Open Weights Aren’t Just Chinese Anymore - DTNS 5368

0:00Being 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. I love days like these at the pool, but my moderate to severe plaque psoriasis can sometimes take me out of the moment.

0:34I'm ready to make a splash with clearer skin thanks to SkyRizzy, Rizinkizumab RZA. SkyRizzy is a prescription-only injection for adults who are candidates for systemic or phototherapy. At four months, most people saw 90 % clearer skin when measured from head to toe, and many were even 100 % plaque-free. People also saw significant improvement in psoriasis symptoms of pain, redness, itching, and burning. Don't use if allergic to SkyRizzy. Serious allergic reactions, increased infections, or lower ability to fight them may occur. Before treatment, get checked for infections and tuberculosis. Tell your doctor about any flu-like symptoms or vaccines.

1:10Now there's nothing on my skin, thanks to SkyRizzy. And that means everything. Ask your doctor about SkyRizzy, the number one dermatologist prescribed biologic and psoriasis. Visit SkyRizzy.com or call 1-866-SKY-Rizzy to learn more.

1:49800 Grainger. Click Grainger.com or just stop by Grainger for the ones who get it done. Losing weight can feel like a roller coaster full of ups, downs, and being thrown for loops. Like when someone brings donuts into the office or your kids' dino nuggets start giving you that look again. That's why there's Noom. They combine GLP-1s to quiet the food noise with healthy habits that help you keep it quiet, even if you get off the meds. Their program is designed to make weight loss a smooth ride, helping you understand what drives you, so you can build better habits to not just lose weight, but keep it off.

2:27Noom. Lose for keeps. Get started at Noom.com. Not all customers will clinically qualify for medications. Individual results may vary.

2:47oh this is the daily tech news for tuesday october 6th 2026 we tell you what you need to know give you important context and we're trying our best we're like the shepherd helping each other understand we'll try our very very best today france's hey mistral mistral over here uh france's mistral i i don't even know what that's from but it's in my head so there you go film noir in general as far as i know yeah i suppose so maybe it's applicable i don't know but france's mistral finds a third way between open and closed what could that be we'll talk partially open uh i'm tom merritt i'm jason howell let's start with what you need to know with that partially open big story so yes mistral released a public preview of mistral large four it's the fourth in the series of mistral large uh you can just call it ml4 or if you're on very familiar terms with this multimodal model you can call it le chonk as in chunky as in it's got a trillion parameters.

3:55Mistral says it was trained on 3 ,800 NVIDIA Grace Blackwell GPUs. So not chonky on the compute side. They didn't have to go anywhere else. They're like, we just used our own compute. It says it is two to three times fewer than the amount of GPUs used by all of the competing Chinese companies that release open weight models. ML4, or LeChonk, can be accessed now through a public preview API in Mistral Studio. And then after safety testing is complete, Mistral plans to make it an open weight model somewhere around October 27th. Look for that. Mistral expects it to outperform other open weight models and possibly outperform some closed frontier models in specific areas.

4:37They say this ML4 is good for security, for finance, and for chip design, which I guess kind of makes sense, since two of Mistral's main backers are ASML, which makes the lithographic machines for everyone and Samsung which makes chips it is at a trillion parameters you might have guessed too big to run on your desktop but once the weights are released companies will be able to run it on-prem or in a private cloud it might even be too big according to TechCrunch to run in a university because it's pretty darn big but for companies that have the space for it you'll be able to run it yourself and not worry about giving that model over to Mistral or anybody else, which I think what Mistral is targeting is like, hey, you want sovereign AI companies, but you don't want to rely on China.

