 Coming up on DTNS, what we'll get from improved protein folding algorithms, new chips from Qualcomm and NVIDIA, and more moon rocks are on the way. This is the Daily Tech News for Tuesday, December 1st, 2020, the Canadian National Day of Podcasting. In Los Angeles, I'm Tom Merritt. And from Studio Redwood, I'm Sarah Lane. And I'm the show's producer, Roger Chang. Joining us, Dr. Kiki Sanford, host of This Week in Science Back on the Show. Welcome back, Kiki. Thank you so much. So good to be here. We have decided to celebrate the Canadian National Day of Podcasting by letting all of our Canadian podcasting friends have the day off from being on our show. That was so kind of you. Yeah. We honor their sacrifice. I don't know. All kidding aside, there's a really good Canadian podcast out there, and they all have special episodes out. So go check them out. There's some good stuff. Let's get into this first. If you heard Good Day Internet, you know that we're very hungry from talking about tacos. You can get that wider conversation by becoming a member at patreon.com slash dtns. But let's distract ourselves with a few tech things you should know. And food-related news there is. Uber announced it's completed its acquisition deal of Postmates. Postmates will continue to operate as a separate service with its own branding and its own front-end. Some back-end operations will merge, but a new combined pool of drivers will also merge. AWS announced that the re-invent conference, it will support the Mac Mini. EC2 Mac Mini instances are now generally available for developers who want cloud-based builds and testing environments for their Mac and iOS apps. EC2 supports Intel i7 Mac Minis, though they say M1 support coming in the first half of 2021. AWS also announced AWS Tranium, its next-gen custom chip that trains machine learning models. Reddit announced it averaged 52 million daily active users in October, which is up 44% from the same month a year earlier. This is also the first time that the company has disclosed that daily metric ever. COO at Gen Wong says, quote, we're sharing daily active users for the first time as a more accurate reflection of our user growth and to be more in line with industry reporting. We're focused on daily usership and increasing this number as we continue to grow our community and scale our ad business. Our chefs have prepared Facebook stories for you three ways today. First, please take note of a dedicated news tab launching in the UK in January. They'll be paying partner news sites to appear there. Second, Facebook's independent oversight board has chosen six cases to review regarding content removed by Facebook. These are its first cases. It has 90 days to reach its conclusions. And finally, the Facebook-backed Libra Association has renamed itself the Diem Association, as in Carpe Diem. It still aims to launch a series of stable coins backed by national currencies. Well, it was rumored now it's official. Salesforce announced it has agreed to acquire Slack in a deal worth $27.7 billion. Slack had lost 40% of its value since going public. Salesforce co-founder and CEO Mark Banyoff said that Salesforce and Slack will shape the future of enterprise software and transform the way that everyone works in the all-digital work from anywhere world. The acquisition will also move Salesforce into more direct competition with Microsoft. Well, congratulations, Slack. Good job for you. All right, let's talk a little bit more about some good news for people wanting to unlock things with their face while wearing masks. That would be me. The US National Institute for Standards and Technology, aka NIST, published data on Tuesday showing that facial recognition algorithms are getting better at recognizing faces behind masks. The report compiled tests of more than 150 separate algorithms, companies that make the algorithms voluntarily submit them to NIST to help with testing. So the tests look at something called false non-match rates, or FNMR. So the FNMR tells you what percentage of faces that should have matched did in fact not match. So an FNMR of 5% means 70% of the masked faces were indeed identified correctly. Of the nine algorithms that an FNMR for masked images of 5% or below, six were submitted later than July this year. So lots of algorithms are still improving at facial recognition of people wearing masks. Keep in mind that the tests are used using regular unmasked photos with digitally applied masks. So the variation in mask type, not representative of a real world sample, obviously see people all sorts of masks in the real world, but still a reliable indicator of improvement. Yeah, so I think you accidentally said 75 when you meant 95%. But 95% may not be the actual rate of success because digitally applied. But it still got better. We still have more algorithms that can do it. It doesn't mean it's coming into your device of preference soon, but this is a good independent gauge of where we are with algorithms. And there were