 Coming up on DTNS, the danger of autonomous weapons systems creeps closer, neural networks get closer to real neurons, and TikTok wants to turn viral recipes into food inside your mouth. DTNS starts now. This is the Daily Tech News for Monday, December 20th, 2021 in Los Angeles. I'm Tom Merritt. And from Studio Redwood, I'm Sarah Lane. From Lovely Cleveland, Ohio, I'm Rich Trafalino. And from a slightly overcast sky, I'm Roger Jang, the show's producer. We were just talking about that brief YouTube TV Disney dispute on Good Day Internet. If you want that and our other wider conversations, get Good Day Internet at patreon.com. DTNS. Speaking of Patreon, thank you to our top patrons. Today they include Johnny Hernandez, Hi Tech Oki, and David Mosher. Let's start with a few tech things you should know. According to analysts at IC Insights, chip makers are on track to spend $152 billion on new fabs and production equipment in 2021. That's 34% up on the year, and the largest annual growth since 2017. Contract fabs like TSMC, Samsung Foundry, and Global Foundries are expected to make up 35% of the spending. And then in 2024, they'll be like, what do we do with all this capacity? These tax authorities announced that the country's top e-commerce influencer, Huang Wei, otherwise known as Via, was fined $1.341 billion yuan for failing to report taxes from 2019 to 2020. That's about $210 million US worth of taxes there. Her social media accounts with more than 120 million combined followers have been taken down. Open Broadcaster Software, also known as OBS, Business Development Manager Ben Torrell, told Protocol that his team has found clear evidence that TikTok's live studio Windows app violates GPL licenses by using open source code improperly. The GPL license under which OBS studio is released requires any modifications made to the code to be released publicly under the same license. Torrell said that the project has reached out to TikTok to discuss the issue. LG announced the 17-inch Ultragear 17G90Q, calling it the company's first gaming laptop. Not going 12th gen, though. It offers the 11th gen Intel Tiger Lake HCPUs, up to an NVIDIA GeForce RTX 3080 Max-Q GPU, 300 Hertz 1080p display, 1080p, and an RGB keyboard arriving in the US and South Korea in early 2022. No pricing yet. The Hugo Awards, an annual literary awards ceremony for sci-fi and fantasy works. You're probably familiar if you're listening to DTNS, but if you're not, this year. The award introduced a category for video games. The games Hades won the first Hugo Award in the category, Spiritfarer, The Last of Us 2, Animal Crossing, New Horizons, Final Fantasy VII Remake, and Blaze Ball were also nominated. Blaze Ball got robbed. Hades, come on. Blaze Ball should have won. Anyway, let's talk a little more about a trend in machine learning that is a little bit of convergent evolution. Back in November, a group of researchers at MIT published a study in the Proceedings of the National Academies of Science, aka PNAS, trends in machine learning, giving insight into the mechanisms of higher cognitive brain function. The study implied that AI algorithms are undergoing convergent evolution with nature without anyone programming it to do so. Martin Schrimp, first author on the study, told InterestingEngineering.com, quote, we've had some success in modeling sensory areas of the brain, in particular with vision, so the team decided to use the process on language processing. They compared data for 43 neural network models to FMRI and ECOG neural recordings of humans listening to words being read. So you'd read out the words out loud. They'd scan the brain and compare that to the actual neural networks processing of those same words. The algorithms included glove, GPT, and BERT. They found that no matter how good the model was at doing what it was supposed to do, they appeared to mimic brain function for language processing with GPT being the most similar to how our brains actually work. Since GPT uses prediction, that implies that the brain relies on predictive modeling as well. However, the models did not predict brain activity for aspects of grammar and judgment, and that implies the prediction function is special somehow and possibly fundamental to language processing, whether you're a computer or a human. They also found that the more brain-like the processing was, the more it matched humans in other behaviors like the time it took to read the text. Shrimp said that models that are better at predicting the next word are better able to predict neural responses in human brains, and the models that predict neural responses can better predict behavior in the form of self-paced reading time. So in other words, you can predict some human behavior in this case. An interesting engineering emphasize that a key takeaway is to remember that none of these models were designed to mimic