AI Slows Down Coding - DTNSB - 5058

0:00Huyen Tue Dao: This is Matt Rogers from Lost Culture East. That's with Matt Rogers and Bowen Yang. This is Bowen Yang from Lost Culture East. That's with Matt Rogers and Bowen Yang. You know when people try a new food and suddenly it's like, okay, hold on, I got a new favorite food. That's the reaction a lot of people are having when they first try Kewpie mayo. Yeah, it's the one with the red cap and the little baby on the bottle. You've probably seen it at the grocery store. And this mayo is different. Mm-hmm. Most mayonnaise uses whole eggs. Kewpie only uses egg yolks, which gives it this rich umami flavor.

0:29It's smoother, deeper, almost buttery. Once people try it, they start putting it on everything. Egg sandwiches, fries, burgers. Chefs use it. Restaurants use it. People who really care about flavor use it. Put it on just about anything. Then you'll understand. Kewpie, the original Japanese mayonnaise. This episode is brought to you by Capital One. Capital One's tech team isn't just talking about multi-agentic AI. They already deployed one. It's called Chat Concierge, and it's simplifying car shopping. Using self-reflection and layered reasoning with live API checks, it doesn't just help buyers find a car they love.

1:03It helps schedule a test drive, get pre-approved for financing, and estimate trade in value. Advanced, intuitive, and deployed. That's how they stack. That's technology at Capital One.

1:14Huyen Tue Dao: Support comes from WISE, the smart way to manage the currencies you need around the globe. Fed up with losing out to hidden fees when you send money abroad with your everyday bank? Choose the smart way. WISE. You can count on the exchange rate you'd usually find on Google. No unwelcome surprises. Plus, ditch that where's my money feeling. Most transfers arrive in under 20 seconds. Join millions, saving billions on hidden fees. Be smart. Get wise. Download the Wise app today. T's and C's apply. This episode is sponsored by HeyGen. AI is great at saving time, but only if it actually fits into your workflow.

1:52Huyen Tue Dao: That's what makes HeyGen so compelling. Instead of spending hours setting up lights, cameras, and retakes, you can turn a script, photo, or presentation into a polished video view in minutes. No production crew required. Whether you're a creator, consultant, attorney, financial advisor, or real estate agent, HeyGen helps you create consistent, professional video content without making content creation your full-time job. It's rated the number one AI video platform for small business on G2, features some of the industry's most realistic AI avatars, is trusted by more than 30 million people and 85 % of the Fortune 100, and supports over 175 languages, so you can communicate with audiences around the world.

2:32Huyen Tue Dao: Here's the wild part. Record yourself for just 15 seconds, and HeyGen builds an AI avatar that makes professional videos for you on demand, so you can post everywhere your audience is, as yourself, without filming every time. Your first three videos are free at heygen.com slash pod. That's H-E-Y-G-E-N dot com slash P-O-D.

2:56Huyen Tue Dao: This is the Daily Tech News for Friday, July 11th, 7-11, 2025. We tell you what you need to know, follow up on the context of those stories and help each other understand. Today, a study shows AI makes developers slower, not faster at coding, and a major ISP gets rid of Data Caps. I'm Tom Merritt. I'm Huynh Wetao. Let's start with what you need to know with that big story. I can't wait. All right, I'm going to talk a lot just so we understand the parameters of this story, and then we're going to hand it over to you, Huynh, because you actually code, and I do not. So this is going to be good. But bear with me because there's a lot of table setting here.

3:39Huyen Tue Dao: First of all, we have this study from METR. It's pronounced METR, but it's spelled M-E-T-R, and it stands for Model Evaluation and Threat Research. It is a nonprofit created to create scientific methods for evaluating frontier models, mostly to assess catastrophic risks, what's called alignment, but also to make good decisions about development. It's founded in 2022 by Beth Barnes, a former OpenAI alignment person who, through a securitist route, ended up founding this. Meter has cooperated with Anthropik and OpenAI and lots of folks and has put out this paper on its own website. So this is not peer-reviewed in a journal, but it's a randomized controlled trial that aimed to see what effect using certain tools had on productivity for experienced developers working on mature projects.

4:34Huyen Tue Dao: That's important. These are projects that have been around for more than 10 years. The projects generally have around a million lines of existing code or more. And it's a small group. The study enlisted 16 developers from these major open source projects. Now we've got 246 total tasks that were evaluated from those 16 developers. But if you really want to expand the conclusions, you need to have a bigger group. They know that. This is just one of those, like we started small to see if there was an effect. Each developer was asked to predict how long it would take to complete each task the normal way and how long it would take if they used a generative model.