5:25How about France? We're friendly most of the time, at least on AI. Well, yeah, we're friendly most of the time. And I think to a certain subset of people who might consider this, we're also not the U.S., right? We're not the U.S. or China, right? We're the third way. Yeah, yeah, yeah. Exactly. And so maybe that's part of the midway between open and closed is not the U.S. and China. Anyways, I love LeChonk. I know it's stupid to start with that, but I picture a character with LeChonk. So I think that was a good move. I just think Mistral, a lot of what Mistral is doing is really smart. I think they're playing an important role at a time where we're so used to hearing about these frontier releases from U.S.

6:12from U.S. companies. And then on the other side, it's the Chinese open weight and everything happening with open weight models over there and everything, but attached to China. And some people, you know, look at that and they say immediately, like, well, then I'm not going to go there. Having a company like Mistral that is coming from, you know, from Europe and Europe has now, you know, and has for a while now, had an AI company that is doing good things and kind of going toe-to-toe with, you know, in that conversation with everyone else. So I think it's a good place for them to be. And I mean, a trillion parameter model, that's nothing to sneeze at.

6:50This sounds impressive. It's chunky. I get it. Chunky. Very chunky. It's a chonker. Mistral is the Spotify of European AI. It is the one you can point to and say, but they're doing well. I know some people in the industry argue about whether they're given more credit than they're due simply because they're Europe's representative in the fight. But I think this is a chance for them to say, no, there should not be truth to that. You know, they're going to say, this is a very competitive model. Is it better than OpenAI or is it better than Claude? Maybe not. But it is designed to be very good for other things.

7:32And, you know, the Avis, we try harder of the situation. And that's what we want. We want competition. on a global scale for this sort of thing. So I'm glad to see that some of the speculation that Mistral was just kind of fade into being a service provider, providing access to open-weight models to other people, turned out, at least for the moment, not to be true. I'm glad about that. Yeah, and also kind of on the flip side here is that there's a lot of excitement about this from what I was reading through, but it's also very tempered with kind of a wait and see, right? Because we know of it, published results are going to need to match what people have, you know, see when they actually get their hands on it.

8:14And what is it actually good for? Yeah. What is it actually good for? Because so many of these happen, what it also just kind of reminded me of on a personal level, and I'm certainly picking this up from other people in the industry, is just a little bit of like model release fatigue. They all say that they're good at agentic and they're all good at coding and They're all good. And, and they're coming out at such a rapid pace. Um, it's, it's kind of, you know, it's, it's a constant game of leapfrog keeps it harder to, you know, pay attention to all the actual details and you don't really know until it can be proven.

8:46So, you know, this sounds amazing. We'll find out once people actually get their hands on it. I call that the sign of, I call that the sign of a maturing market. Yeah. Okay. When it goes from being a horse race of like, it doesn't matter what they do. let's look at the numbers and see who's ahead. That happens a lot in the early part of any product cycle. When you start to hear the people who love the horse race say, I'm just tired, there's too many models. To me, I'm like, okay, so we're getting product differentiation. We're getting models that aren't exciting in and of themselves. We're getting products that are good at models.

9:22This is the classic smartphone situation where people start, oh, smartphone releases aren't exciting anymore, but they're the best smartphones you've ever had. And you have better choices than ever. It sadly has not brought the price down in this current situation. But you saw a very thriving market and a lot of consumer choice. And I think that's what we're looking at right now. The other thing that we shouldn't sleep on is that Mistral says this was all done with reinforcement learning. They didn't use distillation, which is, I think, probably the sleeper story here. If all of this bears out and it's a good working model and it's good at certain things, it's not whether it's first or best.

10:04It's they did it efficiently. They did it without distillation efficiently. Usually distillation is the way you do it efficiently. They trained it on 3 ,800 GPUs with reinforcement learning. I want to know how they did that. And so is everybody else going to want to know that too, I think. Yeah, no, I absolutely agree. I'm very, very curious about that. Yeah, we'll find out. Are you telling me that I can't get a desktop to run this on, Tom? Come on. I need this to check my email. The only reason I would tell you that is to watch you try. I don't know. Maybe later in the show, there'll be a story that we'll talk about that.

