Inside the 1930s Vintage Language Model Called Talkie - DTNS 5257

0:00This 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. Advanced, intuitive, and deployed. That's how they stack. That's technology at Capital One. Hannah, I just Venmo'd you for dinner. Obsessed. I'm literally spending it right now on the lip gloss that's been sitting in my cart.

0:35What do you mean spending it right now? You instantly spend your balance with the Venmo debit card. Stop. Say more. More exactly. The more you do with Venmo, the more you get, like earning up to 5 % cash back with Venmo Stash. Get the Venmo debit card or check out online. Venmo Stash bundle terms and exclusions apply. Max$100 per month. See terms at venmo.me slash stash terms. Venmo checkout, not available at all merchants. The Venmo MasterCard is issued by the Bancorp Bank N.A. Make Mother's Day even more special at Whole Foods Market. Kick off brunch or dinner with quality cheese and charcuterie with no synthetic nitrates.

1:14Then go seafood. There's an abundance on sale at Whole Foods Market, where it's all sustainable while caught or responsibly farmed. At the bakery, grab seasonal treats like their strawberry pretzel cream pie, and you can't go wrong with the ready-to-heat quiche Lorraine, deviled eggs, and fresh-cut fruits to go. Celebrate mom with Whole Foods Market. Parles-tu français? Hablas espagnol? Parli italiano? If you've used Babbel, you would. Babbel's conversation-based technique teaches you useful words and phrases to get you speaking quickly about the things you actually talk about in the real world.

1:48With lessons handcrafted by over 200 language experts and voiced by real native speakers, Babbel is like having a private tutor in your pocket. Start speaking with Babbel today. Get up to 55 % off your Babbel subscription right now at babbel.com slash acast. Spelled B-A-B-B-E-L dot com slash acast. Rules and restrictions may apply.

2:23This is the Daily Tech News for Tuesday, April 28th, 2026. We tell you what you need to know, give you the important context, and help each other understand. And today, a large language model from the year 1930. 30, 30, 30. This language model talks like this. Yes. That's right. I'm Jason Howell. I'm Tom Merritt. I can't wait to talk about this week's big story. Let's get to it.

2:53And I mean, define big, right? Like it's just one of those days where there was no like premier marquee story. So of course, we're going to talk about a large language model from the 1930s, sort of. It's big. It's really big, big and big. It's 13 billion big. It's Taki 1930, and it's a new 13 billion open weight LLM, or in this case, it's actually called a VLM, stands for Vintage Language Model. It's trained on 260 billion tokens of pre-1931 English books, newspapers, journals, patents, case law. But the important thing there is that it has a historical cutoff of December 31st, 1930, which is actually the cutoff for works to enter the public domain.

3:48So, you know, it wasn't just chosen randomly. That basically means that all of this content that's fed into it is legal to use in this capacity. It's developed by a nonprofit team that is using the model to look basically to explore a few key research angles. The first one is because Taki only saw pre-1931 text, it essentially avoids benchmark contamination and will, as a result, better show how much LM's language models can actually generalize beyond their training data. Secondly, by checking how surprised Taki is by the New York Times on this day events, so they're feeding them these on this day events that happen in the future after 1930, researchers can then measure how well a language model can forecast future outcomes as it moves past or beyond its knowledge cutoff date.

4:49So it's a good way to kind of explore that. How prescient are these things if given a cutoff like that? And then third, since Taki never trained on web data because the web wasn't around in 1930, it actually acts as a control group for studying which LLM behaviors come from language modeling itself and which come from being exposed to the modern Internet. This was one of those stories that when I saw it, I was like, well, the minute I saw it, I was like, I'm actually kind of surprised that we haven't seen this sooner because this makes a lot of sense. Like if we have questions about how language models operate and because they're so steeped in web data, because this is the world that we live in, this is the world that they were created in and everything.