5:13Huyen Tue Dao: Because one of the things they were testing for was how long do they think it'll take them? And then how long does it take them? How much time does it really save them versus how much time they think it will save them? Tasks were then randomly assigned so you didn't know when you made your predictions whether you'd be using the models or not. The developers who were randomly assigned, yes, you should use AI models, were allowed to use models primarily Cursor Pro with Clawed 3.5 or 3.7 Sonnet, which at the time they did this study, February through June 2025, those were the current models. Then they did their tasks.

5:50Huyen Tue Dao: They recorded their screen, so there was an independent evaluation of their time. and then if they were using a model, they were asked to estimate how much time they thought they had saved. Meter found that the difference between the predicted time and the actual time was 19 % longer for those who use generative models. In other words, they thought they were going to save time and it took them 19 % longer than they thought. In the exit interview on those who use the models. They were asked how they used them. This is important too. 70 of the tasks use them as normal. 119 of the tasks said, I experimented.

6:32Huyen Tue Dao: I tried some stuff that maybe I wouldn't normally do. And 57 said, oh, I used it as much as possible. We're doing a test of AI. So I tried it out. And so you might think like, oh, maybe that accounts for the extra time. But those who used it as normal were the least slowed down, but still were slower than you would expect. 93 % of the participants that had used ChatGPT, so it's not like that they were unfamiliar with using these models. 44 % had used Cursor, so maybe less familiar with Cursor. All of them received training in Cursor before the studies started, so they weren't coming in cold. None of them showed significant changes in speed as they became more familiar with it.

7:13Huyen Tue Dao: So the scientists think the slowdown probably wasn't caused with unfamiliarity or we would have seen an effect. Those who used cursor did have to switch to a new development environment, so that could have affected speed. Those who used the models added 47 % more lines of code. Often they would add more tests or more comments than those who weren't using models. However, the increased time was also greater on tasks that didn't have those lines of code. So when you drill in and look at the tasks, even when they weren't adding a bunch of lines, they still were going slower. Those who used the models also spent time reviewing the code and reworking it to meet standards.

7:51Huyen Tue Dao: It's possible that those using the generative models spent more time, but less effort. The amount of time spent waiting for models to generate output was not measured. Those who use the models spent less time actively coding. We do know that. But there was added time prompting, waiting on responses, and reviewing the code. And it's worth noting some observations from the author of the second thoughts substack. These projects are less likely to have material that would have been included in training because of the amount of undocumented knowledge carried in the heads of long-term developers like these.

8:28Huyen Tue Dao: The developers are also used to the strict guidelines of their projects, and these models are not. And then there's just the years of experience that make you faster at stuff. So lots to digest there, I know. But Huen, I know you have thoughts. Okay, so just to put my cards on the table, I have used AI for programming tasks and I am just very explicitly very bearish on it. I tend to be a little pessimistic. I have tried not to be a Luddite about this, but I am deeply concerned about a lot of the impacts of AI on our industry. And so on average, I'm going to be a little more pessimistic. And you can maybe just think that I might have some confirmation bias about the study, which is totally fair.

9:08And just to give it more flavor, it's really, it's interesting because this is a very hard thing to measure just because our jobs are not just writing code. and so this is going to be measuring or trying to study a a a particular way of doing a certain part of our job i i will say though okay so the thing is i'm not surprised i'm not surprised by this and this is the thing that kind of concerns me and i think what is kind of really cool about you know generative ai with programming is that yeah for for tasks that you have no experience in that are maybe outside of your usual work outside of your expertise yeah it's really great and i I think, you know, if vibe coding and I make fun of it a lot, but I do think it opens up doors for people.

9:51I use vibe coding, too, just to with things I'm not familiar with. So I don't want to dump on it too much. But it is a lot different from making something just work to as these projects are working on legacy, working on legacy products that have to be used in production for real people over a long period of time where you have to maintain it. You have to maintain a certain level of service and stability. And so, yes, this is where I get concerned. And it's, you know, and like, I think one of the biggest reasons that a lot of the AI is getting promoted to us is increases in productivity. So anecdotally, I will say that I get like, according to the study, a lot of people were very, even people that were very positive on it experienced slowdowns.

10:35And why do developers, or why are developers who are very positive on these tools like them? One of the things that I thought was really interesting, which kind of resonated with me, was a style of using AI. And this is not my style, but this is the way a lot of people do it. Is that, again, they'll write a prompt, have it generate code, look at it, kind of review it as the study set, and then prompt again. And so for a lot of developers, especially in this time of remote work, like the idea of having a buddy, a coding buddy to basically iterate or throw ideas at. And it seemed like some people were excited in the study to do that as well.