humans. They just devolved that way because it worked the best. Practically speaking, this could help advance projects to synthesize things. With vision, it could help to synthesize vision in the brain. For language, it could help synthesize language comprehension. In fact, we might even be able to stimulate the brain in the proper way to either see something or convert info from vision sensors into a format readable for the brain, and likewise the advance on language models could improve learning comprehension or language comprehension for people who have disabilities, possibly even able to translate languages on the fly. Now, at the very least, this all serves as an indication of how well models are progressing. And just one more thing to add to this. We might use it to train brains directly. PC Gamer reports on a separate story that a company called Cortical Labs have trained human brain cells grown in a Petri dish to play pong. A normal algorithm takes about 90 minutes to learn how to play pong. The dish brain did it in five. Now, it's a long way from teaching brain cells to be mediocre at pong to making them do complex behaviors. But as we just heard, we're starting to learn the right training models for human functions. So I don't know, maybe someday we can train a real piece of brain for something and then use it to repair damage or disease. Right. Yeah, I mean, besides playing pong, which while important, imagine, you know, I don't think I want brain surgery to to repair my pong playing ability. But yeah, but perhaps, you know, get to where you need to go in order to fix whatever is wrong or fix, you know, what might be wonky in a shorter correct time, correct vision processing or or language comprehension after a stroke, perhaps something like that. Yeah, now you're talking. What's interesting to me about this is also from another evolutionary perspective. There's always all these questions about like the the the original like evolution of language itself and how that process occurred in the brain and all of that. And I wonder if, you know, knowing that this prediction, the specific prediction aspect is like is really like this core base experience to to language. I wonder if that can advance also like kind of our understanding of how language developed. This is something, you know, like just following that. Yeah, I think it could also have a lot of things. The other thing, though, is really interesting to me is that a lot of these models are black boxes. Like we're looking at outputs of them and we're like, hey, these these outputs are really great. And not even to say like, these should all be open source, which that's a whole nother discussion. What's interesting, though, is like, even just if we were to see under the black box, it might it might be like not comprehensible to actually see like it would take a lot of time to figure out how exactly the models were to be clear. AI being a black box isn't because it's not open source. These are open source models, right? Excuse me. The AI being a black box is because the data is trained and then it comes in an output and we actually just don't know how it does it because it's not designed to show us. Yeah, that's that's the weird thing to me is like, we're we're the outputs on these are so good that they mimic the brain. But we don't 100 percent know like we know how they're programmed. How it got from that code to mimicking the brain. But it does feel like we're on the right track if we're mimicking the brain because that, you know, that's been refined by. So it seems like a pretty good system, right? Shout out to the brain. Yeah, good job, brain. Well, if anyone here has been on TikTok long enough, they know that there are a lot of questions that also involve the for you page. You've probably seen fair share of food and recipe videos. Even if you don't really care about it, it's very prolific on TikTok. These types of videos have routinely sparked runs on food items over the past years, people saying, oh, it's the new TikTok craze. Got to make this meal type thing. And flu or influencers have been part of the restaurant industry for the last few years, kicked off by the popularity of the McDonald's Travis Scott meal. I don't remember that too long ago. So late last week, TikTok announced it's getting more directly involved in this whole thing, launching the delivery only TikTok kitchen and partnership with virtual dining concepts, about 300 restaurants across the U.S. are planned at launch with more than 1000 expected by the end of 2022. These will operate out of national chains owned by virtual dining concepts, including Bukitabepo and Bertucci's. The menus will change quarterly based on TikTok's most popular food trends like baked feta pasta, which, by the way, I tried, pasta chips, corn ribs and smash burgers. TikTok says that profits will go to creators of the menu