 This tenth year of Daily Tech News show is made possible by its listeners. Thanks to all of you, including Paul Boyer, Brad, Kevin, and our new patrons, Tyson and Dylan. Welcome into the club! On this episode of DTNS, Microsoft gets ever closer to acquiring Activision Blizzard, but can they pull it off? Plus, people are starting to complain that chat GPT isn't as good as it used to be. Well, that was fast. This is the Daily Tech News for Monday, July 17th, 2023 in Los Angeles. I'm Tom Merritt. From lovely Cleveland, Ohio, I'm Rich Strafilino. And I'm the show's producer, Roger Chang. Oh, folks, we're going to keep you up to date on all the best tech news that will help you understand the world, starting with the quick hits. Tesla announced its first production Cybertruck model has come off the assembly line at its Texas factory. Tesla expects to start mass production by the end of 2023. In completely related news, Ford announced price cuts for its F-150 Lightning electric truck, the cheapest pro trim down almost $10,000. It now starts at $49,995. And the cuts extend up to its highest platinum tier, which is down around $6,000 to $91,995. This comes as car dealers report three times the number of EVs on their lots compared to last year, according to a report from Cox Automotive. On July 4th, the Court of Justice of the European Union ruled that meta unlawfully collected data for personalized ads without consent. So based on that ruling from earlier this month, Norway's data protection authority ordered meta to temporarily pause displaying ads personalized with online activity or location to Norwegian users. It's the first data protection authority in the EU to do so. This pause will last for three months as of August 4th. Meta can still personalize ads, but it can only be based on the about profile section. The ban would be lifted earlier if meta can demonstrate a way to legally process personal data and give users an option to opt out of targeted advertising. The agency also requested a binding decision from the European Data Protection Board on final measures regarding this ruling. They're begging for somebody to make some rules based on this ruling on here. Just give us rules, man. Talk and they will hear you. Twitter Chairman Elon Musk posted on Saturday that was July 15th. We're still negative cash flow due to a approximately 50% drop in advertising revenue plus heavy debt load. Since Twitter is a private company, it largely does not have to disclose its financials. As machine models gain more acceptance, independent evaluations of them will be critical. Industry consortium ML Commons just announced something called Medperf to benchmark medical AI for buyers. We talked about these kind of systems kind of starting to be tested on the show before and now we're getting some benchmarks for them. At launch this will be limited to mostly models that help evaluate radiology scans. A similar announcement comes from the nonprofit Common Sense which publishes ratings on privacy and suitability of media for children. Common Sense is adding AI products to its ratings and review system so you can see if Claude or ChatGPT are suitable for your kids and you can find out more at commonsensemedia.org. Xbox Live Gold Members will be converted to a new subscription called Xbox Game Pass Core. That will be the least expensive of the Game Pass options and that will start on September 14th. Cost is staying the same so if you're paying $9.99 for example in the U.S. you'll still pay $9.99. It just has a new name. And Games with Gold which is the part of the service of Xbox Live Gold that gave you games for download at no additional cost will be replaced by a small catalog of around 25 games that will be available as part of your Game Pass Core subscription update that list every couple of months or so. Online console multiplayer will remain as part of this new Xbox Game Pass Core which is a little odd because it's not available on the more expensive Game Pass console plan only on the Game Pass Ultimate plan. But that's not the weirdest thing that happened to Microsoft today. Well it didn't happen. They did it. It's not the weirdest thing Microsoft did today. Well this happened to them. Anyway Microsoft deal to acquire Activision Blizzard made its biggest step forward over the weekend. Sony agreed to a deal to keep Call of Duty available for PlayStation for 10 years after any acquisition of Activision Blizzard by Microsoft happens. Rich, stand and applaud. They finally did it. So I have to put on my white flag. Hold on. Yay, yay. This is similar to deals Microsoft has signed with other platforms. They signed an almost word for word version with Nintendo just about Call of Duty. They've also signed more wide ranging deals with other companies like Nvidia. There were some leaked emails that came out. Not like I guess they weren't leaked. They were part of a court case that indicated that Sony was considering a wider ranging deal. But this is not that. This is just about Call of Duty. And the agreement I think also signals that Sony has finally stopped actively trying to block this deal. If they're willing to sign an agreement with Call of Duty because that was their biggest stick in trying to get courts to block the deal from happening. Now that is not the last hurdle for Microsoft. They also still need approval from the UK. The two sides have passed or have paused legal actions and are in talks to determine a restructured deal that would address the competition and markets authorities concerns. Scuttlebud is it has something to do with selling off some