5:35What would it be like to create one of these, but only, you know, cut off from the Internet, cut off from that constant feed of modern information feeding into it and only feed it this wide corpus of information and knowledge from a long time ago? And that can tell us a lot about how good these things are about telling the future and a whole bunch of other things. It's fascinating. Yeah. It's a great example of basically having a control group, right? Yeah. The control is like, let's not let it know anything from modern era and see. Like, for instance, they gave it Python and had to explain how Python worked and said, can we get it to program Python?

6:13And the fact was they could. It was not as good as modern models. But when they sort of controlled for the fact that the modern models know what Python is and how programming languages work, it was pretty decent at a few aspects of it. And so they're confident that if they can get these models improved, make them like GPT-3 or 3.5 level, which they're not quite there yet, that they'll be able to tell for sure where those barriers are. Jason, I do have to take an aside real quick. You made the modern podcaster faux pas of saying, why didn't we hear about this earlier? Which is, of course, going to make someone email us and say, well, obviously, you don't know about rank 4B, Mr.

6:59Chatterbox or Machina Mirabilis. Why didn't you know about those? But no, and I say that teasingly, but this is an area of research that has been ongoing for a long time. Yeah. And so Taki, I don't know why it's getting more attention today, maybe because they've got the chatterbox up there that you can watch happen. Well, that's right. I didn't even mention that. Yeah, yeah. Kind of like clue in and see how it's being used. Yeah. And maybe because it is the most capable that they've been able to do. One of the gates to being able to do this, one of the things that keeps you being able to do this easily is the data.

7:38It's not easy to find pre-1930 data because pre-1930 data had to be digitized. There's no native digital data from before 1930, which means you're at the mercy of optical character recognition. And what they found is that it's much better to do human transcription than to do OCR. So they're working on better OCR to make that happen at more scale so they can have more training data. And also there is contamination. Remember, you're not looking at each piece of work that you put into TrainTalkie before you put it in. You're trying to find it at large amounts, like go to the Internet Archive Project and say, give me everything from before 1930.

8:19But a lot of times there's going to be mistaken metadata, or even if you can get past that, there's annotations that were made past 1930 written into the book that make it into the data. I guess this is the 13th version of Taki. The seventh version of Taki knew about Franklin Delano Roosevelt, and it knew about World War II. And they're like, it shouldn't. It shouldn't know that. So data contamination where stuff from the future, as far as this model is concerned, works its way in is something that they're having to fight as well. Anyway, all of that means the data set stays smaller. Right. And so the smaller the data set, the less advanced that you can make the model.

9:05But they're working on that. They're working on ways to increase that data set and make it work better. and they're just now getting to parity with ChatGPT's original model, like GPT 3.5. And so, of course, they're making slower progress. But like you said, I think this is fascinating too, that we would have the ability to kind of beta test and find out, can the model, when it's trained on old knowledge, develop new knowledge? Can it actually discover things that were discovered in the 50s and 60s on its own? That, I think, is the real interesting thing here because in the last couple of years, there's been so much kind of passion and drive around, oh, isn't it amazing that we have these tools and just think of what they're going to be capable of.

9:50They're going to be able to solve this and this and this in the future. And I'm not saying that that isn't possible. I think there's a lot that these things are going to be able to do that we still don't know even how they're going to be able to accomplish these things in the future as they get built up and more complicated and complex. and people learn how to train them better or whatever the case may be, they're probably gonna do some really great things down the line. But this is just a really great way to say like, okay, so if we cut off the data point in 1930 and then we know that this happened 10 years later, is there anything here in this controlled model that could have potentially led to that same outcome?

10:30And I don't know, I just find that a really kind of cool kind of experiment to explore and see like, oh yeah, it would have gotten there. Or no, humans just think about this in a very novel, different way than a system like this might be capable. It might come to different conclusions. It might lead to different destinations that we didn't consider. But I don't know. I just think it's a fascinating way to kind of shed light on some of the promises that are being made about today's models. And it's fascinating. Yeah. And it's a really interesting examination of biases. You're going to have very different biases as they warn you on the Taki site in a 1930 and before model than you are in the models that we have now.

