 This 10th year of Daily Tech News show is made possible by its listeners. Thanks to all of you including Mike Aikens, Norm Physikus and Chris Allen. Coming up on DTNS, Meta seems to be growing and it seems to be because of AI. A way to watermark text from chat GPT and real-life tractor beams. This is the Daily Tech News for Friday, January 27th, 2023 in Los Angeles. I'm Tom Merritt. And from Studio Redbud, I'm Sarah Lane. Also from Los Angeles, I'm Lamar Wilson. And I'm the show's producer, Roger Chang. But there's so much Los Angeles that Lamar and I are able to comfortably share it. We do well with it. You say, your side, I stay on my side. That's right. We never cross. Who are the majority of the show today? Congrats. We win! Alright, let's start with the quick hits. Where's that button? There it is. Google agreed to make changes to its services to come into compliance with the EU consumer protection laws. Google flights and hotels will clearly show when it's displaying services from Google directly and when it's acting as a third-party intermediary, as well as make it clear that Google Hotels does not verify its reviews. The Google Play Store and Google Store will provide upfront information on delivery costs and availability of repair options, as well as provide additional company details in the Play Store, such as business names, addresses, and contact info. Google will also clearly clarify how to browse other countries' versions of the Play Store in the EU, letting them change a country of residence once per year. Intel reported it earned $0.10 a share on revenue down $32 on the year in Q4. That revenue was $14.04 billion if you keep in track. What's really more important than the numbers themselves is they missed analyst estimates. The analysts say you've got to hit these numbers and if you do, they're great. And if you don't, everybody's mad. This marked the fourth consecutive quarter of declining sales. That's not good. Intel's client computing group, which includes the PC chips, that's kind of what they're famous for, fell 36% on the year to $7.68 billion. Demand fell mostly in consumer and education markets. That isn't too surprising. Its data center business saw revenue down 33% to $4.17 billion. That's a little more concerning. And its network and edge segment fell just 1% to $2.06 billion. Google began rolling out the ability to require a user's fingerprint or a pin to view incognito tabs in Chrome for Android. Of course, quitting Chrome will still remove all incognito pages. The feature came to Chrome for iOS in 2021. Apple analyst Ming-Chi Kuo has sources who say Apple directed most of its engineering resources toward developing three nanometer, three nanometer sized Apple Silicon chips for the foreseeable future. The company hopes these will enter mass production by the end of this year. As a result, though, Kuo says, looks like Apple may have stopped developing its own custom Wi-Fi chip. There have been a lot of reports about that. And therefore, will likely continue using Broadcom chips for the iPhone 15. Therefore, Broadcom stuck up, I think, went up. Good news for Broadcom. Little more Apple news or potential Apple news. The information sources say that Apple plans to let users build their own augmented reality apps for its upcoming headset by using Siri. For example, asking the Assistant to help build something based on scanning and importing real-life objects, like a pet, without having to do all the design from scratch or no code. See, I always go to pets, but it could be a lot of real-world objects. This would be similar to what Meta is already doing, the Quest headsets, let users build 3D environments within horizon worlds, and even iPhone users already have some of this functionality as part of Apple's other augmented reality and virtual reality efforts. Now you know why I'll be chasing Sawyer the dog around later with my iPhone to scan him. Just taking a nap. Alright, let's talk a little more about Meta. Wall Street Journal reviewed internal documents from Meta and conducted interviews and wrote a big story about the future of Meta that was actually fairly positive. Meta is bouncing back from its bad year. It is dealing with the fact that it's getting less personal data from Apple users, and in fact, using less personal data in general to target its ads thanks to some machine learning. So instead of gathering your data and using that, they're using machine learning to predict based on what it does know what kind of ads might be useful for you and productive for the advertiser. In November, VP and de facto head of Facebook, Tom Allison, wrote in a memo to his staff about another piece of AI that's helping. Our internal data indicates that Meta has grown to a meaningful share of short form video. A lot of that is due to algorithms figuring out what reels to show to users, both in Facebook and on Instagram. In October, execs reported a 20% gain in time spent watching reels. Now Lamar, I know you make content for TikTok reels, etc. Are you noticing that? Yeah, absolutely. It's funny this came up because I'm noticing a larger gain in views. I'm noticing more people are in regular conversation are talking about Instagram reels. It used to just be the de facto name TikTok, like Kleenex, right? But I'm hearing people say, no, I'm on Instagram reels and one creator recently said he finds the content is better. And that may be a part of the algorithm working better on that. So yeah, I've been uploading more on there. I'm seeing more views and I'm seeing just hearing more people saying that