AI Gets HIPAA With HealthCare - DTNS 5183

0:03this is the daily tech news for monday january 12th 2026 we tell you what you need to know give you the important context and try to help each other understand today ai companies move farther into healthcare will give you the information you require to stay healthy and safe i'm what's left of tom merit after ces and i'm deep voice rob dunwood let's start with what you need to know with that big story uh lots of stories today around companies using new generative models in the healthcare industry let's start with some research and development nvidia nvidia continues to place a lot of bets around multiple industries we talked at ces about robotics as being a big one for them it wants to have other avenues for success places to sell its products to that end nvidia plans to spend around a billion dollars over the next five years to create a laboratory in the san francisco area with pharmaceutical company eli lily to help engineers understand the needs of the pharmaceutical industry and also help eli lily fine-tune models for use in developing medicine i'm going to guess you won't see anybody call this a circular uh problem but it is the same thing giving a company a billion dollars to then use it to buy a bunch of nvidia chips to create a lab uh to develop uses for generative models in healthcare meanwhile openai followed up on its announcement of a tab for health in chat gpt with a new system called OpenAI for Healthcare.

1:32This is a set of products for healthcare organizations that implement HIPAA compliance, something we talked about on Friday, Rob. You asked and they answered. The products include some workplace automation tools for paperwork, basically helping you file forms and keep track of stuff. It's some data control and of course that HIPAA support, role-based access management, and chat GPT for healthcare is in there too. Models optimized for clinical research, evidence-backed responses. Basically, they cite peer-reviewed studies. They cite clinical guidelines and public health sources in their answers as a way to prevent hallucinations.

2:12They're what I often think are the more useful sorts of these models. They're limited to a certain data set. Anthropics CLAWD is also launching CLAWD for healthcare. This is also a HIPAA-compliant set of tools, tuned a little bit more towards the paperwork side of things, claims and authorizations. Tools for helping plan clinical trials are in there as well, though, and manage the regulatory filings for those trials. And there is a consumer-facing part of this. Claude can also connect with HealthX, Apple HealthKit, and Android Health to interpret test results and medical history for patients. Those are all kind of the good news side of this.

2:49On the other side of this, Google has removed some, or at least two, of its health-related responses from the AI overview search results. This was in response to an investigation by the Guardian, which found that it was giving incorrect recommendations for liver health. So Google stopped serving AI overviews for the search terms or the query, what is the normal range for liver blood tests and what is the normal range for liver function tests? Now, we could definitely spend some time debating whether Google has done enough or whether this is just trying to combat the press, the PR hit they took for this.

3:26But I thought, Rob, that the other parts of this were pretty optimistic. I believe they are. Because when you think about artificial intelligence and what large language models are really good at, they're good at taking a lot of data and comparing it and then giving you an average of what that data is. So one of the problems that dockers have is that they have finite memory. They can only read so much, you know, peer review data that, you know, a lot of times they're working off the best guess. But if you put all that information into a large language model, it can make in many cases much better informed guesses based off of looking at hundreds of thousands, if not millions of pieces of data as compared to looking at dozens to hundreds that doctors actually can put into their mind and remember.

4:14Yeah. And instead of me having to go to a chat GPT and paste my lab core form in there and cross my fingers and hope that that data is never used against me in the future. Having something that's in a HIPAA compliant system that my doctor says, hey, you can ask Claude about this through our standard system that you already use and then get some natural language responses. I really like that. I really like that being something offered to people through their normal healthcare roles. Yeah. The more information people have about their own healthcare, the better their healthcare tends to be. And the fact that you can just go and type questions in, and it's based off of information you've already given your doctor, based off tests that you've already done, it just helps you know more about what's going on with your health.

5:05So I've always said, since we first started hearing about ChatGPT, you know, three years ago, healthcare is where I think this thing can really make its mark, or really make a mark, because making this mark everywhere. But I think that healthcare is one of those places just because a lot of what doctors do is just, well, we think you have this based off of what we've seen before. If you can have more things that are seen, then you can get better outcomes from that. The large language models are very good at doing that. Now, on the Google side of it, you got to make sure that those hallucinations, this is people's health.