some algorithms that did good at this before the pandemic. But since the pandemic, a lot more folks have focused on that. This concerns me a bit because of the potential for just facial recognition in public. I mean, great for unlocking your phone if you don't want to take your mask off. That's awesome. That's kind of a private situation. But what about in public and how many new ways are people going to have to come up with to be able to maintain privacy in the face of algorithms like these? Yeah, I mean, it's a really good point, right? It's a two sided coin because it's technology and it can be used for all different kinds of things. So the positive thing is like, oh, I'd be able to unlock something, but not have to take my mask off, which is more secure. The negative is, oh, if I was wearing a mask to avoid facial recognition software without being a criminal, I just want to protect my privacy, then that wouldn't do it. I guess my response to that would be the solution to facial recognition invading your privacy probably was never really going to be a mask. It has to be control over facial recognition usage, right? Right. Well, at this point, there's so many uses, you know, when I use Apple Pay at the grocery store, which I try to do as much as possible, because it's like that's one less thing that I'm touching, you know, when I'm doing the checkout. And it, you know, I have to use my passcode because my phone is like, I don't know who you are. It's not so lame. So, you know, little things like that is like, yeah, this is actually going to become much more convenient. But yeah, to your point, Kiki, it also means that these mechanisms are more powerful for all sorts of reasons. Yeah. And I don't think that means we should not continue to investigate and advance them. It just means we need to also be responsible in how we use them. Exactly. Or how we let our governments use them. And how we let others use them. Yes. Yes. Well, speaking of non-government uses of all kinds of things, guess who's making all the money these days? Not me. Who could it be? If you guessed Google, Facebook and Amazon, you win. They're making it off both your attention and your direct payments in different combinations. Let's start with the direct payments. Amazon announced Tuesday, the third party sales on its platform rose 60% this year for the period of Friday through Monday, the Thanksgiving Black Friday holiday shopping weekend. Amazon gets a cut of all $4.8 billion of all those sales that it racked up from third parties. It didn't tout how much it made itself directly. I'm sure it made up quite a bit if third parties were making that much. And beyond just Amazon overall, Adobe, which does some estimates of the marketplace in general, estimates online sales rose 22% on the year for Thursday and Friday and 15.1% on Monday. It was a banner year for online sales in this opening holiday shopping weekend in the United States. Three years ago, digital ads made up one-third of U.S. spending, and that's on the rise as well. A report from Group M, which is the world's largest ad buyer agency. They buy ads for companies, estimates that for the first time ever, more than half of non-political U.S. advertising spending will go to digital platforms by the end of this year. Group M expects $110.1 billion to come in for digital ads by the end of the year. That would be 51% of the expected total non-political ad spend. Group M expects that share to rise to 54% next year. And who gets that digital ad money? Facebook, Amazon, and Google together bring in two-thirds of all the money spent on digital ads in the United States. Facebook ad revenue rose 22% in Q3. Google ad revenue rose 10% in its most recent quarter, and Amazon's ad revenue was up 51% in its most recent quarter. Wow, that's quite a jump. Yeah, I mean, the whole Black Friday, Cyber Monday, and everything in between and just online spending in general when it comes to holiday seasons and other particular times around the world, depending on where you are, on the rise anyway. 2020, clearly a banner year. This is not surprising to me. I would think if I was a third party seller on Amazon, for example, just to take that story, and I did really well, I'm not too bent on a shape myself that Amazon also made a lot of money because so did I. But again, it's important to remember that this is a company that's not just letting third-party sellers do their thing for goodwill. It's to make as much money as possible and they're they're being successful. Yeah, I mean, there are those ethical issues. Whether or not these companies have good sustainable practices, how are they going to be helping humanity moving forward? But I mean, this is capitalism. This is the way it works. And they are making money for their investors and for their bottom line. There are things like the Amazon, what is it? The donations that you can be a part of. The smile, yes. And so there is a percentage of those proceeds that do go back to good causes. And