dishes. Now, see, that last is very interesting. I got I got a little confused when they brought in that company, virtual dining concepts. And I'm like, so they're just delivering Bukitabepo. But no, virtual dining concepts owns Bukitabepo. They're going to use Bukitabepo's kitchen to make the viral stuff. And then I started thinking like, oh, so this is like when you rip off a TikTok dance and put it in a video game. And then you get in a bunch of trouble because you didn't create the owner of the dance. But TikTok says, no, we're going to we're going to have we're going to have a royalty system somehow. I'm curious how that's going to work. I'm so curious about the baked feta, you know, or Mr. Beast or somebody can can get a little coin if they're if their menu goes viral. Well, and OK, so I was trying to, you know, unpack this earlier where I was I was, you know, over the weekend, I saw this this news item and I was like, OK, well, sure, viral meals will definitely get attention for a short period of time. Probably not. Probably not sustainable. But OK, short period of time. So yeah, if if the folks who are working for virtual dining concepts have the capacity to help you get that meal. And then, you know, what is the churn a couple of weeks and then nobody wants that anymore? Well, OK, in that in that restaurant back, you know, at the beginning, do they have the capacity to just sort of like make whatever stupid things that people want every five seconds? It seems like a really bad idea to me. If they've got a wide enough variety of restaurants, that's that's the genius of having been like, oh, yeah, well, no, we're using the book at a Beppo kitchen for this. We're not creating a separate kitchen. It's like, oh, yeah, so you just go to whichever restaurant happens to keep the ingredients on hand. You can move pretty fast with that. It only gets a day or two. What's remarkable to me is is TikTok continuing to find to be very aggressive in finding ways to monitor, like to let creators monetize, which outside of all of like the recommendation secret sauce that they have, that that, you know, is the engine of their growth and of their popularity. I think that that's the other thing. This is to keep the content chain churning to me is like the idea that your profits from this, you know, you can have a link back to order through your TikTok virtual kitchen on your video that where you you showed off the Smash burger or the TikTok pasta or something like that. And then you get a cut of that at the end of the day. That to me is like, I mean, they're not moving like super fast on this, but a thousand restaurants and by the end of the next year is like, that's a lot of restaurant. I would call that fast. Yeah. Yeah. Yeah. I mean, just platforms. It's lightning fast. Yeah. I think it's incredibly genius that TikTok is like, oh, yeah, no, the user pattern here is sea baked feta on TikTok decide want to make baked feta look at steps, say, that sounds hard. TikTok says, Hey, would you rather we just make it and deliver it to your house? TikTok user says, yes, I would rather you just make it for me and deliver to my house, you know, step three profit. That's pretty smart. If it works, I mean, you're not going to get any views just getting food delivered to your house people, but you got to make it yourself. Yeah, that's true. The creators don't. You're never going to become a creator if you're ordering it. That's a good point. But baked feta is also really good. So, you know, enjoy. Well, spoiler, the internet is lousy with spoilers, but across blogs and better moderated subreddit, some basic decorum around spoiler exists. There are spoiler tags. You see those on Goodreads and Discord. They can hide spoilery information at first glance and let you uncover it if you don't mind or won't be spoiled because you've already seen the thing curating your Twitter and RSS feeds goes a long way to avoiding spoilers as well. That's a lot of work though. Sadly, the dreaded group chat can still be the Wild West and you just never know when a sort of friend of a friend in a big old group chat just spoils it for everybody. So the messaging app telegram is working on a way to hide sensitive content in a chat. According to a post on the telegram subreddit, the messaging app is testing a way to hide messaging test behind a fuzzy pixelated placeholder that will have a small icon next to it that looks like an eye. And if you want to see it, you click on that and it'll unfuzz it. Only long pressing will reveal the text though. It's not clear if pictures and videos could be hidden to or if this is just limited to channels as opposed to one on one chats. A lot of details yet to come. But what is what has become the norm in many places finally come into telegram. Oh man, I cannot wait. There there's a person that I follow who will not be named very popular on Twitter