 This 10th year of Daily Tech News show is made possible by its listeners. That's you right now. Thanks to every single one of you, including Mike Cortez, Degracia A. Daniels, Erwin Sturr and new patrons. Somebody new patrons. Welcome Matt, Chris and Glenn on this episode of DTNS. VR at work could be bad for you. Authors are very cross with AI. And despite that, NVIDIA launches a more powerful chip to bring down the cost of AI. Read the room, NVIDIA. This is the Daily Tech News for Tuesday, August 8th, 2023 in Los Angeles. I'm Tom Merritt. And from Studio, I don't know what it's called yet. I'm Sarah Lane. And I'm the show's producer, Roger Chang. Roger, is that the official final name of Sarah Studio? Studio, I don't know what it's called yet. Subject to change, I'm sure. Why are you asking Roger? I don't know. He's I was looking for produce serial advice on whether. Get some lettuce while you're while they're cheap. That's right. Good advice. We'll have more advice on produce from Roger. He doesn't understand the title after the show. Netflix just released an iOS app that says it will let you play Netflix video games on your television. Let's get to the rest of the quick hits. Nick, Asia reports that Apple, Samsung, Intel and NVIDIA will all invest in soft soft bank groups. UK chip design unit arm, you might have heard of it. Next month, arm is expected to announce a 60 billion dollar IPO or thereabouts on the U.S. NASDAQ stock exchange. The participating chip makers will each get a few percent at the launch of the IPO, becoming medium to long term shareholders, which should stabilize the stock price once it's listed. And the chip makers will then have at least some sway over arms management going forward. Speaking of chips, TSMC announced it's agreed to build its first chip factory in Europe. It will be located in Dresden, Germany. This was known to be coming, but it's now official. TSMC will receive government investment, some subsidies to bank the plant happen that is subject to EU approval, but it's expected to get it. Bosch, Infineon and NXP are also chipping in. They will jointly own 10 percent of the plant in a joint venture. This comes a week after approval for an Intel plant to be built in Germany as well with production expected to begin at the TSMC plant in 2027. Now, some analysts believe TSMC may face a shortage of skilled workers in Germany, as it has in the United States. TSMC had to delay its Arizona based plant production date because it's still looking for qualified workers. Meta owned Facebook Messenger has started notifying Android users that as of September 28th, SMS messages will no longer be available once their Messenger app is updated. Facebook Messenger has a pretty large Android user base, and presumably some portion of those folks use the SMS MMS integration. Now, with Meta already owning WhatsApp Messenger and Google pushing Android users to switch from SMS to its messages app, which supports RCS communications, SMS does not seem to top of mind for Meta these days. I know it's it's hard to it's hard to wrap your head around Meta killing a messaging app on Android when it's usually Google doing it. But, you know, different different times. The UK's Electoral Commission announced that voters physical and email addresses were accessible to hostile actors as far back as 2021. The Commission discovered a breach in October and reported it to both the Information Commissioners Office and the National Crime Agency. It delayed notifying the public while it removed the actors access to the systems. But that seems to have taken quite a long time. The Commission says the attackers had access to the electoral register systems and email systems, but that it does not know conclusively what files had been accessed. The Las Vegas, Nevada City Council approved the boring companies plan to expand its tunnels in the region. The plan now calls for 81 station and 68 miles of tunnels throughout Las Vegas and surrounding Clark County. A reminder, if you're not familiar, the boring company system moves passengers in Tesla vehicles through a system of tunnels underneath surface streets. Currently, the system includes three stations at the Las Vegas Convention Center and an offshoot open at Resorts World. Construction of the rest of the system is planned, but not yet scheduled, at least not publicly. All right, that's look at the quick hits. Seagraph is happening this week and Nvidia CEO Jensen Huang kicked off the conference with a keynote. Here are some of the big announcements. Sarah, are you ready? I'm ready. Big new chip, the GH 200. Did you see what the GH stands for? I didn't know. Legendary computer scientist, Admiral Grace Hopper. That's cool. Shout out to Grace. OK, that's nice. The GH 200 uses the same GPU as the current H 100, but has 141 gigabytes of memory compared to 80 in the H 100 and a 72 core ARM CPU. It's coming in Q2 2024. Now, GH 200's big advance will be in running inferencing fast. There's training and there's inferencing. Training is computer intensive, but you do it once. And then you've got the model more than once. But you know what I mean, you don't have to do it all the time. Inferencing is what the models do all the time. So when it generates an answer to your question or an image to your request, it is using inferencing to generate that. That is also computer intensive. And Wong said this chip will drop the inference cost of large language models. The two GH 200 chips can be combined to run as one. If you want to do even larger models. This means models would not need multiple systems to run. This is Nvidia's answer to AMD's M 1300 X, which has a