they love it, you know, in everyday life. I'm hearing more people talk about shorts too. I don't know if either one of you have heard that too. I have too. Yeah, definitely. I mean, not real shorts. No reels. Yes. In fact, I know content creators who were like, yeah, TikTok was fun for a minute. But Instagram is where all the people really are and have been for a while and will probably continue to stay. And I think that's a really interesting thought is, okay, well, TikTok has been on the upswing for the last couple of years. I mean, who can stop TikTok as far as, you know, good content that I may want to see based on the algorithm. But Instagram is kind of where I hang out more often. So there you go. Yeah, I'll say this too, that Instagram reels particularly used to be like whatever, you know, whatever got hot on TikTok. It was just like a dump. Just get cross posted. Yeah, just to see what would happen. And they're still cross post have I quote cross post, but I'm starting to see some native things that, you know, are happening or occurring on reels themselves. So they're starting to get their own people, you know, their own creators. I love what you said. I forgot your name. Rich Truffilio. I would completely blank for a second. No, Sarah, hi. Hi. I love what you said on that point, because Instagram, I've always thought is my home. I can write stories there and I feel like that's like my video Twitter and that's like the audience already built in audience is already there. You know, speaking of stories, I need to make more because I noticed the story mentioned that stories creation ran 10% below. And I definitely feel that I've seen more people that I follow are not making as many stories and I wonder why that is that are we focusing on time somewhere else. Maybe on Twitter, like real controversy. Maybe people are seeing that reels get more views and they're like, well, I guess I'll make a real instead of a story. Well, stories also, and I don't I don't make stories ever, but I occasionally look at my friend's stories. And there is a little bit of anxiety built into that because it's like, well, okay, so this is time sensitive. If I look at it, you know, within the 24 period before it goes away, you will know that I did. Not that I'm trying to creep on anybody, but there's a little bit more that I feel is asked of the audience member in order to consume stories. If you're looking at reels or just post in general, then it's a little bit more like Instagram's email where you get to it when you get to it, but it's not so much real time. Oh, I like that. That's a good. Yeah, I like that. I wonder if there's a generation breakdown here too, where the oldest users are watching Facebook and they're seeing Facebook reels. And then the millennials and us Gen Xers are using Instagram and we're seeing reels. And then the younger set are more likely to be looking at tiktok because you're right. You mentioned that earlier Lamar tiktok is not falling down. It's not declining. It's still groaning. I think this is interesting. I also want to circle back to the idea that using machine learning has helped meta bounce back on the ad targeting because it's probably going to be better than using the personal data. It's certainly better for us who want to guard our personal data. And I wonder, given their success in using AI type stuff for both reels and for advertising and for other things. The Wall Street Journal article said Meta's capital expenditures more than doubled most of it on AI over the past two years. Maybe meta doesn't become a metaverse company. Maybe its future is as a machine learning AI company where we're, you know, 10 years from now we're talking about, oh, meta. Yeah, they used to be into social networks, but they're an enterprise level AI provider. Now that's where they get that's where they get all their money from the way we talk about Microsoft getting all its money from Azure now. That's a really good point. Yeah, they can figure out news to it, you know, and get, you know, people are, you know, there's companies back on board with that. AI is great for determining. They're selling their ability to do that to other platforms as well. I think that would be smart. Well, speaking of AI, chat GPT, you may have heard about it on this show, has caused schools and teachers to in some cases ban its use on tests. And people are in general worried about it being fooled by generative text. A lot of folks have been touting different ways to detect if text is indeed written by a chatbot or a human. But let's be honest here, as soon as the detection method gets good, the language models will get better and so on and so on. So we should all give up kidding. But MIT technology review has an article out called a watermark for chatbots can expose tech written by an AI. Scientists at University of Maryland made a watermarking algorithm and tried to including it in a section generated by Meta's open source language model, OPT6.7B. They said they were able to detect watermark texting, watermark text really well. The paper has yet to be peer reviewed, but the method is an interesting one. And you can play with it. And they'll release the code February 15. But Tom, let's explain how that works. Yeah, so language models, a large language model, which is what ChatGPT is using, generate one word at a time. So we like to think they're composing it the way we do, but they don't. They pick a word and then they look at their internals and say, what do we think the best word to come next would be based on the prompt we've got? So