and routinely spoils things and it's stuff that I've already seen, you know, because I'm, you know, it's like succession, the new episode. Let's talk about it kind of thing. Like I'm watching that already. So like it doesn't actually affect me negatively, but I'm like, what are you doing? Why are you doing this? You have like a hundred thousand followers on Twitter. Something like this. And I understand that we're talking about telegram. So it's a little bit different. But something like this that can just sort of be like, hey, I don't need to stop talking about what I care about. I don't need, you know, a bunch of people to mute me or or block me, but I could have an option to just like obfuscate what I'm saying because it's it's, you know, sensitive material that makes so much sense. Though the trick to this though is one I feel like I almost think there needs to be like a a list of words or phrases. Like if you're a moderator of one of these channels, you know, where it's it's around a topic or like a geographically tied or something like that, where you could say like, Hey, we're not going to be talking about any Spider-Man spoiler. So if you say Spider-Man, it's going to be, you know, we're going to apply this automatically before that. Then the other thing is I can imagine the reactions like not everyone's going to remember to fuzz out their response to, you know, whatever the movie spoiler or whatever the thing is. So it's like people reacting to it. I could also see people being like, no, don't don't react to it. We need to now start a separate thread between me and you. I mean, this already exists in discord. It's already exists in Goodreads. And yes, all of those problems can happen and occasionally do, but they're rare. Usually it's somebody new. They're like, Hey, if you're going to do that, here are the spoiler tags. Here's how you do it. It's just a little bit of a learning curve. So yeah, you're right. It's not perfect, but generally it does seem to work when it's implemented, don't you think? Well, it works. But I think there are enough people who just don't follow the rules. I feel like, I don't know, one year from now, wouldn't it be nice if, if because we all live in a world where we're not watching the same thing at the same time, very rarely we are. And there are just, you know, if you're, if you're, if you're cool, you're doing it the right way. Um, if you're not cool, then nobody's going to follow you or be your friend anymore. Yeah, that seems to be what happens on discord and good reads. The rare times that a spoiler is not hidden behind the tags in my experience is somebody new. They kind of get called out. Someone teaches them how to do it. And then that person either leaves or, or doesn't make the same mistake again. I feel like this is good. I think this is progress. Well, if you want to hear us say something spoilery on the show, good luck, we're not going to do that. But you can let us know what you do want us to talk about in our subreddit. Please use the spoiler tags. We're appropriate submit stories and vote on them at dailytechnewshow.reddit.com. Well, every five years, the United Nations Convention on certain conventional weapons meets to debate bans on certain weapon systems. In the past, it banned landmines, booby traps and incendiary weapons. Last week, the group met to discuss banning autonomous system of weapon systems, but failed to reach a consensus that would have started negotiations on an international treaty to govern their use. So discussions continue. The convention defines autonomous systems as robots would lead the weapons that select and attack targets without human intervention. And a story posted on the conversation on Monday, Professor James Dawes of Macalester College, a human rights researcher who focuses on weaponization of AI outlined four primary dangers of these systems that we're talking about. The first is misidentification, shooting at the wrong thing, shooting at the wrong person, bad news. Dawes doesn't think the machines will make more errors than humans. In fact, they may make fewer errors overall. The danger is the scale, the scope and the speed, which can be well beyond what a human could do. A small percentage error can result in much higher death and also damage. Dawes also warns of both low end and high end proliferation of these systems. The low end danger involves systems becoming cheaper and more effective. The high end danger is about nation states with increased sophistication, possibly leading to a race or high race for higher lethality. Sophisticated weapons change the cost benefit analysis of war by reducing the need for a risk versus a sides on soldiers while claiming to have ethical governors that reduce collateral damage. This could increase asymmetrical wars perpetrated by countries that possess autonomous systems against those that do not. When