into it with a robot i mean wonder if you if you married it married it if you if you marry the robot then it's a different conversation yeah let them be supervisors yeah if you can find it glitch could it effectively guilt trip you no it could it could it could leverage your other human relationships i think that it could be very manipulative but but i don't think it could effectively guilt trip you itself it could it would need to bring in the other humans to do that So would it have to be like the secretary for the board on severance?

6:02Like it would be like an LLM that would be repeated through a human? Oh, that's interesting. Yeah. Because then you've got the human interaction thing going on. And even though the board is telling you something, you're looking at a human and you don't. Yeah. Yeah. I think that's a I think that's interesting. When I started posting all of these videos, I didn't think too much of it. I was just like, technology, tutorial, perspective, let me show you how to do these things. And I started noticing a whole bunch of comments from different people saying, like, are these truly photographs? Is this really a moment in time?

6:37This one will actually look at the picture and you can highlight an object, a person in the background, something that you're holding, and you can remove that item from your picture completely. And Samsung has gotten so good at this that they can actually make a basically a generative AI face where that object was removed from so that it'll give you a smile. In my example, I was showing a phone that I was holding in front of me. I mean, maybe not exactly, but like pretty close. Then I started messing around with Lightroom and I found this one was so incredibly good and efficient. In my case, I did a before and after of this picture of Edinburgh Castle.

7:23I didn't like the construction. It wasn't aesthetically pleasing for myself. So I went in and I highlighted all of the construction and it replaced it. In your example, it put in a castle that was blocked. Was that a real thing or did it just decide there ought to be something? So it added it. It just decided. How much are we going to allow AI to change the photographs that we are actually taking some historical moment in time to the point where the picture is not actually truly reality anymore? It depends on what your photo. Like a photo isn't a single thing. photos are more like paintings now than they ever were paintings were never perfect representations of what the artist saw they were what the artist thought would be presentable to people and that's what we're doing with photos we're saying i i want to present this image this is how i want it to look i will already edit those pictures and add saturation or brighten the the undertones or just Just make it more.

8:22Yeah, we all already do that. We all do that. And, you know, in all honesty, our phones are doing that for us anyway. I think we got in the habit of thinking of photos as being harder to alter. Like even when the Soviet Union was doing it, it took a lot of time and resources to do it. That's why everybody didn't alter photos. And now it's easy. So we're back to having to think of photos as alterable, not incontrovertible. They did multiple studies. One of the text exchanges was sprinkled with those kinds of abbreviations like PLZ, THX, GTFO. The other had everything fully spelled out. And the participants in that study rated the abbreviating sender as less sincere and less worthy of a reply.

9:09That effect held across all levels of relationships. Whether they were told that this person was a stranger or a close friend, they felt the same. They went to Discord and messaged random users, asking them to recommend TV shows to watch. One set of messages fully spelled out what they were asking. The other filled it with abbreviations, and fewer people responded to the abbreviated requests than to the fully spelled out requests. They studied participants in a private message portal engaged in speed dating on Valentine's Day, and recipients of abbreviated messages were more reluctant to share contact info than those who got fully spelled out messages.

9:53Science says if you're writing PLZ, if you're writing LOL, if you're writing OMG all the time, that your messages are going to be considered to be kind of a brush off. I thought we were all in the same page with this. We're all shortening things to get done quicker. Here's an emoji to tell you how I feel. You just hit the nail on the head, though, to get it done quicker. If it's truly sincere, you would take the time and the extra thumb motions to spell out the words. Sometimes the emoji responses can trigger me in the same way where I'm like, thumbs up. Oh, okay. Thumbs up is an interesting one.

10:31It has gone from like to like, oh. Because you like it or because you're like, okay. Yeah, like the Jennifer Lawrence thumbs up. The Jennifer Lawrence thumbs up is not a good thumbs up. I always think about that. Because I have Gen Z kids, one of them, and he told me outright, he says, Dad, when you just say okay at the end of the thing, and that to me is the most innocuous one. We've been saying okay long before the internet. And if I put just okay, he says his whole, he speaks for his generation, he says. He says all of them think it's dismissive and tired and I don't really care and it's not important.