11:15Now, some of those are going to be offensive, right? That's where everybody's mind goes. And that's true. But there are also going to be things that you find it do that you're like, oh, that doesn't happen in the modern models. So I think that's fascinating as well. And it's all public domain. So there's no worries about the ethics of the training set. And what you can make this do will show you whether you can do that without having to get proprietary data. Okay. So now I want to take all the recorded music from 1900s to 1930s. I want to feed it into a new music generation system and see what that system, what new musical genre that system creates with just that limited training set to see.

12:03I don't know if there's a direct line that you could draw between what this model could actually inform in that regard versus something like music creation or art. Yeah, exactly. Music is art, of course, but I mean like painting and imagery and yeah, yeah, yeah. So I'd be curious about that. No, that is a fascinating thought. Yeah. Anyway, Taki, you know, a nonprofit group of researchers putting it together. It is not peer reviewed, but you can try it. If you go to Taki-LM.com slash chat, you get cued. Usually I was about 20 or 22 in the queue when I asked it stuff, but you can ask it things. It correctly identified my great grandfather.

12:41It cannot tell me more than one line about Lady Lovelace, and I'm not sure why. It just gets halfway through a sentence and then stops. But yeah, you can try it out there if you would like. I wonder what that's all about. Yeah. Daily Tech News Show is made possible by you, the human listener. Big thanks to every single one of you, including Michelle Sergiu, Kirk Stephenson, Miranda Janelle, and Paul Franz, among so many others. Thank you so much. Yes, indeed.

13:14I'll be honest. Nothing tests my patience quite like sitting on hold, listening to the same four bars of music on repeat. Turns out, all that time adds up. The average person spends over 40 days listening to that dreaded hold music. 40 days! This episode is sponsored by Parloa, the AI platform built to make that a thing of the past. Parloa's AI agents are available 24 hours a day, seven days a week, fluent in any language, and smart enough to remember every customer across every interaction because no one should have to explain themselves twice. From payment processing to roadside assistance, flight rebookings to appointment scheduling, Parloa delivers the kind of customer experience that turns a one-time buyer into a customer for life.

13:57See Parloa's AI agents in action at parloa.com. That's P-A-R-L-O-A dot 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. Advanced, intuitive, and deployed. That's how they stack. That's technology at Capital One. Shipping, billing, admin, payroll, marketing.

14:39You're managing all the things. So why waste time sending important documents the old-fashioned way? Mail and ship when you want, how you want with Stamps.com. Print postage on demand 24-7 and schedule pickups from your office or home. Save up to 90 % with automated rate shopping. That's why over 1 million small businesses trust Stamps.com. Go to Stamps.com and use code PODCAST to try Stamps.com risk-free for 60 days. If your eyes are the windows to your soul and your glasses are the windows to your eyes, then it's pretty important to find your perfect frames. That's why at Warby Parker, we've made shopping for eyewear as easy and fun as can be.

15:19Peruse endless styles in our stores or use our app to virtually try on frames and get personalized recommendations. To find your next favorite pair of glasses, sunglasses, or contact lenses, or to locate your nearest Warby Parker store, head over to WarbyParker.com. That's WarbyParker.com. Hannah, I just Venmo'd you for dinner. Obsessed. I'm literally spending it right now on the lip gloss that's been sitting in my cart. What do you mean spending it right now? You instantly spend your balance with the Venmo debit card. Stop. Say more. More exactly. The more you do with Venmo, the more you get, like earning up to 5 % cash back with Venmo Stash.

15:53Get the Venmo debit card or check out online. Venmo Stash bundle terms and exclusions apply. Max$100 per month. See terms at venmo.me slash stash terms. Venmo checkout not available at all merchants. The Venmo MasterCard is issued by the Bancorp Bank NA.