cloud rights so they don't have a concentration of cloud rights, at least in the UK. The CMA has extended its deadline for its overall investigation from July 18 to August 29. And they have gotten the approval of the competition appeal tribunal which hears appeals on CMA decisions to do that. However, Microsoft has a deadline of July 18. That's tomorrow as we are recording the show in order to close the deal without having to pay a breakup fee. There are some ways that if both sides think the deal can close if they just have a few more days they can avoid having to pay that breakup fee to make the deal happen. But investors really don't like to do that because it weakens the strength of the breakup fees and they want to keep to those deadlines if they can. Rich, lay your odds on this closing on Tuesday. I'm going to give it four to one. Now hear me out. I think it's very possible that, as you were saying, Tom, that Microsoft has some legal and business machinations in their quivers that they can get this to avoid that breakup fee. What is remarkable about this entire process is essentially Microsoft just seeming like they're out maneuvering everyone out here being so ahead of all of the competition concerns outside of the UK. And even then it sounds like they're smartly like willing to concede on some of the cloud issues that the CMA kind of brought up here. But looking at how they dealt with Sony, looking at throwing call of duty at Nintendo, which is something that literally no one asked for, and coming out with these 10-year deals, understanding the optics, the morass of legality that they were getting into, I have no doubt that Microsoft can figure out how to avoid that breakup fee because investors like them some free money. But I don't see it like, if we're going to say this is the date in the Wikipedia entry of when this closes, it will not be Tuesday officially. We'll have like a little weird note and it says citation needed or something. Yeah, I agree with you. I think there's a 97% chance this closes at this point. I know the FTC could still, let's say it closes before the FTC case at the beginning of August, I think it's August 2nd. The FTC could still pursue legislation or pursue the court case to try to unwind it, but that's a tall order, especially given how strongly the decision against the preliminary injunction was crafted. So the FTC might go ahead with it. I don't think they win. It looks like the CMA really does want to come to an agreement, but it's going to take longer than the 18th. They are not going to get a decision from the Appeal Tribunal today and have it done tomorrow or they would have had it done a long time ago. There's still a lot of back and forth to be done. I don't know how long they need. I know they've been given until the 29th of August and I guess the CMA might want to play this out as long as possible, but Microsoft's going to want this to be done before the 2nd when the US FTC goes. So they're motivated to come to an agreement before then. I think by the end of July, this is probably done. Yeah, like in the books done, completed. What's interesting here is it seems like now we have reached the point where most parties have been like, okay, as structured, this deal can go through, but there are these variety of procedural hurdles. The CMA, officially because these remedies were not part of their final part, has to open a new investigation if Microsoft is making new concessions before it could consider that kind of stuff. It's like on its face, okay, we want to accept these, but because this is the way these types of investigations are structured, and for good reason, we have to jump through a little bit of these hoops again, which understandably takes time. I guess they're rubber stamps. It's the damn paperwork. It's always the paperwork. The CMA DMV is a little bit more. No, all joking aside, it really is probably going to be the paperwork, because I think the CMA wants to get a concession, declare victory. We got them to bow. Microsoft wants it done. They want it done before the end of the month. And yeah, I think there was some talk that Microsoft might want to try to carve out Activision Blizzard, because if they were to just go ahead and close it without the CMA's approval, they'd risk paying a fine to the UK for breaking regulatory rules. So I think they were thinking, well, what if we carve out Activision Blizzard in the UK? That would have been really messy. It would have kicked the can down the road. It would have upset the CMA, making them less likely they would want to resolve this. I don't think that's on the board anymore. It really does look like they're going to figure this out. Well, typos aren't just ducking annoying. They can also leak data. If you put a typo in something like an email address, you might potentially send embarrassing emails to a stranger. However, a new report from the Financial Times found that a common domain typo has implications for the US military. That's because the .MIL domain used by the US military often gets typed in as .ML, the country code domain for the West African country, Mali. This isn't theoretical either. Speaking to the Financial Times, a Dutch entrepreneur that's managing the domain, Johannes Zubier, set up a system to catch misdirected military emails to .ML addresses. He was looking for keywords like Navy, Army, that kind of stuff. Since January, this captured more than 117,000 emails since the start of the year. The emails include things like sensitive medical records, identity documents, military base photos. Military doesn't like