10:40I don't even think that one could run a trivia parameter. But yeah, maybe. Maybe. I'm just saying there might be a chance.

10:50All right. DTNS is made possible by you, the listener. Big thanks to Paul Boyer, D. Laser, Brad, and Tom Eastman. Thank you so much for being amazing patrons. You're awesome. I used to be the person who hunched over my laptop mid-meeting, typing furiously to catch every word and missing half the conversation because I was too busy transcribing it. That's the problem Granola fixed for me. Granola is an AI-powered notepad that makes your meeting notes useful. It captures what happens in your meetings and turns it into clean, structured notes with decisions and action items pulled out and made easy to find.

11:25And the best part? It integrates seamlessly with how you already work. No setup, no configuration, no friction. Before Granola, my post-meeting routine was triage, 20 minutes after every call decoding my own shorthand, figuring out who owed me what. Now the notes are just done, organized action items pulled out, before I've even closed my laptop. You get better notes in one place, and you get to actually stay present instead of typing the whole time. Once you try it on a first meeting, it's hard to go without. You can try it totally free. Head to granola.ai slash daily tech. That's granola.ai slash daily tech to get your time back.

11:59Go to granola.ai slash daily tech and try it on your next call. Episode prep gets messy when research, Slack threads, and script trackers are all in different places. That's where chat GPT 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. For 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.

12:35Put 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.

13:13That's how they stack. That's technology at Capital One.

13:43saw significant improvement in psoriasis symptoms of pain, redness, itching, and burning. Don't use if allergic to SkyRizzy. Serious allergic reactions, increased infections, or lower ability to fight them may occur. Before treatment, get checked for infections and tuberculosis. Tell your doctor about any flu-like symptoms or vaccines. Now there's nothing on my skin, thanks to SkyRizzy. And that means everything. Ask your doctor about SkyRizzy, the number one dermatologist prescribed biologic and psoriasis. Visit SkyRizzy.com or call 1-866-SKY-Rizzy to learn more.

14:17Some days, getting the work done isn't the problem. Staying locked in is. AlphaBrain is Onnit's daily nootropic formula for cognitive support, designed to support memory, focus, attention, and mental clarity. Its formula with science-driven ingredients include theanine and huperzine A to support focus and memory, and it's caffeine-free. So when you're deep in the work, AlphaBrain helps support enhanced productivity and tune-out distractions. Get focused on what matters. Shop AlphaBrain now at onnit.com. There's more we need to know today. Let's get to the briefs. Let's do it. Another company coming for those Chinese open-weight models that we were just kind of alluding to a few minutes ago is New York's Reflection with its new Beam AI model.

15:07Beam has 501 billion parameters, but it's very efficient. It uses a mixture of experts approach to leverage about 23 billion parameters per task. Reflection claims that Beam is competitive with GLM 5.2 from ZAI and is near the performance of QN 3.8 Max for coding. But it's twice as efficient as GLM 5.2, so a lot cheaper to run, in other words. reflection claims it'll be three to four times cheaper to run than comparable open models artificial analysis actually got early access to beam still running tests but the early indicators there are that it'll be one of the most token efficient models uh that they've ever seen so i mean if mistral had not entered the chat with a potentially open weight model that we're going to see the open weights at the end of the month.

16:02This would be the biggest story of the day in AI for sure, is that a capable US open weight model is in there. Add this to Mistral and you've got both Europe and America coming in with capable open weight models. And again, we do need to benchmark these things and see just how capable they are. But I'm willing to say on the surface that they're worth evaluating. And especially with the efficiency, which is incredibly important in this compute-constrained world that everyone is working in. I think this is very good for, again, the end users who want to run this sort of thing. And it's also an open-weight model.