16:14All right, there's more we need to know today. Let's get to the briefs. Wall Street Journal reports that OpenAI has missed some internal targets. They wanted a billion new users by the end of last year. They missed that target. A yearly revenue target for ChatGPT was missed and several monthly revenue targets this year have not been met. CFO Sarah Fryer is reportedly concerned about paying for future contracts if revenue does not grow fast enough, which is another way of saying Sarah Fryer is their chief financial officer. Right. I don't know. This is kind of what she does. Yeah. It's like, I haven't met too many CFOs that aren't concerned about paying for future things if revenue does not grow fast enough.

16:54However, Wall Street Journal does say that the Codex tool is growing quickly and becoming more popular. This all could explain why we're seeing them shift more into Codex and why they retired Sora. And Sam Altman and the CFO Fryer issued a joint statement saying, and I quote, we are totally aligned on buying as much compute as we can and working hard on it together every day. This I think could be overblown as like OpenAI is in trouble. It sounds to me like OpenAI is a maturing business that still needs to figure out its way forward and has decided coding, the popular part, is the way forward.

17:30So that's kind of my take on this. Everybody needs compute in order for this to grow. So I'm not surprised that they're saying, hey, we're still buying as much compute as we can. But trying to figure out how to keep that airplane flying while you change the engine is difficult. Yep. Yep. And OpenAI has, of course, publicly pushed back on the Wall Street Journal report. They're basically calling it - the joint statement saying we're totally aligned on buying the compute i know they they threw some clickbait they said some names like clickbait i don't know that that helps us understand the story more that they like called it names uh i also don't think the wall street journal was writing this for clickbait do you no no that's true that's true um and of course they're going to push back on this whether you know yeah because that's not internal reports what it's internal leaky stuff and everything yeah i mean whether whether open ai is in a rough spot or not um i think continually the the conversation that i continue to hear i hear from jeff jarvis on on the ai inside podcast and other directions continues to be just this general sense that you know even a year ago even a couple of years ago open ai was that they were the you know that whatever, the steamroller, they were the engine that was really kind of whether they actually were or not, but that many people saw as the model of like AI success, right?

18:57Open AI, early to market, you know, to really capture on the LLM trend, let's say, and, you know, lots of moving pieces happening and big time success in the last couple of years, and especially the last year, that narrative, whether it's true or not, has shifted to, okay, wait a minute, maybe they came out a little bit further ahead on their skis than they're able to. Now we've got Gemini doing some pretty incredible things this past year. Anthropics certainly has some real strong momentum right now showing that coding and enterprise is a huge growth area and that's where they're putting their efforts and now OpenAI is coming on over.

19:39So whether it's true or not, there's certainly that perception that like, okay, that I think this report essentially feeds into which is uh yeah there's a lot of people with a vested interest in getting you to believe open ai is in trouble and a lot of people with a vested interest and i literally mean a vested interest in this case and getting you to believe that open ai is unassailable uh and the fact is what you just described jason is what's called a competitive marketplace i don't think it means open ai is over their skis it means there are more people skiing now That's true. And so they're not the only one headed down the mountain anymore.

20:14Well, and early on, even a couple of years ago, OpenAI, it seemed like a large part of their business was very consumer focused. Now, with Codex, with some of their more enterprise kind of directions, and I think to a certain degree, this is thanks to Anthropic proving that there's a huge amount of appetite there for those tools in those areas. Now, OpenAI is doing that too for the same reason that you're talking about. this is what it means to be competitive in the AI space right now. You kind of have to have your hands on all things. OpenAI is doing that. So that doesn't immediately signal that things aren't working for them because look what they're doing now.

20:48No, it just means things are tightening up. And yes, you're going to see these kinds of pressures. And there's no guarantee that OpenAI will win this game. There's no guarantee that it'll lose this game either. You know, I was thinking about this when you're talking about Jeff Jarvis, because he covered the rise of Google very well in addition. OpenAI hasn't bought its DoubleClick yet. Google was in the same position before it bought DoubleClick, where it's like, well, it jumped out in front in search, but can it maintain it? And then it bought DoubleClick and dominated the advertising business.