the final danger is assessed, the final danger seems to be a lack of accountability for autonomous systems. And when an algorithm goes bad and it will, does the Geneva Convention apply without adaptation? This could lead to a breakdown of international laws of war. Yeah, go read the conversation.com story about this. We'll have it in the show notes at DTNS. It's a good kind of overview of the problem as it is right now before it's a problem. This is the conversation that needs to happen when it needs to happen. Autonomous weapons systems are still in development. They're not widespread. They're not proven weapons yet. That's when you want to have this. And so it's somewhat disconcerting that the convention wasn't able to put in guidelines because the convention is just a way to say, hey, everybody, this is a problem and you need to decide what the rules are for using it. We think the rules should be this. If you don't even have that, then you're missing that first step. This isn't the solution, but this is the first step toward the solution. So I don't this isn't a story where it's like, here we go, dragon our feet. It's too late. This is a story of Professor Dawes saying, OK, it would have been better if they they had agreed, but we really need to agree now before these become widespread and because they will. And because if suddenly they do become widespread, it could threaten, like you said, the rules of engagement when somebody's like, hey, but, you know, it wasn't me that caused this problem. It was a malfunction, not my fault. You know, you need to have rules about when when that can be applied and when it can't. Yeah. And I think by by continuing the negotiations again, like the conversations are going is not saying this isn't going to happen just to reiterate that just not happening or like we're not getting the process of signing those agreements and stuff like that. Now the the thing that stands out to me is kind of that proliferation use case where it's like because we don't have these treaties in place, then there is like everyone feels like they have to keep up on the the weapons research, you know, on that aspect, which can lead to all of those knock on effects and make I guess make it harder to avoid them down the road. And that to me that to me was kind of stood out in this article. No, I think it's it's a good time for this sort of thing to get more attention. I mean, it's it's not the earliest we could have done it. Don't get me wrong. But but now is a really good time to figure this out. So I definitely think it's worth sharing this around because as Amos points out in a chat room, this would affect status force agreement, law of armed conflict. There's there's so many things that are dependent on this. Once once you get these guidelines, then you have to start modifying other agreements that depend on those guidelines and enforcing them. Yeah, yeah, it's a whole thing. All right, Nika Monford is back with two inventors to tell you about one, an engineer who helped make remote work possible and the other an inventor who advanced modern vision care. Hi, this is Nika Monford, a.k.a. tech, savvy diva of the Snobbo West show. In the teching while black segment, I will bring awareness to a black technology leader who is advancing the tech community through their innovations across the spectrum of industries. Today in teching while black, the spotlight shines doubly on both Marion Kroak and Patricia Bath. They are the first two black women to be inducted into the National Inventors Hall of Fame. This honor has been bestowed on only 600 other innovators, both living and deceased. Engineer Marion Kroak, a tech pioneer who has helped make remote work possible. Currently, the vice president of engineering at Google, in which she leads their research center for responsible AI and human center technology, she developed voice over IP when she worked for AT&T in the 80s. She also holds a patent for the technology that allows text based donations which were used in relief efforts for both Hurricane Katrina and the 2010 Haiti earthquake. Furthermore, she holds a patent for the text voting system used in American Idol. Overall, she has more than 200 patents to her name. While being a member of women in technologies in a national hall of fame, she is a fierce advocate for STEM education in women and girls. Dr. Patricia Bath, a pioneering ophthalmologist whose work reshaped cataract surgery. Dr. Bath invented laser FACO, a minimally invasive device and technique that performs all steps of cataract removal. Also, she co-founded the American Institute for the Prevention of Blindness, as well as the ophthalic assistant training program at UCLA. So now if you want to talk about first, she is the first black female physician to receive a medical patent, the first black woman to complete a residency in ophthalmology at NYU, the