hundred ninety two gigabytes of memory and is also designed for inferencing. OK, so dumb question. No such thing not to be confused with the interference with that actually kind of tripped me up earlier. Yeah, it's I'm inferring from what you said. And so inferencing in the AI sense, and I'm probably stepping outside my bounds here when I say this. But from what I understand, it's the it's inferring at what an what a proper response to your prompt would be based on the model it has. OK, so second question based on what you've said about the GH 200 is, you know, is the fact that it's packing more of a punch. Does that mean that data can just be crunched that more quickly? Is it does it mean that the physical model of this is a smaller model or maybe both? I feel like it's not so much about the size, but it's about the amount of data you can do in a particular chip, right, because you've got more memory. And I think they've tweaked the performance better. That's why they say you can combine two GH two hundreds and one machine and then not need another machine. And as soon as you don't need to spread out, you know, distribute your your task across multiple machines, that makes it more efficient on its own because you're not having it is kind of saving space. You know, yeah, yeah, I guess that's not really it's basically that this is just more powerful. Yeah, it's not that the chip's actually smaller or anything. But yeah, it would save you space in that you would need fewer machines, which uses less power, which means that you need less space in the data center. So if you're renting the space, that saves you new money, etc, etc, etc. It just it's going to be faster and cheaper to run these these these kinds of models. At least that's what NVIDIA is promising. I am these also promising that it's not like NVIDIA is the only one. But NVIDIA kind of the leader in this stuff. Jensen Huang in the keynote made a point of talking about how they sort of bet the company on the future of AI before it was clear that this was going to be as big of a future as it is. And gosh darn it isn't he glad they did because it's bringing in the biggest amount of growth for NVIDIA right now. Well, and NVIDIA has a huge AI chip market share at this point, right? So, you know, 80 percent, right? Is that lots of other companies doing this as well. But NVIDIA is poised to become sort of the industry leader. Yeah, a couple other things out of this keynote. AI workbench will let you create, test and customize generative models on a workstation before having to deploy them to the data center. Available models include Hugging Face and GitHub. That's meant to speed up customization. So it'll still pull in cloud nor cloud resources if you want it to when necessary. But they noted that Gartner suggests close to 85 percent of big data projects fail because of infrastructural roadblocks. So being able to do it on a workstation before you deploy it and have it fail is a big benefit for companies. And NVIDIA also partnering with Hugging Face on something called training cluster as a service. So this is training software to train people up on working with data. It's powered by NVIDIA's DGX Cloud and Hugging Face's training models and data sets. The big data projects failing note struck a chord with me because I thought, all right, well, you know, if big data says, you know, for example, something that comes out of NVIDIA saying, you know, we can we can help you. And then, you know, whoever is running the city council, right? Co is we can't possibly deploy something like this. I feel like that that is that's probably a I know people are working on it and companies are. But that would probably be like an area that seems to be like a middleman thing that we don't hear as much about. Yeah, I think a good example of this would be I'm working at Schwab and I've got a project I've been working on that I think is going to crunch a bunch of financial numbers and make it easier to get some insights on the market. But I have to deploy it. And then I deploy it. It's costly because we don't have those G H two hundreds yet. And it fails because I wasn't able to run it. I had the idea, but I wasn't able to run it until I put it in the data center. Those are the 85 percent of big data projects that failed those kinds of things. And so being able to do on a workstation just saves you all kinds of stuff where you can say, oh, it looks like this isn't going to work, but now I can tweak this and I know what to fix before you ever push it out and and put it in the data center, at least if I'm understanding this. I think so. Well, postdoc researcher in ergonomics at USC, Alexis Sushay, has an article on the conversation called virtual reality has negative side effects. New research shows that can be a problem in the workplace. Some of the negative side effects you might expect include headaches, eye strain, neck pain, you're wearing a thing in your head, things happen. Things you get from a computer monitor also, if you're not careful these days. But some of that is more specific to VR specifically, like dizziness or nausea, even muscle fatigue. The scientists have some ideas for how you can reduce the problems if you decide to use a headset as your monitor for work, maybe part of the time, maybe all of the time. These include taking regular breaks from the headset, limiting immersion to 20 to 30 minute increments and just stop using them all together if you feel sick, because nobody should feel sick while they're working. There are a lot of guidelines for a lot of different possibilities. There's no clear consensus about exactly what they should