11:05Written text is read in the most negative possible connotation. And that is why email tone is so hard to get right for people. And everyone reads okay as like, okay. Yeah. Like, you know, even though you mean it as like, okay. And in MMOs, all this short language has an actual functional reason. You don't want to stop to type too much because you are busy doing things. Someone says, I've got to go to the bathroom. They go, BioBreak, BRB. Nobody gets offended that they use BRB or BB for BioBreak or any of these short names. Because the context is okay for that. The problem is we have come out into the real world and continued to use it.

11:44And I think that there's other generations who are maybe not so much into that. They're playing Fortnite. I'm not saying this is all video game related, but this seems like there's a portion at least of this. I think it's hardwired. I think we are just hardwired to look at abbreviations as, oh, you're in a rush and you don't care too much. This is where LLMs can really come in handy. It's like you have an obsequiousness filter and you just do that shorthand. Maybe we need LOLMs to call something. LOMMs. Dang, man. Write that one down. OpenAI has expressed interest in buying Chrome if the Department of Justice forces Google to divest it, which is unlikely.

12:24this ain't gonna happen but no it's more of like let's talk hypothetical yeah let's let's fantasize obviously the big advantage of them buying chrome as is you know starting there and not building their own browser or acquiring somebody else if in this fantasy world this happened uh it would mean a huge install base of users the idea of using anything but google as your dominant search engine was i mean five years ago it's like what are you talking about i mean Who would that be? What company would be doing that? And now, not so much browser, but just search in general, the AI folks are really coming from these incumbents.

13:05They really are. I hate to say it because I've been very critical of certain generative aspects of AI for a long time now and still am. But when it comes to LLMs and solving problems and proofreading and quickly doing math and some of these like everyday task stuff that the free LLMs are there for, they're really great once you start using them and you realize, oh, my gosh, I don't know if I can go back to the other way of doing this. If there was a world where somehow this makes it all the way through the court and Google is forced to divest itself of Chrome, whether OpenAI buys it or not, is it going to prevent Google from making another browser?

13:41Well, that's what I was going to say. Like, why wouldn't Google just make, like, the Gemini-ser? Oh, Gemini-ser. Yeah. Gemini-ser. It could be called the Gemini-ser. The big story, I think, coming out of this is not just agentic AI. I mean, Microsoft was pushing agents. That's not a big surprise. That's what the entire industry is pushing right now. But specifically pushing two technologies that they're launching, one called NL Web, which will let you use a few lines of code to embed a text field and a submission button in a web page. And boom, that web page has access to a chatbot. uh mcp is often described as usb for ai in other words it lets multiple apps and services plug into each other i interviewed that kevin scott the cto at microsoft gave to the verge he was saying this could solve a lot of problems where sites are saying i don't i want to have control over what your agent tries to do on my website because it's not always going to work very well you use mcp and nl web to make it work better uh and it can make the agents work better because the sites can support it and mcp agents can go out and use that uh protocol to make sure they're going to take the best advantage it's going to make agents work better this is a moment in time that i think we should actually like memorialize a little bit because it reminds me of apps it reminds me of HTML.

15:11It reminds me of like core moments when behind the scenes or stuff that the general public isn't going to really understand, but people who care about tech do, will be able to point and say, hey, in a year when every browser we work with is first and foremost agentic, meaning that we are not opening a window and going to a website. We are opening a window and asking it for a thing that we want and then to just pulling it up, which is going to fundamentally reshape the way that we search for things on the internet. It's going to fundamentally reshape our relationship with browsers. This is the tech that you're going to be able to say, there were things that could do things like this until this happened.