11:11Just let me just use as much as possible, see what happens, experiment. That's all great for sure. And I mean, that's just going to inherently slow things down, right? Just because you're iterating more. So it's hard to like measure the, not necessarily productivity, but the impacts of that. So I admit that. But I do think that in terms of raw productivity, it does, especially when you have experience in this, it does leave to situations where, yeah, you're a little bit slower because I could adjust on this. And also, we're not at a point where I think AI models are able to catch everything that we write code for.

11:48And there's so many times where you write code for something that's just so specific to your project. Or it just is something that, I don't know, it's just hard to see. And yeah, you still have to review the code. And that's like another thing is like, you know, when you're vibe coding, you're not, you don't have like millions of customers depending on you necessarily or thousands of people using your library and thousands of projects. And it's on you to make it as airtight as possible, as resilient as possible. And so when, you know, we write code for these larger projects, you have to review it.

12:23You have to test it. And the fact that, you know, I generate something and I have to review it anyway does slow me down. And it's so interesting because, I don't know, there's just a lot of factors to this. Again, I just don't think this makes a lot of sense to me. And it also depends on how you use it. I generally don't use it much for my day-to-day programming. I am a little bit faster. But there's also nuance to this because there's certain kinds of AI. Like this thing of like, okay, do this thing for me. Generate some code that does thing A and B, that accomplishes C. And then you wait for it.

12:58You get the code back. you review it, you maybe update it so that it kind of fits what you need to do, or, you know, you maybe fix some mistakes or something, and then you kind of repeat this process. I use it differently. I tend to use more like IDE language specific models. Like for example, I'm a Kotlin developer. I use Junie a little bit. I use Studio Bot in Android Studio because I'm an Android developer. And that's a little bit different. I mean, it's similar, but it's not different. Like for example, again, like you mentioned, Tom, that it's used a lot on tests and comments. I generally tend to use it on things that I, that just take me time to type out because I'm a human being and not a computer.

13:34And that's where I find the most productivity. Like I was creating a schedule component and I was like trying to create some test data for it. And so rather than me typing out a bunch of JSON, that's like, okay, well, I have a class at 9 PM and then another class at 9 30 PM. And I want to change the names. I had, I had, I had, I think I had clawed to it and I'm able to generate variation in that data and say, Hey, can you make like the data set have overlapping classes, but like no classes like in this room. And so that actually is like this kind of like very rote brute force. Like I just need you to generate a bunch of things and understand that I need certain situations.

14:09That's where I use it. And, you know, or if it's like an AI that's very specific to the language, it kind of just knows, it's able to understand what I'm writing and just come up with like boilerplate, like, oh, here, here's like the outline of a test and you can fill it in. So it's very nuanced, but I don't know. It's interesting just because I think that we still don't really know how this impacts our job. And I literally was in a meeting yesterday where we were talking about using AI for a work thing. And the PM, the product manager made a specific point. It's like, hey, you know, programmers, and again, like 30 % of statistics are all made up, but like we don't spend all of our time typing and coding.

14:50The statistic that my manager gave out was like 16%. especially senior developers like myself, we spend very little time coding. I actually did not program this week, but I worked my ass off this week. So it's so interesting because I think a lot of the productivity that we get with AI is real, but also there's a perception with it, which is what this, like, yes, maybe there's an argument for less effort. Maybe there's an argument for certain use cases that are just very specific and very niche. So it's really hard to say, but I'm not surprised. And I think that we're still trying to figure out what AI actually is doing and what it actually knows and what it doesn't know.

15:27My husband gave me an anecdote of someone that he works with who is very good, very experienced, who actually thought that AI was making him better and more productive because it caught something once that he didn't catch. And he's a very senior developer. And so I think because AI can be so good at certain things or just like, you know, catch certain things that we can't catch because it's able to kind of survey and consume all this different code. It is making us think, wow, it really is. Is it like, is it as good as I am at my job? And I think we don't really know, you know, as a counterpoint to that.

16:04I found out something that I missed just by reading someone post something on LinkedIn. So I think it's really weird because it's really hard to also attribute.

16:10Huyen Tue Dao: You're not promoting LinkedIn as the savior. No, exactly. I mean, and as a counterpoint to that, within the last six months, I've had we I've had a bot record, like applaud someone for doing something blatantly wrong at work. And it's hard because even at any particular team, people have different experience levels. They have just different areas of expertise. So someone might miss something. And I can't help but also think like, you know, we write bugs. so a lot of times and sorry this is just a lot of dirt feelings i'm not surprised i i don't think i i think the productivity angle again being being a very bearish person is overblown and that we really have to really evaluate what these tools are used for how we're using them because it's i don't think it's going to be especially again on older more nuanced and very strict projects It's not going to be as simple as hit the button, generate the code, fire, like get rid of the developers.