21:21And I feel like maybe they're thinking hardware is that moment for them. Maybe getting some Apple design juice inside the house with Johnny Ive as part of that. Is that enough though? but yeah if you're sitting there wondering like why do i care about these big companies it's because this is going to affect the products that are available for you to use uh so yes you may not care who's investing in whom and how much money is going into what uh but who is able to continue to buy their compute and develop their products uh is going to change the products that you have whether you're a coder using codex or or cloud code or anything else um or just relying on chat EPT or using these kinds of tools in other situations.

22:06It's, I think, good background knowledge to know as a consumer, what are companies doing and how are things going? And the short version here is the companies are very competitive, which is good for you as a consumer. You're going to see a lot more competition for your business because of that. Yeah, 100%. Well, thanks to motang in our subreddit for sharing the valve has a new steam controller that is purpose-built for steam's ecosystem tightly paired to steam for pc steam deck the upcoming steam machine and steam frame vr headset it has very precise tmr joysticks it has dual track pads gyros inside for movement and a magnetic charging and connection puck that can actually support two Steam controllers at once.

23:01The controller itself, like I was looking at it, you know, clearly not in person, but photos looks a bit chunky when you're looking at the photos, not a bad thing necessarily. And reviewers are saying that it's very well balanced. It compares closely to pricier pro controllers. One example given was the Razer Wolverine V3 Pro, and it's going to be available May 4th,$99 in the US. I think it's 149 in Canada and 99 euros. but May 4th. So not very far into the future. Just like a couple of days. Man, tunneling magneto resistance. Just a side note, TMR, that buzzword of the week, because it's going to show up in the Logitech story a little later on about their new keyboard as well.

23:44But yeah, it's, it's basically improved hall effect. And it's, it's, it's the new wonderful thing that people want to have in their, in their keyboards and controllers and joysticks and all of that, that sort of thing. But yeah, I think the biggest takeaway here is that steam had to go and release this without the steam machine, without the steam PC, without the steam frame VR, which means that we are not seeing those anytime soon. They said, you know what? We got to put this out, especially because you can only use this with steam. You can't, you can't use this on other stuff. Now there's workarounds and there's hacks you can do.

24:22Don't get me wrong, but it is designed to only be used with steam because there's some features that get you some really good advantages that work directly with the steam overlay that don't work outside of that but there is a little bit of frustration of like why couldn't you have just put in a switch that lets this work like a regular controller even without those features and steam did not do that i assume because they thought well we're going to sell a bunch of steam machines and these are going to be mostly for that and this is not what's happening I wonder if that's the case, if enough people pushing back on this and saying, but yeah, it should be available to use on some of these other machines, if that's something that they can simply update with a firmware update.

25:04Well, they could. The question is, will they? Will they and will they want? I'm going to guess no. that they don't want to they'll like yeah the hobbyist maybe they won't get in the way of the people making the uh the third party hacks and sure and stuff because valve historically doesn't mind that that sort of stuff too much but i i doubt they spend the time on it they're going to spend all their time trying to find out if they can afford the ram for the steam machine yeah well that's true and this controller by the way does have some features that are not active yet because they're for future hardware like the steam frame it has infrared tracking uh yeah you know somewhere down the line the frame important if you're wearing a headset but yeah that will work yeah not as useful interesting uh youtube's rolling out an interesting thing called ask youtube you have to turn this on in labs it's a search experiment and you also have to be a premium subscriber but it will turn your queries into step-by-step answers uh giving you not just links to videos but mixture of text short clips from videos especially videos that have chapters They'll take you straight into a chapter, as well as your long form videos.