15:56And now all of a sudden there's a whole world that can be interacted with that wasn't otherwise. I think this is the kind of stuff that is huge that we're going to look back on, and it's going to be a big part of the Wikipedia entry. Like right now, if an agent goes to a website that doesn't support MCP, it has to try to figure out that website. And that's a lot harder. It's not impossible, but it's a lot harder for the agent to do than if the MCP protocol is supported and the agent's like, oh, OK, so I just plug into this, this and this, and then I can order your food or get you the recipe or whatever it is it's trying to do.

16:33Now, the question is, is it going to bring the same kind of eyeballs? And that's where you get into other questions of like, all right, well, the free web, for as long as we've dealt with it, ran on advertising. Does that happen in the same way? yeah that was the interesting thing that scott was talking about in the verge article uh in the interview he was saying this can help with that because the mcp gives the website control too so it it could allow you to feed ads into your responses and require them it could allow you to require subscription plans in order to get certain things done and have the agent work like oh the agent could say like actually can't do that because you don't subscribe to that service but would you like me to sign you up i wonder how the the inevitable insertion of ads into responses like this um in this next phase of the internet as we know it yeah um or you know being connected as we know i don't even know if we call it the internet at this point um you know how that looks we are naturally i think ready as consumers to find a way to get this information in a way that is pleasing and interesting.

17:44And even if, and this is kind of another big, this has been a wishlist on the internet forever, micropayments, like Tom is talking about, that might be more of a reality now. Because the problem with micropayments forever, even as the technology has existed, I mean, PayPal's had fractional payments from when Elon was there. The tech has been around for a very, very long time. But it's just the onerous nature of like, oh God, I got to like, yes, I would happily pay 50 cents to read this one article right now, but you're telling me I got to hit a button and then I got to hit another button. But it's going to take me 15 minutes to pay the 50 cents.

18:24So forget. Oh, so annoying. But if you could have it be done at like, oh, do you want to pay 50 cents for this? Yes, computer. Great. Boom, boom, boom. Done. Done. It all happens. Now we're in a different world for it. Tom, do you think that all this is going to look really different in a year? Do you think that the browser is going to look very different in a year? Maybe not a year, simply because what we're talking about here is Microsoft saying, in Windows and Edge, we're supporting it this way. And MCP is a standard that's being supported across things, but it's going to take a while for that adoption cycle to get going.

19:02I think it'll happen faster than Pasci or something like that, because there's more motivation for fun things to happen from this. But two, three years? Yeah. Zoom CEO Eric Yuan used Zoom clips to deliver the opening comments on his earnings call. Now, the point was Zoom said that it is making the custom avatar add-on feature available to all users this week. So he wanted to use that feature to show like, Hey, look, I'm using it too. I look like myself. What I'm waiting to see, though, is that when it's not an avatar, when it actually is just AI driven to where now you just have an AI that looks like you and talks like you.

19:44I'm thinking of us. We have literally thousands of hours of video and audio that is recorded and on the Internet. You could probably make a really good avatar. They could pop that into VO3. Yeah, this stuff is getting better and better. At what point are you going to, are we going to be just this big simulation where basically your AI is talking to my AI and they're, they're, they're doing deals. There will always be those cases where people want it to be clearly not AI. Had it like a crazy basketball situation and you have a black eye and you're like, I can't do this. You know, the AI version of me is just, it's just, you know, everyone know that that's what's happening.

20:25Nobody's trying to like pull the wool over anyone's eyes. Just like using makeup, but with digital. Or a body double. Exactly. And easier. Or a body double. I don't know. I think we're going to get to the point where we no longer think of video as being representative. We know it can be simulated, and we will assume it's simulated unless we know otherwise. Congratulations, Sarah. Yes. Thanks to Hugging Face, we now have open source generative models inside of open source humanoid robots. Hope JR is a full-sized robot with 66 actuated degrees of freedom. It is controlled by a human in their demo to make it clear all the things it is capable of doing by mimicking.