be because we're all different, especially when it comes to kind of that inner ear situation, you know, where some people feel sick and some people don't. So more research is needed to know which risks are the greatest and which methods are the most effective at dealing with those risks. Long term studies are just in the beginning stages. Yeah, because we haven't had these headsets for that long and we certainly haven't had them in situations where people use them for work for very long at all. So you can't do a long term study if you don't have a long term to study. I was very excited when I saw the headline of this conversation article thinking like, oh, so these papers that they wrote are going to have lots of great recommendations and they do. But there are like 90 individualized risks and some recommended guidelines for study for each specific for each one. I was imagining like, oh, we're going to get a list of, you know, five to ten things that are best practices when using a headset as your work monitor. And basically what Sushay is saying is we're not there yet. We don't know enough about the downsides to give good quality guidelines on how to avoid the downsides. We barely even know what all the downsides are. It's very individualized because different people have different downsides. So we don't know which ones are the most common. That's why there's 90 of them because they're like, we can identify when they happen. But we don't know which ones are the most frequent and the ones that most people will deal with. This is all something that could mean nothing if headsets never catch on and work or could be extremely important in a year if suddenly everybody's using the Vision Pro 2 in their workplace. Yeah, no kidding. I was just looking at apartments the other day, you know, and I'm always I'm always sort of like, well, I can do everything in my bedroom, but it would be great to have, you know, the the the office. Yeah, impossible, impossible. Yeah. You know, doing a VR thing kind of negates that need. If everything is going as to plan, you just sort of put it on. And now you're at the different room. Yeah, exactly. And that's great. I I do VR, I've talked about it ad nauseam on the show, mostly in physical for physical exercise, I love VR workout stuff. So the whole kind of like feeling sick, I is not working. I mean, I wouldn't say that when I put on a headset, I'm, you know, it's like full 2020, everything's totally in focus, but it's good enough. But that's like also me, you know, writhing around and doing stuff. If I were to be putting together DTNS in the morning, for example, that would be much more sort of, you know, it would it would matter. Like, can I see text, you know, am I copying and pasting in the right places? Am I feeling a little sick because of, you know, that sort of I don't know where my feet are type thing because because they don't exist in the sphere. I think that's all that's all really important stuff that we should be asking ourselves before we try to push this on a bunch of people who might not want it. Yeah. And I think it's important for it to be studied for for people to actually do workplace studies to say, OK, if you wear it for for eight hours, but you take a break every 30 minutes, does that mean you don't get the neck pain? Like this is the kind of stuff we never did with computer monitors. We all learned whether computer monitors and mice because we all got RSI. Yeah, because everybody just got RSI. And then something like, huh, my shoulder hurts and, you know, maybe maybe we need ergonomic mouse. Let's walk it back. Yeah. So it is interesting to see somebody saying, hey, in case these do catch on, you know, maybe we should try to figure out how to get ahead of this before a bunch of people come back to you saying, like, you know, I've got vertigo and it feels sick all the time. Yeah, you don't want that. Yeah, how many people are going to drop out of meetings because they had to puke because they had vertigo? Like that's that's a new one. You don't you don't have that. I mean, the answer is zero. So if it isn't, let's talk about it. Well, folks, on an entirely different topic, Android aficionados have a show of their own from Ron Richards and Huantui Dao. It's called Android Faithful, devoted exclusively to Android news and information. And it's great. It's fifth episode is coming up. I can't believe it's been around for five episodes already. If you have not already signed up, go sign up. You can catch it live right here on our YouTube channel and our Twitch channel, or you can get it on demand Tuesdays at 8 p.m. East Coast 5 p.m. Pacific or at androidfaithful.com every single day of the week. On Monday's extended show, we were talking a lot about A.I. and copyright in relation to Dungeons and Dragons. If you didn't hear that show, Dungeons and Dragons stated clearly that it would not let its artists use A.I. related tools to make art for the Dungeons and Dragons books. Now, we suspected in our conversation that might be in a large part because of issues of intellectual property. We don't really know whether training on data is a copyright infringement or not. A lot of people have opinions, but it hasn't actually been settled in court or legislation yet. And there's a couple of stories that bear directly on that today. Yeah, OK, so let's start with some good news. Open A.I. announced its GPT bot, which will crawl the web like a search index does and filter out paywall restricted sources as well as sources that violate open A.I. policies and also sources that gather personally identifiable information. Open A.I. hopes that other websites