17:09You don't need them. It's it's and I'm just yeah, this makes sense to me. It jives with my actual lived experience. Yeah.

17:17Huyen Tue Dao: So I think this study is going to get misused and well used. And I don't know in equal measures, but a lot of people are going to point to it and say, see, AI makes you less productive. And that's not what it's saying. I think it's really well laid out and well done and says it is going to make you think you're better than you are, which is something that these tools have been proven to do in lots of ways. So be aware of that. It also is going to speed up your active coding. Like that's in the data. The only things that people who were using the tools spent more time at than people who were not using the tools were idle time, waiting on AI, prompting AI, or reviewing AI output.

18:10Huyen Tue Dao: So what made them slower were things related to the tools. The tools did make them faster at coding, which is why they feel like they were faster at coding because they actually were, right? But it just added all this other time that didn't kind of jive with their estimation of themselves. Yeah. And again, it's so nuanced because I think like anything else, any job, there are just so many nuances and so many different ways that you work that AI, we need to figure out the specific areas, the specific times where AI is most valuable. Yeah. And then understand this is where it could really slow us down and hurt us.

18:47Huyen Tue Dao: And like you said at the beginning, with things you're unfamiliar with, I think it is widely agreed like, oh, yeah, that will make you faster at those things. And I can hear all my friends who work at AI companies going, now it's getting better all the time. Right. And Steve Newman at his sub stack admitted that he's like, yeah, everyone says these things are getting better all the time and they are getting better all the time. So maybe a year down the road, the study will look different. But we don't know. We're just talking about where it is at the moment. Yeah. I mean, I definitely say that I've saved hours of time using AI, but I've also spent hours of time having to explain to people why the AI was very wrong.

19:29And please don't do this or please don't turn your brain off.

19:32Huyen Tue Dao: Yeah. And that's the kind of thing that I tend to get in a soapbox about is that, again, the situation where you're using AI as like a rubber duck, a whiteboarding buddy, that's great. I don't want to take that away. And I think that it does, again, as you said, the biggest thing is like, if you don't know how to do something, it can take hours down to minutes or seconds. And that's, that's super valuable. But the kind of fine, the fine, like the fine details, the understanding, like the whole business context and social context that your work lives in is a lot different. And yeah, I don't know.

20:05There's just, there's just a lot that we need to figure out.

20:07Huyen Tue Dao: Yeah. And it's not, it's not cut and dry. And it may get better at some of those things. It may get better at some of those things very quickly. It may take longer. I definitely feel like on the longer timeline, it'll probably get better and better and better, you know, at a lot of this stuff of being able to be like, oh, just be able to train it quickly on your particular project's guidelines and it will be able to do them. All that kind of stuff is definitely possible. But for now, it's a good, I think this study should be taken as a good perspective setting of these tools aren't bad, but like Quinn said, they're good at some things, they're not good at others.

20:47Huyen Tue Dao: It's nuanced and it's good to be aware of that. Yeah. I actually have been using a story that we talked about. Sorry, I'll start shutting up about this a while ago. No, no, that's why we're here. We talked about like journalism, right? I think, was it Andy Beach who talked to folks about journalism and where AI might live in journalism and where it might be more of an editorial role that human beings take? Whereas, you know, the AI might be generating that. And I'm sorry, as I'm not a journalist, I forgot like what the kind of like. Just the copy. The copy, the groundwork, right? Right. Right.

21:15And so that's what I think, you know, obviously in in programming, it would be very similar. Whereas what I do as a senior engineer, again, I barely coded this week, but I worked my butt off like talking to other teams, learning how this works, going through like specs for this project that I'm on and talking about risks and all kinds of things like that. And so that is more so that, again, it's like a similar role, right, where we're kind of synthesizing information, drawing on experience that we have and say, OK, like in this case, maybe this is the right tool to use. well what about this and this problem that I've learned from my mistakes like that kind of stuff right it's more of a kind of stewarding like kind of architectural perspective and that's great and that's fine the one part that I'm really worried about is for for for engineers coming up today and I've literally had younger engineers like new grads tell me that they're extremely worried and not just from the like socioeconomic of like well we have a job but also for me I'm just like I got that, you know where I got that experience and the ability to do this is from literally coding and making mistakes and having to think for myself.