you can use that. The watermarking algorithm divides a language model's vocabulary into what it calls red lists and green lists. Each time a model is choosing that next word, the algorithm encourages it to choose one from the green list if it can. Now, it's not always going to choose from the green list, but it will do it often enough that more than half the words will be from the green list. And the text will still read very well. Here's an example of why that can differentiate machine generated text from human text. Here's the example. After it chooses the word country, let's say the words, we're going to oversimplify here, the words it could choose are pond, pool, swimming hole, and water hole. They would all make sense given the context of whatever prompt it got. And let's say they all have equivalent scores. So it's not like one is much more appropriate. If pond is on the green list, then the way this watermarking algorithm would work, it would choose pond. Whereas a human, which tends to write more randomly, might be liable to choose water hole or swimming hole or pool. If you repeat that process hundreds of thousands of times, depending on the length of the text you're writing, you'll see the machine language thing use green list words so much that it would make it unlikely that a human wrote the thing. Does that make sense? It makes sense to me. And it's funny because, especially because, you know, we write a lot of things for a DTNS five days a week that I am constantly being like, what's a synonym for pond? You know, let's make it spicier because I'm a human. The algorithm doesn't necessarily think that way. So you may get more, you know, text that you're sort of like, okay, maybe a little repetitive, maybe it sounds like the algorithm just wants to go a certain way. But that is a really interesting differentiator between humans who are trying to be creative while saying the same thing all the time because that's what we do. And a machine that doesn't necessarily understand that. Yeah, you mentioned synonyms. It would be interesting to see how this works in other languages, right? Because our languages quite complex, but so are other languages and would it be able to catch those variants? Well, it's a great point, Lamar. English particularly has a lot more alternate words and synonyms than a lot of other languages. So your mileage may vary depending on what language you're doing this in. But I do think it's clever because we can already analyze word choice and figure out if somebody actually wrote something. There was an author who pretended to be dead and then was writing things on Facebook and they discovered her by doing text analysis and going, wait a minute, she always uses this word and she always, you know, interpolates these two phrases and that sort of thing. So I love this idea, which is to say, let's create a style in which chat GPT writes or anything like it. That'll be detectable. And the amount of work you'd have to do to change it not to remove the watermark would mean that you've actually done the work to make it original to edit it to be your own copy. So it would be self defeating to try to remove the watermark. I guess the biggest hurdle here is you would have to get every usable chat GPT like tool to add such a watermark to its system. Open AI says they're working on something like this, but they're very secretive about stuff. So we have no idea what they're doing. And who's to say that that some open source thing comes along that decides there or that makes it so you wouldn't have to have that element because you could compile it yourself. Yeah, listen, I am all for this and I know we get a whole discussion about about, you know, chat GPT as you already have but I just want to wonder just ponder. Where does this go because as we as Sarah mentioned beginning, as one detection method gets put out there, somebody else is going to counter it. And do, do we just say at some point, hey, let's just give up and focus on, okay, students are going to use this. So let's teach students in a different way. Yeah, or let's just let's have expectations in a different way. If you're just trying to catch student cheating, then yes, you probably it's probably beneficial to figure out how to create curriculum that accounts for the fact that they can use these tools and incorporate those tools the way we have with calculators to be different the calculators but but there's a there's a way to do that. I think there might be some other beneficial aspects of watermarking this way to just be able to tell if if something was machine generated or not in a lot of other ways. If you have ideas about ways that that would be beneficial email them to us feedback at daily tech news show dot com. Also, if you see stories about stuff like this and you want to let us know in case maybe we'd like to talk about on the show we'd love to get your ideas submit them and vote on them at daily tech news show dot reddit.com A paper published in the journal Optics Express is grabbing some headlines because it describes a laser beam capable of pulling objects like a tractor beam, for example, you might say, in what way can it do that universe today is Evan Goff, perhaps Evan Go reported on the article last week and gave a good breakdown of the finding basically scientists have used optical tweezers, which can move things the size of an atom, very small things around. But haven't been able to move larger things, aka macroscopic things. So Tom, what did they do exactly? Yeah, this is not the first tractor beam, like you said, but