26:07So, for example, you could ask, plan a three day road trip from San Francisco and to Santa Barbara, which is a very Silicon Valley example that YouTube gave us. And then it will return specific video segments with titles and channel info, you know, and give you a little chart and some segments and then some full videos. they also suggested you could use it to say where can I find good coffee in part of that conversation so you've got that road trip down then it'll give you places along that route where you can find good coffee and reviews from people who've done that again this is premium subscribers you have to be older than 18 you have to be in the US and you have to opt in on the experiments page I did all that and I tried it for a, uh, imaginary trip from loose day trips from Lucerne, Switzerland.

27:01Um, and it gave me some great stuff. It gave me like a chart of, uh, three different mountains with a video for each one. Uh, and what the differences in the mountains are. It gave me a golden round trip with one person's video that explains the whole round trip. but in text it told me like okay you're gonna you do a boat cruise to a railway uh to a cableway and to a gondola uh so yeah i thought you know what this actually works pretty good it's pretty useful i mean some of some of my just regular google search queries end up with the video i'm sure you've seen it in the yeah yeah you know the very top ai section where it's like oh and here's here's the portion of this one video that answers the question that you're looking for it's kind of like that does it move that video into the main player when you know that is one thing that i was a little disappointed in uh is like so it gave me peak experiences mount pilatus mount rigi and mount titlis and when i clicked on one of them it just opens a separate page and then i have to use the back button to go back so yeah it'd be nicer if they had like a single player at the top that you can just pop the stuff into insert it into the carousel up top or something like that so Okay.

28:14And I'm imagining that's integrated into the ask button, that ask section that we've seen before. Yeah. When you do a search, it'll give you your normal search results and then it'll have that ask YouTube button. You pop that and then it gives you the response in there and then you can ask follow-ups and all that. So interesting to me how through features like this, YouTube is kind of shifting a little bit away from sit down and watch the video start to finish, search for your next video watch start to finish um you know the integration with notebook lm so that you can search the video and pull out information without having to watch it this where it's like okay we find the little clip the little piece of this one video out there so you get the thing you need and then get out it's kind of i don't know in some ways reprogramming how viewers are used to interacting with their videos and um i don't know i'm sure some creators are cool with it and some creators are probably like no i want people to watch my whole video Quit splitting it up into pieces.

29:12Yeah. You're paying me for longer watch time and then incentivizing people to watch less. What's up with that? They do a good job of highlighting the creators in this, though. They're showing you front and center the channel names and everything and all of that. Good, good. But I think you're right. Not that everybody's going to stop watching YouTube the way we watch YouTube now, of course, but this is going to provide another way to do this. I assume that YouTube thinks, well, this will be additive. You'll still watch YouTube the way you watch it, but every once in a while you'll, because people use YouTube for DIY, for help, for planning, like this will be helpful for that.

29:48Right. My daily YouTube consumption appetite might not be about that travel trip that I'm researching, but that one moment that's useful. So my not subscribing and not watching all their videos is kind of a non-issue because I wouldn't have watched them anyways, unless I was working on this particular task. This really only works on the desktop. You're not doing this on your television, on your smart TV. Right. Well, if you got thoughts on this, you're like, actually, the one thing you guys missed is that the best three day adventure from Lucerne is tell us in our discord. You can join it by linking to a Patreon account.

30:22You get the discord access by becoming a patron at Patreon dot com slash DTNS.

30:34Legendary doesn't happen by chance. It happens because it's what BetMGM does. Sign up for BetMGM and receive up to$1 ,500 back in bonus bets if you don't win your first bet. Get into hockey, basketball, and more of your favorite sports with same-game parlays, live betting, and more. BetMGM. Make it legendary. BetMGM and GameSense remind you to play responsibly. See BetMGM.com for terms. 21 plus only. This specific promotional offer is not available in Michigan, Mississippi, New Jersey, New York, Nevada, Ontario, Pennsylvania, Puerto Rico, or West Virginia. Gambling problem? Call 1-800-GAMBLER or 1-800-MY-RESET.