that's useful. But those donations are also tax deductible on Amazon's side, which helped them actually save more money and make more money. I don't know. It's half glass empty, half glass full. And you can buy that glass on Amazon after responding to an ad on Facebook. Yeah, it's just a hard thing to choose these days because these are the places to be if you want to be successful. Yet are we choosing things that are going to help us moving forward? And it's a really tough decision. I think it's useful to occasionally remind yourself that we call them tech companies. But all three of these are advertising and sales companies that also do technology. Amazon does cloud services, AWS. Google does Google suite. Does Android. Facebook does Oculus. But they are predominantly advertising or sales revenue generating companies. Well, we have a couple of chip announcements to look at today. Some of these may surprise and delight you. First Qualcomm announced its next flagship smartphone processor, the Snapdragon 888. It'll offer a fully integrated five nanometer X65G modem instead of the separate chip required by the 865. So it's all integrated at this point. Should reduce power consumption. The sixth gen AI engine, which is an upgrade to the Andreno GPU and 35% faster image signal processor. More details and specs will come from Qualcomm itself December 2. So we will talk more about that tomorrow. But the announcement did come today. Also coming tomorrow in the chip world is the NVIDIA GeForce RTX 3060 for $399, 369 pound sterling or 348 euros. It's the most affordable in the RTX 3000 series and designed to outperform the RTX 2080 chips. The RTX 3060 Ti comes with eight gigabytes of video memory, a clock speed of 1,410 megahertz bootable to 1,665. So many numbers and a dual slot design with a new 12 pin power connector. The card can draw up to 2,000 watts of power, 200 watts of power rather. So NVIDIA recommends a 600 power watt supply. Keep yourself safe. It supports real time ray tracing, DLSS support for image upscaling and RTX apps as well. Now you might be wondering, okay, what do the reviewers say? The reviews generally positive, noting that it's an excellent card at the price point. CNET's Lori Grunin points out that the budget model of the new generation of graphics cards will start arriving in early 2021. But if you need to buy one now, she says you could do a lot worse than the 3060 Ti. I used to work with Lori Grunin at CNET. Her saying you could do a lot worse is high praise. I just want to point that out. Lori needs to be impressed to say something's good. So this is a good card. It's not even the budget card, but it's pretty affordable for the power. Roger, this is rightly impressing people, don't you think? It is. And every year, all the gamer hardware sites always publish the value card. In other words, the card that matches up with the most power per dollar spend. And this looks like to be that card where, you know, it's not the cheapest card. It's not the best performing card. But for dollar, for the per dollar you spend, this is the most performance you're going to see. And that means a lot for people, especially. I mean, when a lot of these cards are hitting around 500, 600 bucks new, I mean, $400 is definitely infinitely more affordable. Yeah, this is not going to be easier to get the other Nvidia cards. I would expect that's probably going to sell out pretty fast as well, which is kind of the problem with these. But if you can get ahold of one, it's a great value. Hey, thanks to everybody who participates in our subreddit. You can submit stories and vote on them at dailytechnewshow.reddit.com. Yesterday, Rich Straffolino and I told you about DeepMind's Alpha Fold system. That's the one that uses attention-based neural networks to predict protein folding structure in a matter of days, avoiding the need to use a huge amount of computing resources to brute force predictions, such as folding at home, or the more meticulous and labor-intensive use of direct observation, both of which take much longer than this. Today, Dr. Kiki, I'm so glad you're here to help us understand a little more of what this advance in neural network training can actually do for us. Can you help us understand a little more about what this advance does? Right. In addition to making it faster and more efficient, eventually it's going to get us to the place where we will be predicting the structures of proteins so that perhaps if we have another pandemic with a new virus, with a different kind of spike protein, we can rapidly synthesize antibodies that would be useful in a vaccine to be able to get something out faster. There are lots of ways that we can use protein prediction to be able to help us medically and in moving forward technologically. Yeah, I think I sort of touched on this yesterday briefly, but I think one of the things that might be hard for someone who's never really thought about protein folding a lot, which I imagine is a lot of folks out