16:40So you can get it yourself. This one, only half the parameters of LeChonk means you might have a chance of running this on a local machine. Yeah, I'm trying to think of what a modifier to Chonk would be to take it down a notch. On put a chunk. Le petit chunk. Le petit chunk. Something like that. Yeah, competitive with China, maybe not competitive with its best models yet, again, remains to be seen. But we're really going through a moment right now where Open, which has been, I think, to a large degree, whether deservedly so or not, has been associated directly with the Chinese models. Now we're seeing this kind of opening of, I guess, an understanding or willingness to take open models in, even if they're less capable than Frontier.

17:33But because of all the benefits that they do afford, China's having clearly a lot of success with it. And so it's nice to see other outfits in the U.S. and in Europe and stuff going down this road. Yeah, I think we're always talking about, at least I'm always talking about, Huawei trying to catch up in chip design. This is in open weights. We're seeing the other end of this. And I think all of the U.S. energy has gone into frontier models because those are the most capable models. but now that we're seeing consolidation and solidification of that market you don't have a lot of room for people to become frontier model providers these days uh you do have room in this part of the market so of course we're going to see people start to fill in that side of it that's super true and um you know at the end of the day competition means choice and choice is a great thing so good to see as you alluded to earlier uh we have a machine that might have a chance of running Beam locally.

18:3419-year-old Zane Javid launched his company Ghost out of stealth this week. TechCrunch Disrupt happening in this week, so you're going to hear a lot of these kinds of stories. Its first product is called Core, C-O-R-E, no relation to Scott Johnson's podcast. It's a personal computer, desktop computer, designed specifically to run agents. But it's not a desktop computer meant to sit on your desktop with a keyboard, mouse, and a monitor. It is, however, running an NVIDIA RTX Pro 4000SFF Blackwell GPU. That is a data center quality GPU and specific software. That's where Ghost says this isn't just about putting the right hardware together.

19:17They have specially designed a browser, a file system, and software that can process information and carry out tasks without constant prompting. The idea is for this to be well-designed to use the hardware to work on its own when you want it to. Now, it will come with Quen 3.8 Next, Quen 3.8 27B, and Gemma 431B pre-installed. So Google's open-weight model and a couple of Quens from Alibaba. You can install your own. You can go to Hugging Face or anywhere else and install your own if you want. They just thought it's nice to give you a step up and pre-install some. and you access this computer through a phone app.

19:56That's what I was saying about no monitor, no keyboard. I mean, you could probably figure out how to do it, but that's not what it's designed to do. It's designed to be an appliance that you access through the app. You can type, you can talk, but you tell it what to do and it just goes and does it. Everything is encrypted. No third-party data needed. Ghost can't see your data without your permission. Neither can anybody else. You decide how much autonomy to give the agents and they made a point of saying if ghost shuts down this machine is designed to keep working even though ghost will provide updates to the software that software will continue to work should ghost ever go out of business ghost is also building a firewall that will monitor every outgoing network request on the computer and see if the agent is performing a sanctioned action or maybe trying to exfiltrate data it shouldn't uh it rejects those requests and flags it for user review that's not out yet that is something that they're working on uh the gpu in this thing alone is worth three thousand two hundred dollars right now the case is estimated to be worth eighteen hundred dollars because it's a very pretty case uh you have a terabyte hard drive in there you have 24 gigabytes of ecc ram tech radar estimated the worth of the parts at five thousand dollars but pre-orders opened Monday for$3 ,499.

21:15One of the things Ghost said is that they got a deal on these GPUs when they were$1 ,700. So that's part of the reason they're able to sell it this way. And by the way, you can't buy one right now. It's already sold out. If you were lucky enough to get one, shipping begins October 31st. If you want the next round, you need to put down a fully refundable$499 deposit. This is the kind of thing that as soon as people heard about it on Monday, they've immediately tried to order it. and they ran out. They're assembling these in San Francisco at their headquarters. This is not something where they have a big factory yet set up for them.