21:13But you could have it controlled by an algorithm. They haven't set a price yet, but they said around$3 ,000 for Hope JR. $3 ,000 for a human-eyed robot. if you watch the demo video that we just saw, is pretty impressive. It's great. It doesn't seem like that much. No, that's less than some MacBook configurations, right? Very much so, yeah. So I got a question for you guys. Are humanoid robots, is that the way to go? Because it seems to me like if I had a robot at the house that was able to do stuff, a trash can with an arm coming out of it seems like it would be way more useful and more sturdy and more capable than a humanoid robot maybe it's just me but i i think you know i'd rather have an r2d2 than an actual c3po as far as the type of robot that i want to do things around the house some of the stuff that i see of like you know future of the kitchen you're like i don't want that thing in my kitchen all the time if i only need to use it to i don't know make the best pancakes every once in a while but if it was uh a a you know a robot friend that could be also a vacuum cleaner or also uh you know do the windows in a different room with with a variety of power sources then that that would be interesting would you prefer an adult-sized robot or like a child like something like that's smaller so it can fit into more you know different this should be an option on tinder i don't want the robot do you want a big humanoid robot or a smaller height i don't want the robot to constantly be asking me to get something off the top shelf what was the robot on buck rogers that said beady beady beady beady what was it yeah do you want it to be tweaky size or do you know i want it to be regular like Just regular human size.

23:08Average human adult size. It should be able to carry me in case I fall. But not like a fireman carry. It should be able to get taller when it needs to and then squish down. It's a robot. Some sort of giraffe-a-bot. Exactly. Because, I mean, you're all taller than me. You can reach that top shelf that I have to use a stepstool for. But I'm scrappier. No, no. So wait a minute. I can get down way easier than you. I mean, I guess the humanoid robot can use a stepstool, too. Or if you get something like a Sarah-sized robot, you could get several of them. They could do like an acrobat where one climbs onto the shoulders of the other to reach the thing of a top shelf.

23:47Yeah. That would be not disruptive at all in my kitchen, which is a small kitchen in general. They could be smaller, but you get like five of them, and they can combine into a big super robot. It's like Legos. Like Gortrod. It's like, whoa. Oh, we built a whole city. A gas pack, a hugging face. With all of the talk about Apple intelligence being de-emphasized by Apple in this announcement, thought it was smart for them to confront that head on and say, we're going to open with Apple intelligence. We're going to admit that Siri is the greatest thing since sliced bread someday. But that's not coming right now.

24:22That's coming in the coming year. and went right into what is probably the most notable of their AI announcements, which is a foundation models framework. Any app can tap into the on-device Apple LLM and use it in their own app. We've had a lot of stories that say, well, the reason they're saying that is it's a mess behind the scenes, and we've talked about that. But that doesn't mean that they couldn't be right. The question is in whether we will see those large language model features come and prove them wrong. I think Apple's inability to capitalize on artificial intelligence will quite possibly be looked at as the moment when they stop becoming a leader in innovation and technology.

25:12They have such an advantage here owning the trust of their user base because they have bet on privacy, because they have bet on hardware. I do think that if you are putting this out, then what you are not doing is focusing on how can we, Apple computers, the leader in customer service, integrated hardware and software, take this obvious burgeoning technology and make it the best it can for consumers. I just think they stepped out too soon. I think last year when they announced Apple Intelligence, I believe they should have held back and did or are in the process of doing all the things that you mentioned.

25:58Tech journalists were saying Gemini, Copilot, Claude, Mod, Xi, ChatGPT, they're off to the races. What are y 'all doing? And Apple was like, we're doing some stuff. Number one, if they did what you said, where they reacted to the press, then they are even more lost than I think they are. Well, consumers are saying that too. Yeah, sure. But Apple's never cared about that. Apple's cared about you will get the product when we are ready, and it's going to be something that is materially a step beyond where things have gone before. We have now seen them push out their biggest hardware failure of my lifetime with Apple Vision Pro.