there, is why can't you just look at it under a microscope and see how it folds? Do you know? Yeah. This comes down to biochemistry and physical biochemistry, if you want to go back to your graduate level courses. Every molecule has physical interactions that are based on physics. Certain molecules have bonding angles and we know the angles that they bond at and we know the strength of bonds that can be made between different kinds of elements on the periodic table. When you have something like an amino acid chain, which is a chain that you would look at, that chain will end up folding up based on the attraction level between different units in that chain. It's like little tiny magnets that pull it together. How do you predict how that chain is going to fold if it's made up of a number of different random units? You can see the amino acids, but you don't know how they're going to end up in practice the shape they're going to be in. Is that right? Exactly. Yeah. You don't know exactly whether they're going to create what's called a helix shape or whether it's going to create a beta fold sheet where long stretches of the amino acid chain actually fold up over each other like a fan. Being able to predict these structures will help determine the overall structure of a protein because a protein is made up of lots of different amino acid chains. It's called the tertiary structure. You have the primary structure, secondary structure, and then you have all these molecules that come together to create the final protein. That is again another level of complexity because once you have the folding of that original amino acid change, how does that link up with a completely different folded amino acid chain? You have all these levels of how different parts of the molecules attract to each other or repel each other that end up determining that final structure. Then the final structure goes on to interact. Say it's the spike protein for SARS-CoV-2. That interacts with a receptor on a cell. How does that structure create a strong bond with that receptor or a weak bond with the receptor? There are lots of different ways that it can play out and be able to predict it. It has been really difficult. I've kind of thinking about it as like I could tell you like, oh, this car has this much steel in it and this much plastic in it. But the arrangement is the difference between it being a pickup truck or a coupe or a hybrid or something else. That's going to be useful for different things. When it comes to the eventual structure, when you're fighting a disease or when you're trying to make a material that's useful for something, knowing how it's going to interact depends on that protein structure. Have I got that right? The protein structure is really the determining factor of how these different units interact and how they will function. In biology, we talk a lot about structure determining function. This is a really huge example of this. As we move forward in synthetic biology, it's going to also help us synthetically design proteins because if we know that we have a target and we want to create a drug to hit that target, for instance, you know what the target looks like. How can you create a molecule that will really bond with that target? Being able to design an amino acid sequence that will fold up the way that you want it to to create that bond is going to be hugely influential. To bring it back around to the algorithm, I guess what we're getting here with the alpha fold system is the ability for it to be trained to say, try to guess. We know how this set of amino acids fold. Try to guess. We teach it to do that till it's good at saying, I know what to look for. Then we set it loose on things that we don't know how they fold. It does it faster than the old brute-forth method of just trying different folding until it fits the data we have. The folding at home has been that example of the brute-force. I know that's one that's been used as well. The other thing that we have done is the fold-it game, which uses human pattern recognition. Aside from this new advancement in the neural network machine learning, pattern recognition has been the only thing that's really one of the few things that's rapidly pushed us forward in identifying the structure of a lot of proteins. There's brute-force. There's people. We can recognize things really well, but now we're training artificial intelligence to do that recognition for us. It looks like it's going to happen a lot faster than what we've been able to do. Even when we're using our intuition and fold-it, we still don't know, right? We're still just helping the brute-force go a little faster. In this case, there's no brute-forcing happening at all. It's not guessing. It's going, oh, no, I'm pretty sure based on this amount of metal and plastic that you've got Prius, and it gets it right most of the time. This isn't going to change things today, but it is going to speed up the process moving forward. Always in these reports, I'm