21:48I mean, especially if you're talking parts worth five grand, even though they got a deal on those GPUs in advance. They only got a deal on so many of them. They only got so many. So what happens on the other side of that? But I think this is, I would guess that we're going to see a lot more of this in the next couple of years. Now, granted, the timing is crazy with the component shortages and high-cost memory and all that kind of stuff. But we talked about this a lot on AI Inside, me and Jeff Jarvis, about kind of the thirst or the need for a certain level of user to want some sort of an AI appliance like this, appliance a.k.a.

22:27a computer designed to run AI models and everything, that doesn't come with necessarily a lot of that complexity that is required for people who want to install their own models or whatever. Like, can it just be easy? Can it get to a point to where putting these models on is basically as easy as like installing an app or something like that or picking up an appliance and being like, you know what? Hey, it does all these things. It does it really well. Yes, it's easily updatable and I can just start using it to manage my life. Also, it's local, you know, and so that's going to make a lot of people happy.

22:58But it's ridiculously expensive, you know, compared to what we're used to spending for consumer computers. But that's just where we are right now. I know. It's the price of a flagship foldable phone. I know, which is weird. Then when you put it in that context, then you're like, oh, maybe it's not that much at all. I don't know. It's more expensive than a flagship foldable phone. But that was the first thing I saw. I was like,$34.99. That's what top end PCs cost when you max out. They're like, well, this is maxed out. It's got a GPU in it. And how much more expensive is it from a flagship phone these days, crazily enough?

23:32Yeah. Yeah. The other thing that occurred to me when you were mentioning the ability to just say install an agent without having to do it easily, because this comes pre-installed with Gemma and Quinn, you theoretically could ask those to install Beam or something else. Well, that's true. Yeah. It's like using Internet Explorer to download Firefox, right? Yes, totally. Back in the day. Like, I wonder how many people will just use it that way. I don't think it's quite where you were talking about where it's like super easy and meant for people who don't know a lot about this. I think it's still meant for super users at this point.

24:14Yes. But it's pointing the way, right, to that becoming an appliance where you're like, oh, yeah, I just have my agent is plugged in over there in the corner. And then you access it on your app and you just talk to it and ask it what you want. um we are good we're going to see a version of this from microsoft and videos joining microsoft for the surface announcement tomorrow we're going to talk about we'll mention the details on that a little bit later uh the dgx spark is five thousand dollars which is what tech radar estimates the price uh of the parts of of the uh of this one to be so yeah i i do like core i i've almost forgotten the name of it because i'm just thinking about that gpu this is i saw somebody describe it as It's like you buy an NVIDIA GPU and a free computer comes built around it.

25:01They're not kind of far off the base. Yeah. Let us know if any of you are like, I'm definitely giving a flyer on this if I can get one. Or if you were lucky enough to get one, feedback at DailyTechNewsShow.com. Yes, indeed. The Documentation Foundation said in a blog post that it will not add AI to its software for the foreseeable future. Part of the reason is that the organization doesn't want you to need an internet connection for any part of the software to function. You can still install extensions. You can still connect to local models if you choose to. LibreOffice says no AI integration meets its requirements, such as users' data staying on the device and no reliance on a single provider.

25:48So having, yeah, no AI, I guess, in the install process, if you really want it, being able to install into it, I know that's going to make a lot of people happy. I look at this story and usually I'm in the position of this isn't as bad as it sounds, trying to explain that. And then I take a lot of heat for like, you're just sticking up for the man. This isn't as good as it sounds for people who hate AI. I think a lot of people are like, good for LibreOffice, sticking it to the man. No AI. I'm going to download LibreOffice right now. And you should. It's excellent. You should use LibreOffice. I like the way they describe this, which is we're not against AI.

26:32We just want it to meet the criteria of not being crappy and protecting your data. And nothing does that yet. So we'll let you add it if you want. We're not going to tell you what to do. But we're not going to add it until it meets those criteria of protecting your privacy and all of that sort of thing. So I fully expect there to be a blow up when LibreOffice finally finds something that meets its requirements and integrates it down the road. But for now, I think this is going to be useful for people who do want control over their data with AI and for people who are like, I want nothing to do with AI.