it's the biggest one we've made. It's still very tiny. You're not going to be pulling the Millennium Falcon in with this one. What they did is a laser can push an object by heating up one side of the object. The difference between the temperature on the two sides of the objects is something called the Knudsen forts. I think it's just Knudsen, but it's spelled with a K in low pressure atmosphere. So you'd have to have it lower than Earth. They had to have a really low pressure gaseous atmosphere to test this. You can do that and it will move an object. The laser will push the object think solar sales. In fact, they were able they said they were able to do three times the magnitude of a solar sale. The scientists were using a graphene composite. So this wouldn't necessarily work on just anything. That's that's one lab limitation. But they could push that with the laser in low pressure and then figured out how to pull it as well. They coated one side with silicon dioxide, which reduced heat conduction, hit that with the laser. That meant the opposite side without the coating was heating faster. That caused gas molecules to be released from the side that was hotter. And so, you know, it's kind of propelled it forward. So they were pushing, they were hitting the laser on one side, but it was hitting the other side and then you were able to pull it with the laser. See, I, I hate them. This can only be done in lab conditions. I want to see. You gotta start somewhere, huh? I know, I know. It just it sounds exciting when we hear laser beam. But but yeah, lab conditions only and then, you know, like, very specific conditions, but it still sounds cool. It sounds really cool. And I'm with you, Lamar. I'm like, okay, lab conditions, but like, how was this going to help my life when we get better at this? How will laser beams control objects that They will make it easier for me to do my household chores, I guess. Yeah, we're we're decades away from that, probably, if not hundreds of years from that. But but is that the the only thing that makes scientific endeavors like that worthwhile? Well, of course not. No, there's I mean, just to do it is like, OK, this is this is the thing. Let's figure out how, you know, everyday tasks could be eventually tackled by, you know, this. Well, and also, also, it's not it may not be good for everyday tasks, but optical tweezers have allowed science to be able to manipulate atomic scale stuff in a way that they couldn't before, which allows them to do other work, which then ends up making advancements that do lead to iPhones and things that we use every day. So even if the connection isn't direct, this could do a lot of stuff. One of the things they think it would be useful for is space. So the the concentration of the atmosphere that they were using to do the lasers in was similar to the concentration on Mars. So immediately, everyone said, well, wait a minute, back in 2011, NASA was investigating the use of possible tractor beams in its Martian rovers to be able to bring in samples in a way that that was not capable, you know, with mechanical things that could break down or get grit in them or whatever. Would that be something they could do? Could you use that when you're flying something through a comet tail to be able to to grab a little stuff? Would it work on an asteroid? So there's all kinds of things we could learn by using this possibly as well. Yeah, that's why I get irritated when people say, like, oh, you know, why spend all this money and space is not going to go and doing things like it absolutely can have practical uses down the line. So let's let's not stop exploring. Let's not stop trying new things because it can and will have application at some point to benefit us. And also tractor beam, like they really work. It really does work, right? That's the goal. Yeah, we need some sharks, but the lasers. Okay, anyway. Yeah, don't give it to the sharks. Shark with freaking tractor beam. Yeah, that just makes it worse. All right, we got another we got another scientific advancement to talk about as well. Oh, we really did. So scientists at Sun Yat Sen University, that's in China and Carnegie Mellon University in the US, published a research paper. Well, they wrote a research paper that matter published demonstrating how they created a shape shifting robot that can switch between liquid and metal states to navigate a variety of environments. Now you might say this sounds like Terminator two. We're not totally at that level yet that these robots are they're small. They're actually modeled after see you see cucumbers in shape, but they can navigate obstacle courses. They can remove or deliver objects to a model of, let's say, the human stomach, not a real human stomach yet, but a model of one. They can also liquefy to escape a cage before reforming back into their original humanoid shape. I love this. The scientists use the material gallium gallium, if you're not familiar, can easily shift between soft and rigid states in ambient temperature. It's a soft metal, meaning that it has a melting point that's really low 29.76 degrees Celsius. That's about 85.57 degrees Fahrenheit at standard pressure. So that's less than, you know, a healthy human has in its body. So it could melt easily in your hand, for example. The scientists then added magnetic particles to the gallium to create a magneto active solid liquid phase transitional machine. If you're like, what does that mean? Essentially, it gives the robots mobility. I think the best part of these scientists is not only that they were