5:38You don't want to give improper data. But I think that this is a place where they can absolutely get it right. And I think it would benefit millions. Yeah, I think these are good stories to compare because of that Google story, where if you are just out on the wild web using a model that is not particularly tuned to anything in particular, you are likely to get improper results. and that can be a bad thing. Whereas if you are in a model that is specifically designed and tuned for an industry, has compliance to its regulations, is overseen by a doctor that's going to review it and say, well, hold on, why is it doing this?

6:17Give feedback to the model. I have much more confidence in that. And I really, really, really want my doctor to spend less time filling out reports and more time working on my health. Absolutely. And when we first started hearing this stuff back last week at CES, my wish was, you know, this is really cool. I wish they were just making HIPAA compliant. And now it's HIPAA compliant. I think it was a day later after having that conversation that we start to see these things are coming out and they are HIPAA compliant. So you can be as safe or, you know, your data is as safe as what HIPAA compliant allows for it to be.

6:52Yes. Good way of putting it. and like I said, you know, stuff could stuff happen. Yes. But that's always the case with data. It could happen with your, your data currently. So it's as safe as it can be based under HIPAA compliance. Yeah. HIPAA is not like, like some magic spell or something, but it, but it's helpful. DTNest is made possible by you, the listener. Thanks to Justin Zellers, Carmen Bailey, Chris Benetow, our new patrons, Joseph, Brian, Ramen, Deep, Kirk, and Alexander. We've got razors from Craig and Roman and an annual plan update by Alan. Hey, thanks everybody to all the new folks in the new year joining us.

7:29Really appreciate that. Catching up on a bunch of you joining over the CES week because we're a little distracted there, but it's good to have everybody, every single one of you. Thank you very much. All right, let's get to the briefs. A few bits of Apple news to note today. Apple and Google announced a partnership to use Gemini models to power some features based on Apple Foundation models, including an upgrade to Apple's voice assistant Siri later this year. Meanwhile, CounterPoint Research estimates say that Apple shipped the most smartphone units of any company in 2025. Overall, smartphone shipments grew 2 % in 2025 and Apple grew its shipments to 10 % to take 20 % of the global market.

8:09Samsung grew 5 % and finished with 10 % of the market. Xiaomi, Vivo, and Apple took the next three slots, with nothing and Google remaining outside the top five, but posting 31 % and 25 % growth for the year. Meanwhile, Apple Senior VP of Services, Eddie Q, posted a note Monday to announce that Apple Services broke records in 2025, with total hours viewed on Apple TV, up 36 % on the year in December, and Apple Music setting records for listenership and new subscribers. Yeah, I wrote our script wrong. It's 19 % of the market for Samsung. So it's right there behind Apple. But good news overall for Apple, I think it's less good news that they have to rely on Google.

8:52But I think good news for us as consumers that they are not going to stand by and use an inferior model. This is somewhat the same as Google using, or using Google search and Safari instead of Apple creating its own. And maybe that will be fine. Maybe that will be better, and maybe Apple will use this as a way to get the time to make its foundation models more capable as well, the way they did with Apple Maps. Remember, Google Maps used to be the maps for the iPhone, and Apple's first launch of Apple Maps was very unsatisfactory, but Apple Maps is quite competent now, and in some places seen as a better alternative.

9:35So I think that's good news for Apple iPhone users. And then the fact that Apple is able to say like, hey, in developing markets, we are shipping more. I think sometimes you jump to the conclusion that, well, Apple's more expensive, so it won't do as well in those markets. It's those markets that help propel it to this global market share. And of course, services is where they see their future. So I guess that's why Eddie Q is out there touting like, hey, don't worry about Apple TV. It's doing fine. There's some good stuff on Apple TV. I find myself watching more and more of it. As far as Gemini is concerned, I think Gemini is pretty good.

10:10It's if it's not the best model, it's in it's in the top two or three. It is it is very good. So Apple deciding to to run with Gemini kind of makes sense. If you've been if you've been reading the tea leaves, we kind of thought that this is where we were going to ultimately get. But it makes more sense for Apple to say, let's partner with a company that is doing it very, very well. We'll put some of our apple juice on it and make it our own. But let's partner with a good model out there and not be lagging behind everyone else trying to get our own model up that just isn't up to par with some of these other models.