27:09And the big win is being committed to you should be able to use the software without an internet connection, which I think is absolutely wonderful. Yeah, I absolutely agree. I feel much the same way here. I read this more as just putting the user in control of the experience a little bit more than maybe they would. When you talk about companies like Microsoft and Google and everything, they're baking that in so heavily. Not everybody really, truly wants that in there for a whole multitude of reasons. And yet it's becoming more and more or less and less controllable. There's no ability to kind of remove it.

27:48So I see this as being a give the user the ability to add it in if they choose. I think that's a good thing. It goes back to more control in the hands of the user is a good thing. And a good reminder for folks, LibreOffice is out there. If you're shaking your fist at Office 365 or Google Docs or something, just run it locally if you don't need it to be in the cloud. We also would like you to consider buying something from our store. Now, if you're not an AirPods user, we got T-shirts, we got mouse pads. Mouse pads are good for a lot of things. Amos was using a mouse pad to cover his head from the hot sun during our meetup in Los Angeles when it was 100 degrees.

28:29See? You had no idea. But yes, it's useful in that scenario. So many uses. But if you are an AirPods user, we do want you to check out our AirPods case. A couple of you bought these recently. I want to hear your reviews. It's just an AirPods case for charging your AirPods. It has the DTNS logo on the front. And then on the back it says, do not disturb me. I'm listening to DTNS. which, whether it's true or not, might be a way to just keep people off your back at the coffee shop. So go check it all out. It's available at dailytechnewsshow.com. slash store.

29:28Advanced, intuitive, and deployed. That's how they stack. That's technology at Capital One. Your daily supplement routine doesn't need to be managing a handful of different bottles. Onnit's Total Human brings key research-based vitamins, minerals, herbs, and amino acids together in convenient day and night packs. One all-in-one product designed to make supplementation easier. The day pack supports cognition, focus, and cellular energy production. At night, the Night Pack helps you wind down and supports a calm, relaxed state. Less time managing your supplements, more support for your day and night.

30:06Shop Total Human now at onnit.com.

30:12These headlines are quick and will make you smart. I guarantee it. Guaranteed. Microsoft will show off new Surface hardware, as we alluded to a little bit earlier, in conjunction with NVIDIA. tomorrow. That's Wednesday, October 7th, starting at 1 p.m. Eastern, streaming on the Windows YouTube channel. What if someone's listening to this on Wednesday? It's not tomorrow then. That's true. I broke the wall. What if someone's listening to it yesterday? Then I want to talk to that person. I want to interview that person for a Saturday episode. We have already emailed you in the past. anyway uh we already recorded the episode and then jumped back yes uh and they're right here according to google support pages starting on friday october 9th which could be tomorrow if you're listening to this on thursday uh free users will no longer get access to gemini flash and pro models leaving flash light as the only free option flashlight the five dollar a month will keep access to Flash, while the$20 a month plans and up get pro access.

31:20And if you're on the $100 a month ultra plan, you will soon get access to Gemini for Argon once it becomes generally available. Oh, really? Okay. The Wikimedia Foundation said OpenAI agents posted edits intended to repurpose a citation tool to access data from third-party sites. They also tried and failed to compromise the wikipedia etherpad note-taking tool for the same purpose the agents also made millions of api requests crawled millions of pages and made hundreds of thousands of queries to the wikidata query service straining the system as one could imagine wikipedia foundation just like you need to take responsibility for this it wasn't horrible it didn't deface wikipedia it did maybe cause an outage maybe not we're not sure but we didn't like it so please take it take responsibility uh investigators have determined that attacks against seven south korean financial firms at least seven of them uh show traces of an open source security tool called artex ai artex was developed in china and can use whatever third-party model you want you can plug it into something.