26:39and we have seen them launch Apple Intelligence when, to your point, they couldn't do things that were interesting. I mean, hell, they couldn't do the things that they were running ads about on television, let alone things that were actually interesting to the consumer. When they put out an AI product, you would have said, oh, my God, this is great. Now I can use this technology to interact with all this data that I have on my phone, My address book, my health data because I have an Apple Watch. Like there's a lot of stuff that I trust to Apple that now I want to be able to be remixed and fed to me.

27:16And if they were further along, if they were where Google is on AI, if they – I mean let alone where OpenAI is, the features they would be able to plug into their phones would be a huge step forward. They were saying our interface actually reacts to you because of what we know about you on device. Sure, but that's machine learning from 15 years ago. But it works. I mean, number one, we'll see. I think what Apple's trying to say is point where it is being done better in a way that makes people more excited to buy that instead of us. and you can say that in a world that open ai doesn't develop hardware yet this is a problem for them and i'm a huge apple fan i'm looking at a macbook i have an apple watch i'm talking on airpod pros i get a new iphone every year because i pay them monthly i am as ensconced into this universe as one can be it worries me that they are not on the ball here at its most cynical, Apple is a 2007 supply chain company that will slowly bleed out its relevance because they are not on top of the technology like they used to be.

28:36Wall Street Journal saying that Walmart and Amazon are exploring issuing stable coins. If you don't know what a stable coin is, it's a coin that's pegged to a currency. Usually it's pegged to the US dollar so that one stable coin always equals one dollar. Apparently Expedia Group and some airlines have also discussed issuing their own stable coins. The reason they would do this is if you have stable coins on a blockchain, you can handle transactions yourself and you don't have to pay fees to the banking system. You can have people input money, turn it into stable coins, and then do instant transaction with no fees.

29:10Meanwhile, two other stories today that made me want to bring this. Shopify is offering USDC, which is a US dollar-based stable coin, to a limited number of merchants. Those merchants can accept USDC from Circle as payment, and they can get paid through USDC in their local currency. And then Coinbase announced it's offering an American Express card with a$30 a month subscription plan. So if you're a$30 a month subscription person on Coinbase, you can get an American Express card. It really struck me that these weren't startups. These weren't things you have to learn to trust. We're talking about Walmart.

29:48We're talking about American Express. We're talking about Shopify. Like these are big, fairly well-trusted names. Are we at the point now, Rob, where we are about to see this become mainstream? How does a government control its monetary policy if it doesn't control the money? So the government controls dollars and cents right now. They don't control Bitcoin. They don't control stablecoins. So I'm curious to see how are they going to regulate this? We're talking about stable coins that are U.S. dollar linked, but you could just as easily get a stable coin that is linked to the ruble or linked to the yen or what have you.

30:27That's a really good question. And I think that's why these companies are looking at stable coin, because they're almost always dollar oriented. You're not wrong. You can put it in whatever currency you want, but they're almost always dollar oriented. Even the U.S. is examining issuing their own stablecoin for U.S. dollars. And if each stablecoin is equal to one U.S. dollar, then U.S. monetary policy is going to affect that. These are just meant for transactions. They're not meant for saving. So that's another way that this doesn't totally cut out the government from it because you're cashing these out into whatever currency you want.

31:03With Shopify, you can cash it into rubles or yen or euros or whatever you want. So it goes back into the government regulated systems in those cases. Okay, that sounds great. Somebody's got to be mad about this, though, because somebody was getting that money. Banks, banks not going to like this. The banks are already trying to figure out how can we exist in this new world to where we still get some of those transaction fees that we get right now. One of the things they're doing is creating a bank consortium to issue a stable coin so that they're like, well, if we fail on that front, maybe we can use our stable coin.