like, five to 10 years, but I honestly think that with this kind of advancement in five to 10 years, we will be seeing the results from this advancement. Yeah, that's good stuff. Thanks for helping us understand that a little more. That's really helpful. I do hope I helped. You helped me. Yay, good. Well, on November 19th, just a few weeks ago, the US National Science Foundation shut down the Arecibo Observatory in Puerto Rico due to the potential for collapse. Monday night, cable failures caused the instrument platform to fall into the dish below, kind of what everybody was worried about. Two other cables had failed earlier in the year, making it too dangerous to even inspect, much less prepare the remaining system. Arecibo Observatory officially opened quite a while ago, November 1st, 1963, and played a role in multiple discoveries like pulsars, as well as the search for extraterrestrial intelligence. Over the last 15 years, the NSF cut funding. An auxiliary cable snapped in August and a replacement cable had been ordered, but then one of the main cables snapped on November 6th. So this all was, you know, it happened quickly. The cable snapped despite only bearing 60% of the strain required for what it was thought to be its minimum breaking strength. Scientists used Arecibo in 1974 to broadcast a picture detailing humanity's achievements for anybody that might be listening out there. Arecibo also appeared in television and film. You might remember, Goldeneye, Contact, and others. Yeah, a bunch of others. I'm of two minds about this. One thing, Arecibo is so iconic, it's sad to see it fall into disrepair. But on the other hand, how many other buildings from 1963 are still up and doing their purpose, especially something as important and precision oriented as this? It lasted a long time. It did last a long time. And it has been struggling for funding for many years. Like you said, Sarah, the funding has been cut previously and they transferred management of it to the University of Florida for a while. And there are other telescopes that have been coming online that are bigger, that are better. And we do have better technology now. So we could build something bigger, better, faster, stronger. The other angle to it is that this has been really historically and in modern times it's been useful to the Puerto Rican people. This has been a huge instrument for advancing astro-radio technology and the profession within Puerto Rico. And for them to lose this tool is going to really impact their ability to train future scientists in their country. It's not just going to affect our ability to be able to see near-Earth objects that might come and hit us someday. It's going to also affect people here on the planet. Yeah, yeah. I mean, we have other ways to see near-Earth objects. It's not like Arecibo was really even the main way and certainly not the only way. And I guess the good news is that they are going to continue to operate the scientific outreach from Arecibo, even if they don't have the ability to operate the telescope anymore. So hopefully that fills at least part of that gap. Part of it, yeah. But they also lost an additional building where some of their work, that outreach work, is done when the structure collapsed today. So it's going to be hard to replace it. Maybe it'll get replaced, rebuilt there in Puerto Rico. Maybe there will be structures built elsewhere around the world that will be NSF funded. We have yet to see the priorities as we move forward. Well, one of your priorities may be acquiring some moon rocks. And we have good news, if that's you, because China's Chang'e Five Lunar Lander successfully touched down near the Mons Rumkur Volcanic Complex in the Oceanic Prela Sarum. The Lander will spend the next couple of days gathering up to two kilograms of regolith to bring back to the orbiter for a return trip to Earth. It has been 44 years since the Soviet Luna-24 mission last brought back moon rocks. Yes, those samples and others brought back to the Soviet Union were brought back to the Soviet Union. And US Apollo missions were from an era where the rocks were more than three billion years old. Three billion. Now the rocks collected by China are from an era no more than 1.3 billion years old. So still pretty old, but much less so. It'll provide valuable information for estimating the age of other surfaces in the solar system. It's exciting. Yeah, we're getting some new new new moon rocks in multiple ways, right? They're new in that they were bringing them from the moon, but also they're younger rocks, which will help us figure out like, okay, if we see this many craters, then that associates with this age of rock and we get more precision measurements and stuff on that. I'm just surprised that we haven't seen more of a push in, you know, I mean, we had the space race against the Russians, you know, many years back, decades back. And today, the Chinese are really pushing forward on going to the moon and their technology's advancing, their