32:35You'd point it at a model and it uses it. It doesn't have a model itself, but then it powers its responses. It's meant to be a security tool. They've updated the terms to say you shouldn't be using this for illegal purposes, which probably should have been in the terms in the first place. But yeah, we're going to keep getting these kinds of examples of people using tools meant for one thing for another thing. And completely unrelated, our tax is Atreyu's loyal horse and companion in the 1984 fantasy film, The Neverending Story. I don't know. Actually, I did. That's why I looked it up. I wonder if that's on purpose or accidental.

33:10I don't know. It's our text versus our tax. It's one letter off. Okay, okay. I don't know. It would be super nerdy. I would appreciate it. I'd be into it. UK Communications Regulator Ofcom is investigating whether Instagram properly assessed risks, particularly child safety and illegal content risk before launching the instance feature in may instant posts vanish after the user has opened them meta says we told them all about it beforehand uh but that doesn't that mean anyway uh carnegie china released a survey of where top ai talent is working uh the share working in the u.s has declined that has been the story uh top talent worked in the U.S., 46 % in 2022.

34:02The share is down to 34 % in 2025. 34 % of the world's top AI talent now works in the United States, while China's share rose from 27 % to 41 % in that same period. However, Chinese talent is still flowing to the U.S. with 46 Chinese AI experts working in the U.S. for every one U.S. AI expert working in China. China is just keeping more of their experts. They're not pulling them in from the United States. South Korea had the second highest retention rate behind China. And meanwhile, the report found that no European or Indian institutions made it into the top institution rankings in 2025. Mac rumors, sources, sources to Mac rumors anyway, say Apple will have two product launches in October, sometime this month, one around smart homes as previously reported by Bloomberg, and another at the end of the month, though it's unclear what products would be announced then.

35:01They're guessing MacBooks, which makes sense to me, but we don't really know. The Emmy Awards are going to move from their practice of rotating among the major broadcasters in the U.S. each year to being available for free on Amazon Prime Video worldwide for six years starting in 2027. Going fully online. Interesting. Google has reached a deal with Constellation Energy to boost output at 11 nuclear reactors in Illinois, Pennsylvania, and New Jersey. That will make up 25 % of the power Google is contracting for. And Skydance closed its merger of Paramount and Warner Brothers. The new company will be called Skydance Media.

35:41Skydance Studios? It'll be called Skydance. And in the press release announcing the closure, said to expect, and I quote, significant improvements to its direct-to-consumer streaming products, which will unify into a single service over time. So I guess HBO Max and Paramount Plus may merge after all over time. CEO David Ellison has referred to the combined offering previously, but some people thought that would mean a bundle. He's also said he wanted HBO to be able to operate with independence so that HBO can candidly do what it does incredibly well. Casey boys who runs HBO is going to be in charge of all the direct to consumer.

36:20So I guess maybe they just merge paramount plus into HBO max and still give HBO independence. I don't know. I'm very confused now because they also pointed out paramount plus and HBO max separately in their video talking to investors earlier this week. So I don't know. I don't know if they're going to merge or not, but it sounds like maybe they will. Sounds like something is going to happen. That's the news today. Something is going to happen. We end every episode of DTNS with some shared perspectives. Today, Hugh has a very good question. Yes. Thank you, Hugh, for this. Hugh wrote a longer email, and I read it all and appreciated your daily sports headlines feedback, Hugh, heard.

36:57And we will get to you on that as well. But I liked this question, Hugh asked. why does there appear to be no legal or financial consequences for the companies that have agents that evade control? I feel certain I would face severe consequences if I did similar things during an interview or a training development exercise. This is an excellent question, Hugh. Why are these rogue agents able to get away with it? Why isn't everybody just suing OpenAI? It is a complicated answer, but I will attempt to give you one. First of all, you can't prove the agents had intent. When you want to prove criminality like fraud or computer fraud and abuse, you do have to show a measure of intent.