31:37If you can't be able to join them, you just do it better. It's the stable coin you know and trust. Yeah, exactly. If you're a business, you are paying those bank fees constantly and they build up. And if you can cut them out, it's a huge savings. And you speed things up. You can make transfers super fast, super instant, which is already happening in a lot of parts of the world that they're getting fast. But you can make them literally real time. The big problem with a lot of these mixers are you can have 12 channels in. It's always going to do two out. It's always stereo out. So you lose – also you lose a lot of flexibility when – even if it's over USB.

32:15You're really only getting stereo out. So a lot of that flexibility of, oh, yeah, I'm recording all this stuff. Well, you're kind of baking in, especially when it comes to like music production. Sometimes it sounds good in your headphones. Then you're trying to master it or something like that. I barely know what I'm doing there by the way. But you don't have that flexibility. It's a pain in the butt. So it would, yeah, having something with a recorder that can do each track as its own recording would was definitely something that was appealing to me. But what's your what's your takeaway, Tom, with these X-Real glasses?

32:46Yeah. Wait, if you are somebody who heard me say I can watch video on a plane comfortably and you're like, I travel a lot. $649 is worth it for me. Sure. If you're somebody who's like, actually, I could use a big old monitor and I don't mind plugging these into my plugged in laptop or desktop at home. And that's cheaper than a monitor you would buy that's 171 inches, which it is. Sure. For most people, I'd say wait. Xreal is actually a partner with Google on developing glasses. They're going to have better versions of these out. So this is really an early adopter situation. You did mention earlier that, you know, somebody has to think about cleaning it pretty regularly.

33:32Obviously, some components can come out and be cleaned in a dishwasher or, you know, hand washed. Do you feel like over time it still feels pretty safe to put food inside? Absolutely. I've had no issues with, you mentioned the toaster ovens. A lot of ovens in general can have like grease buildup this i could i take it out and i wash it and there's no grease buildup and it's been pretty easy for me neilay patel the the editor-in-chief of the verge does this great article where he's basically like just buy a print you know but just buy a brother and if you search like the verge just buy a brother or the verge you know um uh laser printer you're going to get neilay's article which is kind of a rant uh and it's a wonderful rant uh Because it goes talks about printing and why it shouldn't be a problem and why so many vendors are nightmares and search engine optimization.

34:23And it's a whole thing. Is there a darker side to these models, Justin? That is the supposition of the New York Times, which has some extraordinarily sad stories. People whose mental health deteriorated when they began having dark conversations with Chad GPT, which eventually pushed them, at least according to the direction of the article, into worse mental health positions. This is a dark, dark element of what can happen, but how much of a problem is because of the technology specifically? It really depends on the person. You know, what is the thing that you're going through? What do you need help on?

35:11And, you know, is this going to make you better or worse? And you could say that about a human therapist as well. you could be like nope that therapist was not for me yeah not the not the person that was right for my brain i think every new thing uh faces this question so violent video games right did they cause people to be more violent uh when when violent video games first came on the scene a lot of people were like how could you argue that they don't look at that look at that video game that's showing you how to shoot a bunch of people. And then this person who watched that and went and shot a bunch of people.

35:52And so there's an, there's an analogy there here, which is how often does this happen? You know, does, does this really encourage this or is it somebody who is prone to that? And the New York times article tried to make the argument that these are people that were not necessarily indicate, had indications of mental instability before they went down this rabbit hole. And one story they told, I thought they'd told fairly well was somebody who started to realize, oh my gosh, I'm getting sucked into this. And I think that's a concerning thing. I don't know how widespread it is. That's not me saying it's not widespread.

36:23I'm saying I don't know how widespread it is. And we should know how widespread it is before we make a determination about this. There should be an alert system or a safeguard or something that is triggered when certain patterns are created that says, hey, are you sure you want to be going down this road? And it looks like those safeguards failed in a couple of instances. The problem to solve here is with a technology that can be as engrossing as this is, what are the safeguards you need to build for worst case scenarios? I don't know if the problem is, does this technology take an even keeled person and make them paranoid and deluded?