10:44Yeah, I agree. I think Gemini is probably one of the better ones right now. But to your point, it's in the top three. You're going to have good uses for ChatGPT, for Claude, for Gemini, depending on what you're using it for. and they're going to keep jumping ahead of each other. The next release of one is going to put it ahead of the other two. And don't count out the mistrals and others out there as well that do things differently. But it's not like Gemini is settling for RC Cola here, right? And no disrespect to RC Cola, it's quite tasty. Now, the thing that's going to be interesting to me is that this kind of a little bit feels like how Apple is using Google for its search.

11:28You know, it made Google the default search. They got paid by Google for that. This isn't necessarily the same. I think Apple, in some instances, is actually paying Google for this. It's the opposite. So it'll be interesting to see how that relationship continues, you know, going forward. Scientists led by Eiji Watanabe, an associate professor of neurophysiology at the National Institute for Basic Biology in Japan, have been studying whether deep neural networks could replicate what happens in the human brain when we see optical illusions. This is just one of those like, huh, interesting to understand how the human mind works and also understand how these neural networks differ from how we work.

12:10Do you know about the snake circle, the rotating snakes illusion? Have you ever run into that before the snake is like eating itself as it as it spins around yeah it's not just the it's not just the it can it can be other things besides the snake eating itself but yes yes that's the one where you look at it and it looks like it's moving but it's not actually moving it's just a static image so they used a model called prednet which is trained to predict future frames in a video, hence the Pred for prediction, and based on previous frames it's seen. So it gets trained on seeing things and then tries to predict what the next frame would be.

12:51Watanabe's team trained it on natural landscapes. So they did not train it on pictures of snakes in circles. They just gave it natural landscapes. PredNet was then shown the rotating snakes illusion, and it was tricked by all the same ones we are. There are certain versions of the illusion that you look at and like that one doesn't look like it's moving to me. So PredNet was not tricked by those, but it was tricked by all the ones that we see as moving. The one difference they found, I found this interesting, is when we fix our focus on one of the circles in that illusion, that one will stop and the others will kind of in our peripheral vision continue to move.

13:32PredNet always saw the circles as moving. It didn't matter whether they instructed it to focus on one or the other. This is really interesting. When we are building, basically, these are computers that are looking at things and are perceiving them in very, very similar ways to we as humans perceive them. and they're getting tricked just like we get tricked you know i i do wonder why you know what is the difference when we actually you know when we focus on the one circle and it stops moving that doesn't happen for uh you know you know for these models i wonder you know what that could be and is that is that is that a deeper uh a deeper feeling of of perception for us as far as you know our level of focus are we able to focus more so than what these computers are they see everything so they still see everything.

14:24I don't know how that works, but it is interesting work that they're doing here. Yeah, and that exact question that you're bringing up, I think is worth the exploration to find out like, oh, okay, so we sometimes say, well, neural networks are basically just versions of our own brain, but they're not. They're close, but there are differences. And here's an example of one of those differences and getting to that answer to your question, why is that, I think will be very instructive. both on how our brains work, but also how these neural networks differ from how our brains work. Microsoft is retiring lots of things to start the new year and to caught our eye.

15:05The Microsoft Lens app for iOS and Android will be removed from stores on February 9th, 2026, and scanning will stop on March 9th, 2026. Users will switch to OneDrive scan feature. This was supposed to go away in mid-September, but was delayed. And starting in February, Microsoft will remove the option in Microsoft Word that lets you send a document to Amazon Kindle. Microsoft recommends using Amazon's Send to Kindle website. Now, the Lens app, I sort of get, which is like, that's just a feature of the OneDrive app now. So they'd rather you just use the OneDrive app. And that kind of makes sense to me.

15:38But the Kindle thing, I guess not enough people were using it, right? Never say nobody because somebody out there listening right now was like, I use that all the time. But I was just going to say somebody is very, very upset about this, but I imagine that it's just a few people because there's just other ways to do it. And I honestly, I've done it before, but I didn't know you could do it from within from within Word. I thought you just went to the website and just say, hey, send this to, you know, send this to Kindle. So I would just grab that send a Kindle email address, find the Microsoft Word file, attach it and send it that way.