37:44And it certainly wasn't OpenAI's intent to do this, nor could you show that the agent intended to cause harm. There's also not a human actor. It's a bit like brakes failing on a parked vehicle with nobody in it. There was no driver. It just happened. Now, it could be negligence if the previous driver didn't properly park it, right? But in pursuing negligence, you have to prove that the actor, in this case, the company, OpenAI or Anthropic, could have foreseen the danger and didn't take proper actions to avoid it. And experts, both in and outside of these companies, say that these actions were unexpected.

38:25So I think in a court of law, it'd be a good defense to say, like, we never intended this to happen, nor did we think it would. So that's we're not being negligent, at least under the law. Then it becomes a product design issue. All right. So did you make a defective product? That becomes particularly tricky to prosecute under existing laws, because what is a defective AI product? And so a lot of companies are falling back on insurance to protect them, saying, you know what, if we get any monetary damages, we'll just file a claim with our insurance for it. The unclear legal landscape, the protection of insurance have kind of combined to lead most organizations to at least see what the investigations have to say, talk it out among themselves rather than go to court.

39:12And that's now. That's going to change. The EU and other governments, including U.S. regulatory agencies in some states, are enacting or enforcing more expectations on the safety of these. So that will become a little clearer. You'll be able to have a law you can go into court with specifically for these instances. Some lawyers expect claims to eventually be filed along the same lines as you'd file for an oil spill, right? No, they didn't intend the oil to spill, but it damaged the community. And so, you know, they need to pay some kind of restitution for that. So we might still see some cases.

39:48But the reason you don't see everybody running to sue them right now is all of this. Like, it is a complicated legal landscape. And companies are willing to be like, wait, this isn't going to happen again. Also, the damages have not been particularly costly either, at least not yet. Yeah, good summary. You know, I think at the end of the day, also, there's really no way to absolutely guarantee safety in these products because they're still, to a certain degree, they're still trying to figure out exactly how they do what they do. You know what I mean? I mean, there is a way to guarantee safety.

40:21It's to not allow them access that they had. Yeah, well, yes. I think there's a lot of ways to guarantee safety. The question is, can you get the training and assessment you need, especially the security assessment, right? This isn't just about improving the models. It's also about if we're going to see what they're capable of, we have to see what they're capable of. Yeah, that's true. And there are ways to do that more safely. They just didn't expect that the water would flow into that part of the house. There was a leak they didn't anticipate. And now everybody's walking around like, well, you should have.

40:55You should have seen that hole and maybe they should have. Right. And so. So, yeah, I think even even if they say, well, I don't I don't know if there's a way to guarantee safety. That's an argument for the to go in front of a judge and say, well, then they shouldn't be making these. Right. Yeah. Fascinating. We're in the eye of the storm right now. We'll see where that leads. What are you thinking about? If you have insight into a story, you can share it with us. Leave feedback at feedback at Daily Tech News Show dot com. Yeah, big thanks to Hugh for that question, contributing to today's show.

41:30Thank you all for being along for Daily Tech News Show. I was going to read a new patron's name today, but none of you signed up for patron yesterday. You want to be the name tomorrow? Sign up. Your name will be in the show tomorrow. Patreon.com slash DTNS. Bye. The DTNS family of podcasts. Helping each other understand. Diamond Club hopes you have enjoyed this broker.

41:59This 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. Some days, getting the work done isn't the problem. Staying locked in is. AlphaBrain is Onnit's daily nootropic formula for cognitive support, designed to support memory, focus, attention, and mental clarity.

42:42Its formula with science-driven ingredients include theanine and huperzine A to support focus and memory. And it's caffeine-free. So when you're deep in the work, AlphaBrain helps support enhanced productivity and tune out distractions. Get focused on what matters. Shop Alphabrain now at onnit.com.