37:09What is the worst case scenario? At what point does the LLM have to be like, nope, can't do it, buddy. Can't do it. And can it stop people? People who have had mental instability throughout history have fixated on technology far less sophisticated than chat GPT. We went through an entire generation where we worried about whether or not Catcher in the Rye was signaling people to murder celebrities for some reason. if there was hidden messages in various different pieces of media, the human mind is built to read more into things than is actually there. And on the most delightful stuff, it's the reason why we bond to art.

37:50And on the most dark and troubling, it is stuff like this. I think the solution here is after day three of a conversation about whether or not you're in the Matrix, Is it the responsibility of the platform to cut the conversation off, to put up a signpost that says, hey, by the way, I'm just funning? Yeah. Yeah. Like what if you want to use this as a storytelling platform for yourself, for your own entertainment? And you're like, I'm just role playing. Like should that not be allowed just in case or what are the signs? I will end this conversation on two notes. It's the most important being if you are seeing any situation where you feel that you are slipping into something, then please, please, please do not only go to an LLM to solve your problems.

38:38Please talk to a real human. And also, I am very, very concerned about stories like this when we get into a really hard level of fear, uncertainty and doubt that historically we have seen panics arise over technology and art that the examples ignore other societal things like substance abuse. family issues, economic issues that are usually many variables, usually the reasons why these things happen. And we understand that, but it's not novel. It's not newsworthy. Heavy metal is video games are professional wrestling is, and now chatbots could be, and I'm not saying there's not a problem because I do think that there's a technological question here that does need to be discussed.

39:35But the I just want to take my stand and stay when we when we get to the this technology is killing people question, we need to be very, very, very sure that we are ascribing things accurately. New York City filed a lawsuit against Facebook, Google, Snapchat and TikTok, alleging that their platforms fueled a youth mental health crisis by fostering social media addiction. Where is the medically valid scientific proof that this is happening? I think a lot of people suspect it's happening. I think it's reasonable to wonder in what ways this is damaging. But that's far from being able to say in court, they were gross negligence and creating a public, the public nuisance thing is just a legality.

40:23You know, that's just the way you file this. It sounds weird, but it's, but, but they're saying you should have paid more attention because we all knew this was harmful. And I don't think that's true. I think we suspect it might be harmful and we don't even know exactly in what ways it is harmful. It's like, what is social media addiction? What does that mean? And at what point does it become something where it's like, oh, I just watch MTV all day, which was me when I was a teenager, to today being on a social media network too much. I'll give you an example. When I was a kid, I used to watch TV way too much to the point where my dad physically unplugged the TV, put it in the box, and stuck it somewhere in the garage.

41:07I never told him, but I knew where he kept it, so I physically unboxed it. I would watch the shows that I always watched, and then I would unplug the TV, put it back in the box, and stick it back in the garage. People use addiction to be like, oh, I'm so addicted to social media, right? Meaning I use it a lot. That is not the medical definition of addiction. And there are different kinds of addiction. There are physical addictions that are dependencies, and then there are kinds of addictions that cause withdrawals. And then there are psychological things that I think have a different name from addiction but could be used in this court case to say it is like an addiction, right, and it causes harm.

41:43The problem is even if you just get away from the definition of addiction and focus on mental health harm, all the studies that have been done are cross-sectional, which means they're just from one moment in time. They're not longitudinal. They are also correlative most of the time, which means we haven't ruled out that there's something else that causes people to use social media a lot that also is the cause of the mental harm. and that the social media use is just a symptom of that other thing and not actually the cause. And until we rule all those out, it's going to be difficult to prove gross negligence.