16:10I didn't even, I wouldn't even think about going into the Word document and opening it up, but I guess that's probably why they're That's how I thought it worked. Yeah. Wall Street Journal talked to memory makers. These aren't people who implant memories in your head, but, you know, folks who make RAM and SD drives and all of that. They talked to memory makers about their plans to increase capacity to build memory given the surging demand for its data centers. This is an essential question when we're looking at the rising price of RAM and solid state drives. hey y 'all gonna build some more factories to make more of these so that you don't have an under supply seagate plans to increase its expenditures the most because it maintains a policy of spending four percent of revenue on capital expenditures sandisk will raise its capex 18 gave us a hard number there but citing a law a lack of long-term supply agreements for nand flash memory as a reason it's not going to spend even more.

17:11Micron's CapEx spend will rise less. NAND flash chip factories can take years to build, and they don't want to build them, and by the time they're done, they don't need them anymore. Sandus Chief Executive David Geckler told the UBS conference last month, perhaps the demand side should think about making commitments that are longer than three months at a time. This is the classic problem of are we in a bubble or aren't we in the bubble? They know that if it's going to take us multiple years to build these plants and we start building them, we've got to build them. But will the supply be the same?

17:44Will the bubble have burst by the time these facilities go online? So if I am the memory maker, I'm very concerned about this because it's not like they're just throwing a warehouse up. These are very, very high-end manufacturing facilities that take a lot of work to go into them. So I would absolutely, if I were one, I would want a longer term commitment. But if you are on the other side of it, well, I don't want to do a longer term commitment because I think we're in a bubble as well. It is probably going to be at some point. So it's kind of like they're in this circular dance of, you know, who's going to be, you know, are we going to give you a long term contract or are you going to build a factory where maybe once you have the factory built, we'll give you a long term contract.

18:25I'm going to need all the RAM you can build for the next two years. Great. Can you put that in your contract? Well, I don't know. I think it's also worth noting here. It's tempting, and I'm sure some of you are thinking of this, to say, well, these memory makers just want to keep those prices high so they can make more money. And the fact is that won't make them more money because they are raising the price because they can't make as many as demand makes. If they could easily make more, they do it in a heartbeat because selling more units, even at a lower price, will make them even more money.

19:09So if you're attributing this to greed, the greedy move would be to try to build factories as fast as possible and get as many of these made. But like Rob just explained, there's a worry of like, well, But by the time we do that, do we not need it anymore? So it's not about just keeping the price high. This is about estimating how long that demand is going to stay high. Absolutely. Memory is not like a Bentley Rolls Royce or like a Maybach to where they are literally niche made. We're going to make this car for you. It's something that they mass produce, but it takes a factory that takes, you know, a lot to go into that factory to get it done.

19:51So they just want to make sure that if we put this investment out, the bubble hasn't burst and you are still going to pay us because of this thing that you've done. It makes perfect sense. I understand where everyone is on both sides of this. So we'll see what happens. If any one of these companies or a smaller company can figure out how to increase their capacity within the three month contract timeframe that everybody wants to sell, they will do it. Oh, absolutely. I'll be interested to see if somebody does. Finally, some quick headlines that are just good to know if you want to understand the news as you go along in the future.

20:24WCCF Tech has seen a series of Cinebench and multi-core benchmark comparisons of new Snapdragon X2 chips against the Apple M4. The M4 beat X2 Elite Extreme in four out of five synthetic CPU and GPU tests. Good news for Apple, I guess. In release notes for his holiday coding project, Linus Torvalds, founder of Linux, said he used Google Antigravity to vibe code the visualizer tool of his audio noise guitar pedal related audio effects app. Linus always does like a holiday project. This was his holiday project this year. If you want to take a look at it, though, you can find audio noise on GitHub.

21:05Speaking of Linux, a fix for NVIDIA GPUs on Noval driver made it into Release Candidate 5 of the Linux 6.19 kernel. A stable release is expected around February 8th. The most popular app on the App Store this past week in China was an app whose title translates as Are You Dead? The idea of the app is you push a button every day, and if it goes two days without you pushing the button, it automatically messages an emergency contact. Hmm. Okay. It says a lot about the mentality of people in China right now. I guess so. UK media regulator Ofcom has begun a formal investigation into X's chat box, GROC, over some users creating sexualized deepfakes of women and children.