22:15My only concern is that I don't have a problem with a staff senior engineer sitting down and like, you know, whiteboarding through something and like doing crazy stuff with cursor at ChatGBT. Because I trust them to like look over things and be like, okay, let me like refine this. But I don't know where the experience, the human experience that builds that senior engineer and that stuff engineer is going to come from when we have a lot of pressure in the industry to just sit down, just use it. If you don't know what's wrong and or if you don't know where the gotchas are in some code that chattybt pushed at you and you just don't have that experience, I don't know how you're going to gain it.

22:51You know what I'm saying? Because it's just, well, it's not going to be as easily gained because someone's going to have to give you the time. I don't know. It's just, that's the part I worry about.

23:01Huyen Tue Dao: And the tools aren't good enough yet for you to be comfortable treating them the way you treat C++ versus assembly, right? Yeah, absolutely. Because there were programmers in the 80s making the same arguments of like these interpreted languages, like you really need to understand COBOL and you need to understand assembly to know how to code. You know, object-oriented programming is great and all, but eventually object-oriented programming got so good that you, yes, it's nice to have an understanding of that, but you don't need to use it. And that is possible here too. Like maybe this gets good enough where you, as long as you have an understanding, it's enough, but we're not there at the moment.

23:37Huyen Tue Dao: And so it's uncertain, like, well, are we going to get there or are we going to stay in this mushy middle that we're at? Yeah. And I mean, like, and this, it's hard because AI has kind of accelerated that process. Like that process. I mean, and I totally agree with you. I've used that same analogy of like, like punch, punch cards versus like assembly versus like higher level languages, which is, which is actually like a good thing. but it's also like this the speed of evolution i i'm concerned about because you know we've had decades and maybe that's wrong of me to be like an old programmer and be like well you know i earned this you know over decades i mean maybe that's like myopic of me but at the same time we're still human beings and we live in human like structures and and and and and we live in a society yeah you know and i don't know if our industry and the people that live in it can i guess keep up with this interesting speed at which the industry is changing.

24:27And I wonder if it's going to be a boomerang or I wonder if it's just going to, we're going to be in a spot where we're going to have to, I don't like saying this, sacrifice some people's like current, like, I don't know, just we're going to be messing things up. Yeah.

24:40Huyen Tue Dao: Is it going to be an easy transition or a rough transition? It might be a rough transition. Yeah. Yeah. Because it's easy to look back at the eighties and say like all those people who they said how fast things were changing. It wasn't that fast and it was fine. Well, yeah, we could say that because we know how it turned out. Much more disconcerting to be sitting on the edge of the precipice going like, I think we'll survive the fall. We usually do. You know, that's a whole different situation. Yeah, I was literally apologizing to a new grad three weeks ago saying, I'm sorry. I know that we were putting, like all of us as an industry are putting you in a weird situation and I don't know the answer.

25:17Like I don't know how it's going to turn out. But I don't know. I just feel like some of us will just hopefully be there as a safety net. I don't know. We'll see. Hopefully it ends up okay. Well, yeah.

25:29Huyen Tue Dao: Gwen, thank you very much for giving us so much of your perspective on this. Like it is incredibly valuable. I can't thank you enough. I appreciate it. When I saw this on the run and I was like, yes, I have thoughts about this. But hopefully they're helpful. I think so. I think they really are. Okay. All right. Well, I guess we'll just move on now and say DTNS, including long rambles like this, were made possible by you, the listener. Thank you so much to Hitek Oki, Chris Zaragoza, and Jim Hart. Yeah, without folks like that, we don't have the ability to sit down and have these conversations. So thank you very much.

26:07Granola is an AI-powered notepad built for the way real people actually meet. Here's how it works. You take rough notes like you normally would. And in the background, Granola securely transcribes the meeting. Then, it turns everything into clean, structured, actually useful notes when the meeting ends. And the best part, Granola works through your device's audio, which means it integrates seamlessly into the video conferencing tools you already use. No setup, no awkward bots. It's just your normal meeting, with superpowers. You get to actually listen instead of frantically typing every word, and still walk away knowing exactly what was decided, who's doing what, and what comes next.

26:43My first time using it was in a meeting that had a lot of moving parts and was pretty technical. I couldn't believe how organized and complete Granola's notes were. If meetings are eating up your day, Granola is a no-brainer. You can try it totally free. Just head to granola.ai slash dailytech. That's granola.ai slash dailytech to get your time back. Try it for free at granola.ai slash dailytech. Episode prep gets messy when research, Slack threads, and script trackers are all in different places. That's where ChatGPT work comes in. Give it the goal and connect to the tools where work happens, like apps and files, and it can help pull the pieces into a draft you can review without you having to re-explain the project at every step.