able to do this, but then once they did it, they said, well, one way we should definitely demonstrate this and take video and put it on YouTube, not just in our article is to create a little Lego looking figurine and have it melt through bars as if it's Terminator getting out of jail. Yeah. Yeah, the robot's like, I can't fool me, man. I'm just going to liquefy and I'm going to go where I want. This is how you do science. Yeah, they say it's not Terminator, but no, I mean, we're on the path. I feel like they're saying, look, we know you're all going to say this is Terminator. So why don't we just go ahead and demonstrate it as if it was Terminator? Let's not pretend that we're not all thinking it. Let's just do it. Yeah, I love that. All right, let's check out the mailbag. Let's do it. This one comes in from Chris in Nova Scotia. Hello, Chris. Chris says, I just wanted to share my experience with chat GPT as a high school visual arts teacher. Chat GPT is excellent at generating drawing prompts after experimenting with it for a few weeks. I figured out how to get it to create prompts that tie directly into the outcomes for my courses. And it even does well as creating prompts for beginner and intermediate and advanced techniques. I'm so impressed that I'll be recommending it to my students as a tool this upcoming semester. And I'm looking at building an entire project around a prompt that it generated. Chris says, I know many teachers are worried about the impact of AI and education. These are also conversations happening at my school. But I hope that my experience can help inspire some out of the box thinking. Oh, Chris, this is great. And this goes right to what you were talking about earlier, Lamar. Absolutely. Yeah, I think it's great. Like, instead of making it a bad, you know, making it a bad evil thing, how can we make it a positive thing? It's out there. So how can we use it positively in high school? And I think that's awesome. I think that's always true of any big new advance. We're still having the debate about screens, right? Oh, are screens good? Are they bad? Not just for kids, but for us and screen time and all of that. But I don't think anyone would argue that a smartphone is entirely evil. And we should all just, well, take that back. Somebody's going to argue this. Not many people. Many people think that, yeah, there are good things about it. I think the majority of us understand that there are positives here, that if we can figure out the negatives are worth pursuing. So I think we're just beginning to deal with chat GPT. And yes, there are misuses that could be put to absolutely. And we, and I'm not saying we shouldn't pay attention to those and figure out how to limit those. But there are great examples of why it's not something you should just kill in the basket right now, right? You should look at like, okay, but it can also come up with these great prompts that Chris would never have come up with otherwise. And that's actually making his class better. It's going to make them better artists because they've got more varied prompts. They're going to start thinking about art in a different way. Yeah. I think we need to pay attention to all of the aspects of it. And this is a great example. So I appreciate Chris sharing it with us. Indeed. Thanks, Chris. And thanks to everybody who writes in feedback at daily tech new show.com is where to send that email if you say, I have a thought as well, we would like to hear those thoughts. Keep them coming. Also keep Lamar Wilson coming to the show because it's so much fun when we have you. Let folks know where they can keep up with your latest. Yeah. So I just, I got the MacBook 16 inch M2 this week with that big 96 gigs of RAM. So I did an unboxing of that. And I did a tour of my desk area. Also got a second Mac studio, studio display monitor. So got a nice two monitor setup, did a nice tour of that. You'd like to check that out is that YouTube.com slash Lamar Wilson. It was really fun. It was just really fun. Everybody was asking about every little device that I had on my desk. So that was kind of cool. Oh, fantastic. Love that. We also want to extend a special thanks to Tom S. Tom S. You know who you are. But for anybody else who doesn't know, Tom S is one of our top lifetime supporters for DTNS. And we like to thank some of our lifetime supporters. And today it's you, Tom S. Thank you. You could be a lifetime supporter like Tom S. Just start today. It's not too late. Patreon.com slash DTNS. And Sarah, tell them what they'll get. Well, if you, if you're a patron, you get all sorts of things. But one of the fun things is our extended show called Good Day Internet. Just a reminder though, you can catch DTNS live Monday through Friday at 4pm Eastern. That's 2100 UTC. Find out more at dailytechnewshow.com slash live. We hope you all have a great weekend. We'll be back on Monday with David Spark joining us. Talk to you then. Our mods. Beatmaster, W. Goddus one, bio cow, Captain Kipper, Steve Godorama, Paul Reese, Matthew J Stevens, a.k.a. Gadget Virtuoso and J.D. Galloway. Modern video hosting by Dan Christensen. Music and art provided by Martin Bell. Dan Looters, Mustafa A. Acast and Len Peralta. Acast ad support from Tatiana Matias. Patreon support from Dylan Harari. Contributors for this week's show included Lamar Wilson, Megan Maroney, Scott Johnson, Justin Robert Young and Chris Christensen. Thanks to all the patrons who make the show possible. This show is part of the Frog Pants Network. Get more at frogpants.com. I hope you have enjoyed this program.