31:07Available in the U.S. For New York, call 877-8-HOPE-NY or text HOPE-NY 467-369. For Massachusetts, 1-800-327-5050. For Iowa, 1-800-BETS-OFF. For Puerto Rico, 1-800-981-0023. First bet offer for new customers only, subject to eligibility requirements. Rewards are non-withdrawable bonus bets that expire in seven days. In partnership with Kansas Crossing Casino and Hotel. Security program on spreadsheets, new regulations piling up, and audit dread? It's time for Vanta. Vanta automates security and compliance, brings evidence into one place, and cuts audit prep by 82%. Less manual work, clearer visibility, faster deals.

31:48Zero chaos. Call it compliance or call it compliance. Get it? Join the 15 ,000 companies using Vanta to prove trust. Go to Vanta.com slash com.

32:04All right, now some quick headlines. that are good to know might make you look smarter in the future yeah despite 600 google employees signing a petition objecting to this the information says google has signed an agreement with the u.s department of defense that will provide gemini for workers to use with classified materials the u.s government could already use gemini for unclassified now it will be able to use it for classified materials for any lawful government purpose okay then a filing from chinese EV maker BYD showed Q1 profit fell at its fastest pace since 2020 due to increased competition in China and slow sales.

32:45Apple will now let app makers offer a reduced monthly price if you commit to a 12-month subscription. So they already let you do a yearly subscription at a reduced rate. Now you can still pay monthly and get a reduced rate. You just have to commit to 12 months. Oh, OK. That's interesting. April 28th is the date that all smaller gadgets like phones, tablets, cameras and headphones must have must have USB-C charging capability in the EU and the European Union. Yeah, most of them have already met this deadline, but, you know, worth marking that today is the day. Today's the day. Yeah. Australia's government has proposed legislation, so this hasn't passed yet, but they have proposed legislation to add a 2.25 % revenue tax, specifically against MetaGoogle and TikTok, starting July 1st, that would be offset if they reach deals to pay news publishers.

33:42They would get a bigger offset for smaller publishers. And the more publishers they reach a deal with, the closer they get to offsetting the entire 2.25%. Don't reach any deals. You got to pay the whole 2.25%. That is a big chunk of their revenue, to be honest. And it's meant to replace that 2021 news media bargaining code, which everybody agreed to and has been abiding by. But they say it has not had the intended effect. Very interesting. The U.S. Federal Trade Commission estimates that people in the U.S. lost$2.1 billion in 2025 to scams originating on social media up eight times since 2020.

34:23Wow. Eight times. Yeah. And those scams are perpetrated on Facebook, Instagram and WhatsApp, according to the FTC. Oh, OK. In totally unrelated news, the FTC is suing Meta. Wow. That's odd. Speaking of lawsuits, a nine-person jury has been chosen. And today we'll hear arguments in Elon Musk's lawsuit against OpenAI, Sam Altman and Greg Brockman. So, yes, Altman and Brockman are employees of OpenAI, but he brought the suit against the organization and the two individuals. Musk alleges unjust enrichment and breach of charitable trust. Basically, he says OpenAI was supposed to be nonprofit and it's not anymore.

35:05I think I saw a quick clip of this right before Showtime where the judge is like, and please don't go to social media and just start blasting about this case all the time. Basically, looking at Musk and to Altman to say, like, please, let's keep it in here and not on social media. Good luck with that, Judge. Yeah, good luck. Good luck. Snapchat launched AI-sponsored snaps that let brands offer agents that customers can ask questions and get recommendations from. Ah, so you can talk to the ads now. Logitech, this was the story I mentioned earlier. Logitech has launched the G512X gaming keyboard that supports both mechanical and analog switches.

35:47So that tunneling magneto resistance, TMR switches, has also an 8K polling rate, dual dials, RGB light bar, and some game mode presets. It's available now for$180 or$200 and shipping May 2nd. Okay. All right. We end every episode of DTNS with some shared perspectives. And today, Sunil and Craig both wrote in with some information about the IPv8 draft. Yeah. Yeah. So this is a little bit of a lengthy email, but Craig and Sunil both did a great job of explaining this. And because IPv8 became not just a quick hit, but then a response from Andrew yesterday, I think people were like, wait, I know a lot about this.