27:25For example, ChatGPT work can gather team feedback from Slack and use it to suggest updates to a project brief in Google Docs. You stay in control reviewing the draft and deciding what moves forward. Put ChatGPT to work on your most ambitious ideas and projects. Get started at chatgpt.com by selecting work mode. Available on Plus and Pro plans.

27:50This is Troy Millings from Earn Your Leisure. On Earn Your Leisure, we're all about smart money moves. And here's one most people skip. And it starts with Insurify. Most of us never shop our car insurance. The policy renews every year. The rate creeps up a little at a time. And nobody calls you to tell you. You stay with the same company because it's easy. And that convenience costs you real money. That's what Insurify fixes. Real quotes from more than 120 top insurance companies side by side. About two minutes, free, no phone calls, no pressure. And if there's nothing better for you right now, Insurify keeps watching and tells you when that changes.

28:26So you're never stuck overpaying. Millions of people have used Insurify to shop for their car insurance. So insure with Insurify. Go to Insurify.com. That's I-N-S-U-R-I-F-Y.com. Insurify.com. This episode is brought to you by Capital One. Capital One's tech team isn't just talking about multi-agentic AI. They already deployed one. It's called Chat Concierge and it's simplifying car shopping. Using self-reflection and layered reasoning with live API checks, it doesn't just help buyers find a car they love. It helps schedule a test drive, get pre-approved for financing, and estimate trade-in value.

29:04Advanced, intuitive, and deployed. That's how they stack. That's technology at Capital One.

29:37at Range Rover dot com.

29:42Huyen Tue Dao: There's more we need to know today. Let's get to the briefs. All right. Well, Belkin announced Thursday that it will end support for its Wemo smart home products and the app used to control them on January 31st, 2026. Any products still under warranty at that date may be eligible for a partial refund. Products that can be configured to use with Apple HomeKit will continue to function. Affected devices were sold between August 2015 and November 2023. Now, listen, I'm the kind of person who's always like, wait, these stories aren't as bad as it sounds. Let me explain. This is older stuff. Not that many people have them.

30:18Huyen Tue Dao: This is not that story. This story sucks. This is Belkin. You know, when you buy Belkin, you're like, well, they aren't going to go out of business. And it's stuff that's only a couple of years old. I have bulbs. I have smart bulbs that I've been using for 15 years. You know, like this, this is, this is not great. I do not like this story. No, I'm pretty sure I have a Belkin smart plug somewhere in this room right now. So yes, it sucks. And I understand the decision, which is Belkin's just getting out of this business. And they're like, we'll support your warrant. We'll try to give you a refund if you're within your warranty.

30:54Huyen Tue Dao: But we're not going to be in this business. So we're not going to spend money supporting it anymore. I get it. But as a smart home user, and I don't have any Wemo devices. I went and checked really fast because I wasn't sure. I'm not going to be directly affected, but this is not good. And it's going to turn some people off of smart home devices, honestly. Apple published a preprint for a paper submitted to the International Conference on Machine Learning titled Beyond Sensor Data, Foundation Models of Behavioral Data from Wearables Improve Health Predictions. It describes a wearable behavioral model or WBM that it trained on behavioral data.

31:31Huyen Tue Dao: So step count, gait stability, mobility, VO2, stuff like that, compared to what is called a PPG model, which is based on like direct data, like heart rate or ECG. The behavioral WBM outperformed the PPG model in 28 of 57 health prediction tests. The WBM was better in all but one of the dynamic tests. These are things like detecting pregnancy, sleep quality, respiratory infection. The one dynamic test it was not as good at was diabetes. PPG was better at that one. So the headline isn't WBM means you don't need heart rate sensors. It's combining these models gave the most accurate results. So being able to pull in this behavioral data really does improve their ability to have some predictive health information for you.

32:23I love this. And generally, I totally jive with this because I'm a big health nut. And I also use, I presume, a lot of devices that probably use a PPG model. But as we're all adopting them as a society, you can kind of see that this, again, more anecdotal than anything else. But the small kind of like, oh, this doesn't quite make sense. Or like, you know, people are all different. And heart rate is kind of a very person-independent thing. So as you said, it's really just nice that we can have like this kind of not necessarily redundant, but complementary information to make all these things better.

32:52I'm so into it.

32:52Huyen Tue Dao: Yeah, make it more specific to you. All right. Well, Intel will spin out its RealSense robotics and biometrics business, launching it with a$50 million round of funding. Intel's vice president and general manager for incubation and disruptive innovation will become RealSense's CEO. Yeah, it's not probably going to change much for you, except maybe you'll see some more RealSense technology getting into stuff. uh samsung's coo wong june chui not to be confused with the pitcher for the doosan bears told cnn that samsung is exploring more wearable devices he said quote we believe it should be wearable something that you shouldn't carry that you don't need to carry so it could be something that you wear glasses earrings watches rings and sometimes maybe a necklace samsung has watches and a smart ring already and is working on glasses, including a product in partnership with Google.