21:50Meta announced that in Australia over the past month, it has closed 330 ,000 Instagram accounts, 173 ,000 Facebook accounts, and almost 40 ,000 threat accounts believed to be held by people younger than 16. This is in compliance with the new Australian law. Yeah, they're not messing around. They really meant it. Alphabet Swing announced an expansion of drone delivery service areas for Walmart to 150 more stores, including Cincinnati, Los Angeles, Miami, and St. Louis, with more areas to be announced. I wonder if it's going to come to my Walmart. I got to keep an eye out on that. Google announced the Universal Commerce Protocol, or UCP, for agent-based shopping.

22:31This was developed along with retailers. So Shopify, Etsy, Wayfair, Target, Walmart all participated in this. The idea is to make it easier for agents to talk to retail systems and then speed up the process on your end and make the process smoother on their end. Meta has hired Dina Powell McCormick as president and vice chairman. She leaves Merchant Bank BDT and MSD Partners. She previously worked for 16 years at Goldman Sachs and served as U.S. Deputy National Security Advisor from 2017 to 2018. and Assistant U.S. Secretary of State from 2005 to 2007. And finally, Amazon getting back into making physical stores.

23:09Kind of looked like they were pulling back from that for a while, but now we know what their strategy was. Amazon plans to build a 229 ,000-square-foot big-box store in Orland Park, a suburb of Chicago, to sell groceries and general merchandise. So previously, Amazon just had Amazon Fresh Grocery Stores, and they still have a bunch of those. but this will be interesting because this sounds more like a Walmart where you combine the groceries and the general merchandise. I would say 230 ,000 square feet almost. That is a big store. That's not a little shop on the corner. That's a big store. Well, folks, do you like hearing about these stories on the show?

23:47You want to let us know a story that you think we should talk about? Maybe get some feedback from other listeners on the show to find out if they would like to hear us talk about it. You can do all that in our subreddit. Submit stories and vote on them at reddit.com slash r slash daily tech news show. We end every episode of VT Nestles from shared perspectives. Today, Mark has a trick for installing your smart lights. You're going to like this one. It's about Govi. Mark was hearing us talk about, I think it was Sarah was talking about having to get on the right Wi-Fi network. This was in our troubleshooting episode that we did over the holiday break.

24:23Mark wrote, here's another 2.4G 5G hack to add to your list. I think gigahertz, not like cell phone 5G. I recently installed Govi lights in our home. You program the lights and you must have 2.4 gigahertz Wi-Fi going to do it. But we have the most recent Nest Wi-Fi router. I saw online a trick that if you grab your phone and watch the Wi-Fi connection, it will drop down to 2.4 gigahertz at a certain distance. Once the phone connects, I could see the Govi controller. Once configured, everything works fine. So if you're a little confused by this, sometimes smart lights or smart home stuff won't let you set it up over a five gigahertz connection.

25:10They need a 2.4 gigahertz connection. And so you have to go into your router and manipulate it and change it. And maybe your router doesn't have that. And then you're in a whole other world. And what Mark was pointing out is sometimes at a certain distance, they just drop down to 2.4 gigahertz. And so you just go to that distance, connect the app for the smart appliance, and then you're good. Yeah, that is old school. It's like, you know, I have back in the day, actually, let's really test how long we can make a patch cord. Can we really, can we make it as compared to just 100? So, you know, this is one of those things to where it actually works.

25:48Go beyond the range, let it drop down, configure it there, then bring it back in. It's already configured. Yeah. That's fantastic, Mark. Thanks for passing that along. What are you thinking about? Got some insights into a story? Share it with us at feedback at dailytechnewsshow.com. Yes. Thank you, Mark, for contributing to today's show. Thanks, everybody. You listening right now for being along for Daily Tech News Show. You keep us in business, folks. Become a patron. You know, we got to pay for all those rides around Las Vegas covered in CES. So join that up right now, won't you? Patreon.com slash DTNS.

26:28The DTNS family of podcasts. Helping each other understand. Diamond Club hopes you have enjoyed this program.