36:29Let me share, which is exactly why we do this segment. So IP, if you don't know, is the internet protocol address. It's the number behind the domain name. You often hear it called like the phone book of the internet. It's also used for a lot of other things, but it's most famously like dailytechnewshow.com is actually 385.7.whatever, whatever. The most famous of those is 127.0.0.1, which is local host. IPv6 has a larger address space because it uses 64-bit addresses instead of 32-bit addresses. Those four numbers make you 32 bits. And the IPv8 proposal we mentioned previously gets to 64-bit numbers, but in a different way.

37:13Craig described it as IPv8 proposes 64-bit dotted decimal addresses. So you take the four numbers, the n.n.n.n.n of IPv4, and you add some routing to the front, r.r.r. So you get eight numbers. The first 32 bits, the r's encode a routing prefix, specifically an autonomous system number, if you need to know. And then those last four behave just like IPv4. The first four bytes are zeros if the last four bytes are just an IPv4 address. That's why it is interoperable with IPv4. Now, some folks are critical of this approach. Sunil noted, while this might seem to provide more addresses, the limitation is in how older devices handle such inputs.

38:00Older devices may not interpret or parse the proposed address correctly, leading to ambiguity or undefined behavior in routing and security. If parsing expectations differ across systems, it could introduce unintended routing or security gaps. For example, if the old logic is set to read just the first four groups, then it's just going to read the R numbers, not the N numbers, and that would take you to the wrong place. While backward compatibility is a benefit, this approach might be a security bogey in mixed environments. Also, the difference between IPv4 and IPv6 is the underlying architecture.

38:36Again, Sunil wanted to clarify. The difference can be loosely compared to RGB and CMYK color models. Both represent color. Red is both RGBFF quad zero and CMYK zero hundred hundred zero, but they do so using different structures and different assumptions. In a similar way, IPv6 is not just an extension of IPv4. It's a different approach to addressing and network design. And then Craig noted that this protocol, IPv8, has a lot of headwind, especially the OAuth authentication. And there are other IPv8 proposals. This was just the latest. Finally, Sunil adds, there have been exploratory ideas around the value each N could take in the IPv4, similar to improving accessible colors from 65 million to over a billion in RGB by extending the range from 255 to 1023 without changing the structure.

39:35However, in networking, such an approach runs into compatibility limits. The value of each N is constrained to one byte and values beyond 255 would simply fail validation rather than extend the address space, which makes that kind of scaling impractical. But the upshot is as good as it was that that IPv8 proposal was submitted, we are far from having a solution. We are far from being finished with this kind of stuff. And I'm amazed at the synergy of these two folks in the community. I know, right? Thank you, Craig and Sunil. And you being able to integrate it all, because while I understand it better than I did before, it's hard for me, like it's still going to take some percolating.

40:16It's a lot, yeah. Yeah. Well, thank you folks for sharing your insight. This is exactly the kind of thing that we encourage. So if you have similar insights, share it with us, feedback at dailytechnewshow.com. Yes, indeed. Thank you again, Sunil and Craig, for contributing to today's show the way you did. And thank you for being along for Daily Tech News Show. You can keep us in business by becoming a patron. It's right there if you're watching the video version, patreon.com slash DTNS. We'll see you tomorrow.

40:54The DTNS family of podcasts. Helping each other understand. Diamond Club hopes you have enjoyed this program.

41:07This 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. Advanced, intuitive, and deployed. That's how they stack. That's technology at Capital One. Security program on spreadsheets, new regulations piling up, and audit dread? It's time for Vanta.

41:42Vanta automates security and compliance, brings evidence into one place, and cuts audit prep by 82%. Less manual work, clearer visibility, faster deals. Zero chaos. Call it compliance or call it compliance. Get it? Join the 15 ,000 companies using Vanta to prove trust. Go to vanta.com slash com.