42:24I think that's where I go. I'm not trying to say, oh, no, social media is fine. I'm saying it could be harmful, but we don't know. And can you prove gross negligence? Not yet. We need to wait to figure that out first. I wouldn't be shocked if we found that social media use did cause harm, but more often was used to self-medicate for other problems, right? Like there's probably a dual situation going on. It's like a disassociative thing. Right. Like, yes, it doesn't help and it might actually worsen symptoms of a preexisting condition. But when you are depressed, you might spend a lot of time online trying to get that dopamine hit that makes you feel a little better.

43:07And, you know, what this perfectly describes is those spots where when I was younger, I would play video games excessively because, like, as Tom described, self-medicated. Yeah, yeah, yeah. Because video games allow you an amount of control, especially if you're in a point in your life where things seem very, very out of your control. There's this one bit where you get to control the world, you get to control yourself, and you understand the rules. We would be negligent if we didn't do our research. The American Society of Addiction Medicine defines addiction as a treatable chronic medical disease involving complex interactions among brain circuits, genetics, the environment, and an individual's life experiences.

43:45people with addiction use substances or engage in behaviors that become compulsive and often continue despite harmful consequences. Is social media compulsive? And not just I use it a lot, but like I cannot literally stop even if I want to. And does it have harmful consequences? Is it getting in the way I use social media and my grades are declining or I'm not spending time with people or I'm not able to hold a job? That kind of stuff. There's a lot of conversation about what is going to put us in an AI bubble. Is it the over-promising on capacity? Is it the circular financial engineering of taking money from one company and turning around and using that money to buy their chips?

44:25Is it over-promising and under-delivering? But the simple fact is that every conversation about compute is a conversation about data centers, and every conversation about data centers is a conversation about energy. Like finally, there was a conversation with Satya Nadella in which he said out loud that chips are not the problem. Power is the biggest barrier to AI growth. One gigawatt is equivalent to one nuclear plant. In order to meet, and the United States, just side note, has built three nuclear plants in the last 30 years. In order to meet the electricity demand that just the U.S. is forecast to require, we will need 80 gigawatts of additional capacity every year for the next 20 years.

45:20I mean, it's just 200 to 400 utility scale solar farms, 800 wind farms. The scale of power that they just casually talk about is unfathomable. We're not going to try to get there with only nuclear plants. There will be a mix and match of all of the above of the things that I said. But even the idea of building four nuclear plants a year and then 600 wind farms and then 22 natural gas plants, there is no reason that every single inference needs to come from the entire corpus of human information. That's just inefficient. One thing that was pointed out to me is that AI doesn't care where it goes to school.

46:00An LLM doesn't need to be. There's a sense that if you build a data center, it has to be online 24-7 with like what they call five nines reliability like that power can never go out if you're just training that doesn't matter so you could build a big facility that is powered by intermittent solar power with battery as a backup and maybe that power is non 99.999 percent of the time but it doesn't matter because you're just training and then you have a smaller always facility for inference that, you know, translates that information into answers. Like there's a lot of clever solutions that I think we are going to need to demand from hyperscalers and that they will start to demand from themselves because they'll realize like we wrote a whole lot of checks that the American utility infrastructure cannot cash.

46:48If they could underwrite or at least subsidize solar on my house, I will help contribute to powering their future. But I mean, think of all the roofs that don't have solar panels that you could easily tap. I talked to NVIDIA at the beginning of this year. I mean, the chips get like 100 times more efficient with every new architecture. They genuinely do. I think, like, Roger, you're totally on, like, I can't say who, but I was talking to somebody recently and they're effectively working on, like, SETI for compute. I thought you meant like three solar panels for my roof. But yes. Merry Christmas, Roger.

47:25but that right like if you if you let me rent some space for compute you get free power for life like that is on the table that does it for 2025 thanks to everyone who supports the show on patreon we could not do it without you our holiday programming is not done yet we'll be back on tuesday with our 2025 predictions results show talk to you then

47:51Happy holidays and sending good vibes to the DTNS extended family. Diamond Club hopes you have enjoyed this program.