will let it crawl their sites in order to make its products like chat GPT work better. But if a website doesn't want to participate, it can disallow GPT bot in its robots.txt file, just like sites can already do that to stop being indexed for search. Sites can even customize directories to allow GPT bot to access some, not all, but some of its site. So that's good news. But a lot of folks are so wondering if Open A.I. has the right to use web information for training at all. Even if sites have left it open to be crawled, did they do that on purpose, which leads us to our next story? Yes, there's a lot of abuse of these generative models going on out there. One recent example, author Jane Friedman discovered that someone was selling generated texts under her name on Amazon. And it took her story going viral to get Amazon to remove them. That is just one example of the kinds of things that cause people to get very upset and people getting upset has now affected Benji Smith. Benji Smith launched something called Pro'scraft in 2017. It's good to note, especially if you see headlines about this, that Pro'scraft has been around since 2017. It had scanned more than 27,000 books and used an algorithm that is not AI, not a generative algorithm, but used an algorithm to rank them by various statistics. Those statistics included things like word count, occurrence of passive voice, the percentage of adverbs, and a measure called vividness. These are all things to help writers evaluate a text. The only sections of a book that were shown to customers were the most passive and the most vivid page of a given selection. Benji Smith, who created Pro'scraft, thought it was a useful tool for new authors to upload and compare their own work. He did not mind their work. He was just using the algorithm that was drawing from published texts to say, hey, you're about as vivid as Herman Melville or something like that. It was a way to improve the writing and it was incorporated into his Shakespeare with an X word processor so that writers could use it while they were writing. Smith said because he only published small pieces of text alongside summaries of statistics, he thought it was a fair use of the texts and so he didn't seek permission. Hmm. OK, then Monday, the Internet noticed and backlash commenced because that's how the Internet works. At which point, Smith removed Pro'scraft and said, quote, your feelings are legitimate and I hope you'll accept my sincerest apologies. End quote. He also noted that Pro'scraft never made money. He was never he didn't take VC money for Shakespeare. He was working on other jobs to pay his bills. He also said, quote, I would love to rebuild this library with the consent of authors and publishers. I truly believe these tools are useful for creative people. But now is not the right time. I understand and I am sorry. End quote. Smith's algorithm was not generative AI. As Tom mentioned, nobody noticed Pro'scraft for six years. But the sensitivity to such things has been ratcheted up by situations like Freedmen's, which we talked about just a few minutes ago. So do we think that Smith deserved the ire, Tom? Yeah, my my immediate thought is that. No, somebody noticed this on Monday and took it out of context, which is all too frequent. Somebody, you know, the mob mentality, all grab our pitchforks and our torches and go after the thing. And the reason that works is because there are bad actors who do things that look like Pro'scraft like Freedmen ran into. And people are tired of having to deal with them. So they just jump to a conclusion. Frankly, I think they really picked the wrong example here. It sounds like Smith was very sympathetic. The fact that he just took Pro'scraft down and wrote a heartfelt apology, didn't try to make excuses. She didn't try to say like, hey, you guys are so mad. I guess I'll take it down. Was like, no, I understand. I agree with you point just saying, I'm sorry, it was supposed to be fun. This was a labor of love. Yeah, I'd like to do it right. If that's at all possible, which, you know, I've already seen other people reacting like he's lying. That's what these startups always do. And it's unfortunate because I feel like not only is Pro'scraft a really interesting and potentially helpful piece of software for people, but could be made better with generative AI. If we can all figure out how to navigate the difference between responsible uses and irresponsible uses and more and more, it just sounds like people are like, yeah, I'm tired of trying to tell the difference. Get rid of it all. Yeah. Well, I don't know. I don't have a solution for that. I was I was not familiar with Ben G. Smith and Pro'scraft. I, you know, I have to say, I think that he handled it the best way that you can in a situation like this saying, hey, listen, you know, the technology six years ago when this launched was not what we're talking about today. Today, I am sorry, and I will not, you know, go forward using the same method. And, you know, I hope that people who enjoyed what he put out there maybe have suggestions for, you know, how this does not, sorry about that. My microphone got very upset about that. It took a dive. Yeah. Yeah. So it's it's a brave new world. You know, we're kind of right now as we go along. That's how I feel about that. It strikes me just now, just as we were talking about this, that there is a this is a great example of the difference between the attitude in the 90s and early 2000s, everyone would assume that Smith was an independent developer, the words like shareware were still in existence and would have would have rushed to his aid and