ability. Now they're bringing stuff back. I mean, why are we not hearing more? We've got moon missions planned, but I feel like it's not as much of a race. It's not so... That's because everybody's getting along better now, Kiki. It's more collaborative. Right. We're all about, you know, having people pay a lot of money to go up into space for like 10 minutes. That's our thing. I'm all about collaboration. If we can work globally... We have a huge NASA... NASA has a huge plan to put the first woman and the next man on the moon and all that. So it's happening. That's out there. I want NASA to be friendly. Can we just be friendly and have lots of friends and then make alien friends? I like it. I do like the idea of making some alien friends. In the meantime, we'll rely on humans to be our friends, like people who email us. Indeed. Mattia wrote in about our conversation yesterday about Microsoft saying, hey, we want to make meetings better. You might be remote now, but you might be in a conference room later, and we can gather a lot of data and make everybody happier. Mattia says there's an old adage in the business world that basically says managers will create metrics based on the data that's easy to obtain and not based on the behavior they wish to encourage. It would be nice if Microsoft makes sure that the productivity data is usable and not just a way to sell more licenses. That being said, the ultimate responsibility is on an individual company. And sadly, it's easier to create metrics than spend the time determining if those metrics drive positive and productive behavior. Great example of this is measuring lines of code. This inadvertently creates bloated code, encouraging 10 lines of code when maybe two may have been sufficient. Another example is measuring defects found in developer's code. This encourages developers to avoid refactoring any complicated code that they may find. It also creates a contentious, not collaborative environment between testers and the development team. The approach of creating metrics based on what you can measure and not based on the behavior that you want to encourage also leads to long drive through lines and other real world situations. Well, the good news is Microsoft is trying to lean into that concept. Corporate Vice President Jared Spetaro wrote this morning that Microsoft will change its productivity score data to only aggregate data at the organization level. No one will be able to access data about an individual user going forward. It was never designed for that kind of surveillance, but as we mentioned yesterday, researchers did discover how it could be used to look at individual activities. So Microsoft's going to update its privacy disclosures and the user interface to make clear the product is designed for assessing organizational, not individual productivity. And again, make it, maybe not absolutely impossible, but make it very difficult for someone to find individual data. They're going to, before it was really easy. So they're going to, they're going to stop that. Well, thanks, Mattia, for the very thoughtful email. Good stuff. And thanks everybody who emails us with feedback. If you have some, please do the same. Feedback at DailyTechNewShow.com. Shout out to patrons at our master and grandmaster levels. Today they include John Atwood, Daniel Dorado, and Craig Meyer. Also, big, big thanks to Kiki Sanford for being with us today, host of This Week in Science, and all around very smart person. Kiki, where can people keep up with your work? Thank you. Sometimes I don't think I'm that smart, but people can find me at thisweekinscience.com where you can find This Week in Science. You can find us on Facebook, YouTube, and we also are streaming on Twitch now. You can also find me on Twitter at Dr. Kiki, d-r-k-i-k-i. No, Kiki, you are a smart virus that infects my brain and makes me smarter. That's how I think of it when I listen to This Week in Science. I love going viral. Hey, folks, we want to send you a holiday card. If you're a patron and you have given us your address by December 10th, we will, but we have to have an address to be able to send it. If you don't want to give us your address, that's cool. We just can't send you a card, but hope that's cool. You can check if we've got your address by going to patreon.com slash pledges, find DTNS if you back more than one Patreon, and look in the right-hand column. If you see your address and it's correct, then you're good. If it's not correct, edit it. If you don't see an address, click Update and give us the address. If you want that holiday card, become a patron now if you're not already or update your address dailytechnewshow.com slash Patreon. We are live Monday through Friday for 30 p.m. Eastern. That's 2130 UTC. Find out more at dailytechnewshow.com slash live. Thanks to everybody who joins us live and we'll be back tomorrow with Scott Johnson. Talk to you then. This show is part of the Frogpants Network. Get more at frogpants.com.