33:46I love this and it makes a lot of sense. I think like things like, you know, big HDR or VR, VR headsets are wonderful, the Vision Pro, but I think like we're moving into that, that kind of section of technology. We're away from like the cool conceptual, but maybe practically, you know, infeasible devices to, hey, small stuff that I don't even have to think about that's on my body most of the day,

34:07Huyen Tue Dao: which makes so much sense. I don't know what a necklace would do for me necessarily, but I'm curious to find out what Samsung comes up with. Right. Yeah. The only thing I think it was like maybe for older folks, almost like that. What is that old commercial with the life alert where you. Oh, yeah. I've fallen and I can't get up. I can't get up. But but but better with maybe like calling emergency like that kind of thing. But like they don't have to say anything, you know, but you can do that. So but maybe they don't want to wear a watch all the time. Maybe the necklace. Yes. Well, yeah. Yeah. No, there's there's there's definitely some some space there for stuff.

34:37Huyen Tue Dao: I'm interested. Let us know. HMD Global, which has the license to make phones and tablets under the Nokia name, announced it will scale back its U.S. operations. The Verge reports that it appears to have stopped selling phones in the U.S. under the HMD and Nokia names. HMD says it will continue to support customers of its phones through its global support team. Yeah. HMD did not say it was because of tariffs, but it's because of tariffs.

35:08Huyen Tue Dao: Russian pro basketball player Daniel Kasatkin was arrested at De Gaulle Airport in Paris on charges of being a member of a ransomware gang. He's being held in extradition custody on the request of the U.S. Kasatkin played college ball at Penn State, plays for NBA Moscow now in pro ball. Kasatkin denies the allegations and his lawyer says he bought a used computer and added he's useless with computers and can't even install an application. Okay. Which, you know, I would maybe be insulted if I was Kasatkin, except that if that gets you out of this somehow, then sure, go ahead. Yeah, sure. What is that?

35:49What's the computer? What is this thing? Okay. Okay. Okay. Comcast announced that new customers can sign up for a service that does not include usage caps. There are four unlimited plans with speeds from 300 megabits per second to two gigabits per second. The unlimited plans are for those who use Comcast's Wi-Fi gateway, so you can't bring your own modem. Comcast also told reporters that it will offer the plans to existing customers as well.

36:18Huyen Tue Dao: You should read the Stephen Levy report in the Wired magazine talking about the clause in the agreement between Microsoft and OpenAI on superintelligence. The clause says that if OpenAI achieves artificial general intelligence, Microsoft would not have access to those models and the terms of their contract would become void. OpenAI defines AGI and its charter as a highly autonomous system that outperforms humans at most economically valuable work. The agreement also says Microsoft won't try to develop its own AGI during the terms of this agreement, which was probably fine back in 2019 when it was signed.

36:53Huyen Tue Dao: And AGI seemed a long way off. But now that OpenAI wants to renegotiate the deal and its CEO is out there saying, I don't know, we might get AGI this year. Microsoft wants to change the terms of that deal. So it's a good read. I'm just pointing people to it if you want to understand that relationship a little better. All right. Well, eSIM vendor Aerolo just completed a round of funding that values it at$1 billion, making it a unicorn. So unicorns are not as rare as they used to be, but Aerolo is an interesting one because it resells wireless data services to travelers. Aerolo was founded in Singapore.

37:30Its founder now lives in Dubai, and it has its headquarters in the U.S. state of Delaware.

37:35Huyen Tue Dao: I used Aerolo when I went to Japan. Oh, really? Yeah. It was super easy. You just on the app, you buy your eSIM, it installs it for you. It tells you how many gigabytes you have left at any given time. You could add credits super easy. Like I'm not surprised that it's getting valued as high as it is because it was a pretty good service when I used it. I love that too now. And I think especially as someone who's been traveling internationally for a while and the old style of like renting a whole little like portable Wi-Fi or just having SIMs. It's kind of, I feel like it crept up in me a little bit.

38:07Like, oh yeah, eSIMs are, you know, basically not, I'd say not standard, but, you know, it's increasingly available to people. And this is such an amazing thing. When I saw it, I was like, oh, that's not the sexiest unicorn I've ever seen. But it makes a lot of sense.

38:19Huyen Tue Dao: Yeah. It's a very practical unicorn. Folks, what do you want to hear us talk about on the show? One way to let us know is our subreddit. You can submit stories and vote on them over there at reddit.com slash r slash daily tech news show.