first black woman surgeon at UCLA Medical Center and the first black woman to chair an ophthalmology residency program in the US. Now, if that isn't enough, she also received five patents over the course of her career. These two phenomenal women have most certainly earned their place in history and are more than worthy of celebration. History is being made in real time every single day, so let's celebrate it now. When we are aware of all innovative voices, especially those in underrepresented groups, the tech community thrives. As we wrap 2021, tune in next year where we will highlight a new slate of black tech innovators. Thank you, Nika. Of course, you can catch Nika Monford on Snobo S-Cast every week talking about the latest Apple stuff with Terence Gaines, aka Brother Tech, and also she'll be coming up on our 2022 predictions show at the end of the year. See what Nika is predicting along with the rest of us for the coming tech year. Well, it was good to hear what Nika is predicting because she's a smart cookie. If you're also a smart cookie, you might play Minecraft, you know, if you do, if you don't. Well, that's my cookie. But if you do, building, functioning, computer processors in Minecraft is not new to you if you're actually familiar with the platform. These are made using the unique redstone material in the game to create logic gates. Some users have gone as far as to make these look like real-life PCs. Minecraft builder Sammy Urie spent several months on a particularly complex processor, dubbed the Chungus II, or Computational Humongous Unconventional Number and Graphics Unit, because, of course, that's what it meant. This simulates an 8-bit processor with a 1-hertz clock speed and 256 bytes of RAM. Teeny-tiny makes it powerful enough to play interchangeable 2D games like Tetris or Snake in Minecraft, with each program also virtually building into the game to look like a giant game cartridge. Ha! These are always cool, aren't they? Yeah. Love them. We're approaching Zillog Z80 status. I can't wait. Yeah, yeah. This is Nokia candy bar phone level. You can play Snake, right? Not bad. All right, let's check out the mailbag. Let's do it. This one comes in from Greg. Greg says, I was listening to last week's episode that had a conversation about technology solving a counterfeit goods problem. So I thought of the whiskey industry. Greg says, there's a big problem with this. There have been spectrometers that can analyze right through the bottle, but that might be more of a Sotheby's kind of thing. The McCallan has used a unique quirk design that doesn't fit the same one once open and has a QR code that sends a browser to a page on that particular whiskey. This is all stuff that Greg says that the whiskey industry is trying to do to get out of counterfeit issues. Greg says, but I just read about Buffalo Trace using on certain bottles NFC tags inside the quirk, which can not only verify on to on to authenticity, but can also tell you if see if the sealed looking bottle that looks sealed might have actually been opened. Very cool idea to try out at least. Greg says, I'm amazed that blockchain hasn't been added to that as well, but anything to make me feel better when I finally snagged that bottle of Pappy. So you can trace your Buffalo Trace with the NFC tag. Yeah, basically make sure it's funny. You said the blockchain thing. I had my mouse over. I had my mouse over the C in NFC, and I thought it was going to say NFT, and I was like, oh, no, there is blockchain. It's too late. It's good stuff. Thank you. And I will now I mean, who doesn't love the McCallan? It's it's just like a basic building block of Scotch appreciation. So I'm going to have to look for that unique cork design now. Yeah, I mean, I guess I'm not really a person that likes the McCallan, but I love the idea of something that is valuable, being able to be traced. And I will accept you next time I mentioned the McCallan. I apologize. Thank you, Tom. Appreciate that. If you if you have feedback on anything that we talk about, could be whiskey or Scotch related, could be but anything. Please do send it our way. We love to get your feedback. Feedback at DailyTechNewShow.com is where to send that email. We also have a brand new boss and we would like to thank our new boss, Stefan. I'm going to go ahead and think that you're like the SNL character, Stefan. Stefan just started backing us on Patreon. Thank you, Stefan. Maybe Stefan hard to say. Either way, we're very appreciative that you started backing us. Please don't leave. Please don't. Please don't. I mean, we will thank you every day. If that's what it takes, thank you, everybody, for being part of the show. By the way, we're live Monday through Friday at 4 30 p.m. Eastern. That's 21 30 UTC. You can find out more at DailyTechNewShow.com slash live, and we will be back tomorrow with Chris Ashley. Talk to you then.