said, like, how can we make this work better? Let's let's donate text. Let's, you know, fill him up with Gutenberg stuff that's public domain. Let's help him out. But we are so far away from that world of technology to the automatic tech. It must be bad if you say you're independent, you're probably lying. Big tech companies are behind everything and they try to pretend to be small people to fool us. That that seems to be the general attitude towards this. And that's a shame. Well, Tom, you know, it isn't a shame. Actually, I don't know. I think this might be also another could be a shame, depending on who you are going to Evo 2023. If you're not familiar, it's the biggest fighting game event of the year. Also a big deal for eSports took place in Las Vegas over the weekend. It's also run by Sony. So take that into consideration when I tell you the rest. Some of the competitors play Station Fives running all the games through various matches, which took some time. You know, you're you're kind of you're you're going hard for a few hours. They were melting their controller connectors so hot that they fried their USB adapters and a handful of posts seen on X and also Reddit lamented similar heat issues. The PS five only has one USB port on the front of the console. Two more on the back. If you have one, you already know this, but a lot of people still don't. So it might kind of make people wonder, is it the positioning of all the stuff that's making the heat exhaust not work correctly? Or are we just running these things too hard? Some players said even the connectors seem too hot. Yeah, the point being, there's two USBs on the back right next to the heat exhaust, right? So that that has melted at this game. We're again, they're pushing these things to the limit. This is not regular daily use, but also it's run by Sony. So their own equipment should probably be up to the task. Yeah, like I love to, you know, imagine, you know, somebody at Sony being like, let's just see. Let's just see. Because they have to know that this is a potential. It's not only melting cables. It's also like causing the USB assembly to come out because the parts around it have melted, you know, like that's crazy. Yeah, yeah, it's it's. I mean, listen, melting hardware, anything is usually a bad idea. No, it's a bad thing. Yeah, I'm not going to I'm not going to fight you on that one. I'm trying to come up with the upside to melting hardware. Unless you kind of do it, I'm so good at this. I melted my PS5. Hey, that's what Sony should say. It's like it's just how hot our event was. Yeah, just everybody was so enthusiastic. You were playing too hard. Yes, you were playing too hard. Yeah, why don't you take it? Take a step back, y'all have a milkshake later. And of course, this kind of thing does get confused because now you have a bunch of fanboys like, see, you should have used an Xbox or a PC. And I'm like, OK, maybe. But it's a Sony event. I think less than that. PS5 shouldn't melt USB controllers in any particular. No, no, yeah. Or if that's possible, Sony should say just a warning. Yeah, six hours in, things might get weird. They won't. They don't want to say that. You don't burn down your grandma's house. They don't want to say that if they're the sponsor of the evil event, though, because it would mean they couldn't use. Of course they don't. But they should. Yeah, I know, yeah. All right, let's check out the mailbag. Let's do it. This one comes from Oblo. This was in response to our conversation. We were talking about ice cream on last week's show. I think it was last Friday. Oblo said here in Thailand, durian ice cream is a real thing. I think we had all sort of like chuckled that. Why would anyone have durian ice cream? Because durian is a Roger brought up durian ice cream. Particularly aromatic fruit. Oblo says the real stuff, not bad. The artificial stuff they have at Dairy Queen is horrible. Apparently Dairy Queen has durian milkshakes, at least in some regions around the world. Oblo says, by the way, I love the real durian fruit. Yes, durian. I think my reaction to Roger is we were trying to come up with an ice cream that everyone would love. Not everybody loves durian. So maybe not the pick if you want widespread acceptance. As Roger's saying in our chat right now, it's pungent. Pungent, yes. That is a good way to describe it. It is, I think I said last week, it's an acquired taste. And someone was like, if you don't like it, it's not going to be acquired. You're not going to acquire. You may never love this. But people say that about, I don't know, kimchi, too. So, you know. True, that's true. It's the more you know, the more you know. Well, techno mentioned in our chat notes that today is National Custard Day. So any of you enjoying a nice durian custard out there, send us a picture. Indeed. Thanks to our patrons again. You make the show possible. So excited that we have new patrons coming in the door. Everyone make them feel welcome. If you haven't joined yet, now's the day. Patreon.com slash DTNS, because if you're a patron, you can stick around for the extended show. Should journalists just stop covering everything related to Elon Musk or everything he says? What we think and our own policies on such matters, stick around. Just a reminder, you can catch our show live because we do it live Monday through Friday, 4 p.m. Eastern, 200 UTC. You can find out more at dailytechnewshow.com slash live. We are back doing it all again tomorrow. I was God Johnson joining us. Talk to you then. This show is part of the frog pants network. Get more at frogpants.com. Diamond Club hopes you have enjoyed this program.