38:36This is Troy Millings from Earn Your Leisure. On Earn Your Leisure, we're all about smart money moves. And here's one most people skip. And it starts with Insurify. Most of us never shop our car insurance. The policy renews every year. The rate creeps up a little at a time. And nobody calls you to tell you. You stay with the same company because it's easy. And that convenience costs you real money. That's what Insurify fixes. Real quotes from more than 120 top insurance companies side by side. About two minutes, free, no phone calls, no pressure. And if there's nothing better for you right now, Insurify keeps watching and tells you when that changes so you're never stuck overpaying.

39:15Millions of people have used Insurify to shop for their car insurance. So insure with Insurify. Go to Insurify.com. That's I-N-S-U-R-I-F-Y.com.

39:25Huyen Tue Dao: Insurify.com. Being a serious business owner takes hard work, patience, and a drive to keep moving forward. So it's important to find a serious business card that goes the extra mile. With the VentureX Business Card from Capital One, you earn unlimited double miles on every purchase. And with big purchasing power, you can spend more and earn more. This is your sign to take your business to the next level. The Capital One VentureX Business Card. What's in your wallet? Terms apply. See CapitalOne.com for details.

39:58We end every episode of DT &S with some shared wisdom. Today, Derek is helping us understand.

40:04Huyen Tue Dao: Yeah, he has some thoughts in response to our discussion earlier in the week with Andy Beach about the WPP ad agency and its problems with AI. Derek writes, as one living in the ad world, it's been hard to escape the bigger news that rocked the industry in the past few months. WPP lost some of its largest clients to a competing agency, including Coca-Cola, Mars and Paramount. Heard of them? He didn't write heard of them. It's gotten so bad WPP is suing this other agency. Current economic and political instability also leading to more conservative and lower forecasts for ad agencies in the coming years.

40:38Huyen Tue Dao: That's not to say AI is not shaking up the advertising industry. It has already started to disrupt the industry across the board with advertisers, agencies, and media companies, but everyone's still figuring it out. If anything, advertisers have been looking to lean on agencies for guidance and education on how to navigate a potentially changing landscape? How does paid search work in a world where people are shifting over to AI-generated results, etc.? AI will definitely reshape the ad industry and force agencies to adapt, but the ad agency isn't going away quite yet, nor is it the central cause of WPP's current woes.

41:11Huyen Tue Dao: Thank you, Derek. Appreciate that information, that insight. All right. Well, what are you thinking about? Do you have some insight into a story? Please share it with us. Feedback at dailytechnewsshow.com. Thanks to Derek for contributing today's show. Thank you for being along for Daily Tech News Show. You can keep us in business by becoming a patron. If you're not already, please do. Patreon.com slash DTNS.

41:41Huyen Tue Dao: This week's episodes of Daily Tech News Show Briefing were created by the following people. Host, producer, and writer Tom Merritt. Host and writer Jason Howell. Co-host Rob Dunwood. Co-host Jen Cutter, co-host Wen Tui Dao, producer Anthony Lemos, producer Roger Chang, editor Hammond Chamberlain, editor Victor Bognat, science correspondent Dr. Nikki Ackermans, social media producer and moderator Zoe Detterding, our mods, Beatmaster, WSGOTAS1, BioCow, Captain Kipper, Steve Guadurama, Paul Reese, Matthew J. Stevens, a.k.a. Gadget Virtuoso, and J.D. Galloway. Mod and video hosting by Dan Christensen.

42:14Huyen Tue Dao: Music provided by Martin Bell and Dan Luters. Art by Len Peralta. Acast Ed support from Tatiana Matias. Patreon support from Bobby Wagner. Our guest this week was Andy Beach. And thanks to all the patrons who make the show possible.

42:31This show is part of the Frog Pants Network. Get more at frogpants.com.

42:38Huyen Tue Dao: Diamond Club hopes you have enjoyed this broker.

43:07get pre-approved for financing, and estimate trade in value. Advanced, intuitive, and deployed. That's how they stack. That's technology at Capital One. This episode is sponsored by Aerie. If you know me, you know that my philosophy is all about looking good and feeling good every single day. And that honestly comes down to the first layer. I'm talking about bras. If your bra is digging into your shoulder, or maybe it's riding up in the back, or you are dying to take it off at the end of the day, then that bra is not the right fit. And what I love about Aerie is that they have 59 different sizes in really soft fabrics and really good fits.

43:44Because at the end of the day, your bra should hold you in and not hold you back. Visit Aerie.com to find the right bra and your perfect fit.