that. Military itineraries like room numbers where high ranking officials are staying and that kind of stuff. Tom, this sounds bad. Is it bad? It could be bad. The important part here is that the contract that Zubier had to operate the domain for the country of Mali and today, July 17th. And so now Mali is in charge of its own domain again. Mali is very cozy with the Wagner Group, the Russian paramilitary organization. It seems to be fairly cozy with the Russian regime and with the United States engaged in backing Ukraine in the war in Ukraine that Russia is fighting. It seems like a bad idea for military emails to be going to Mali where they could be passed along to Russia. So that's one of the reasons this is a more sensational story than usual. The thing to keep in mind is that this isn't the military accidentally sending things. They actually have filters within the DoD to check to make sure .ml has not been typed accidentally and block emails from going to .ml. It's a blocked, you can't send email to Mali without special approval if you're working within the DoD. So it's not going to be people within the DoD sending emails to each other. That would be horrible. It is still a possibility and probably the majority of what's being caught by Zubier up until now is third parties. There's lots of military contractors out there. And if one of them doesn't have a similar filter in place, they might accidentally be sending information. And that's where you get them adding health records and stuff. I imagine this kind of intelligence is much lower level and much less useful than if it was coming directly from within the military. It's probably not impossible to get elsewhere. But it's not great. Not great, Bob. I'm not going to lie. Yeah. For a lot of more high-end correspondence, there are dedicated systems outside of email that are much more trusted. Those are not as open-ended. So it's harder to enroll a vendor or something like that in there. But what's really interesting about this, this shows a problem with enterprise security in general in that what the DoD did in blocking .ml emails and stuff like that is they were addressing inside a risk. Someone inside their organization, inside their four walls metaphorically, messing up and sending something the wrong way. And that is a whole branch of risk in any kind of large organization that is becoming more well understood. We're looking at behavior monitoring and stuff like that so we can see whether someone's acting maliciously or accidentally in either way kind of stopping that. But what represents what email is coming in is kind of third-party vendor or just third-party risk. And email is just this wide open system where, I mean, if we look at wider security stories, we've seen like the move it, manage file transfer, you know, exploit on there. Having all sorts of implications for giant companies like Shell was the latest one I saw that was implicated in a database with that. But we've seen all sorts of giant companies being vulnerable to that. Their security posture was fine. The DoD is doing what it needs to do from this. But they are only as secure as, you know, they're outside vendors that are contacting them. Obviously, they can do things. If this is a contractual relationship, they can say, you must have these, you know. Yeah, put these filters in place. Yeah, stuff like that. Yeah. And it should be noted, Amos, our producer who was in the military, says those filters are notoriously hard to get around. And he said, it's a fun meta-game we play. To try to figure out if they could get around the filters. Yeah, some fun background from them on the ground experience. Yeah, so it sounds like they actually work when they're properly implemented. You just need your third-party contractors to implement them. And yes, I'm sure some of you are like, also train your workers not to make typos. Yes, do that. That will only have limited effectiveness. People are going to make mistakes. I have all kinds of things come to Tom Merritt email addresses. Usually regarding training camps for football. I have been on several college football training camp invitation lists for many years. I have no idea why, but they're always to Tom Merritt. So some Tom Merritt did not get invited to training camp. Probably could have gone pro if they'd put in the writing email address, but for a typo. Well, folks, whatever you're typing on, we hope you enjoy it. If you're typing on an Android phone, we got something special for you. Android Faithful. Ron Richards and Huan Tui Dao bring you a podcast devoted exclusively to Android news and information launched right here in association with Daily Tech News Show. If you've been missing all about Android, you can still get Ron and Huan and all. Michelle's been on the first couple episodes as well. Episode two happening Tuesday, 8 p.m. East Coast, 5 p.m. Pacific right here at twitch.tv slash good day internet or available in your podcatcher of choice. Subscribe to it right now at www.androidfaithful.com. ChatGPT remains one of the poster children for large language models. If you talk about AI with somebody on the street, they probably heard about ChatGPT, but that has not stopped it from seeing a decline in usage. Recent data from similar web saw a 10% drop in traffic compared to last month. We talked about that briefly last week on the show. Sensor Tower reported a decline in app downloads as well. A recent piece from TechRadar's Christian Goitin cites several users claiming the ChatGPT quality decreased since switching to OpenAI's GPT-4 model. So it's not just the interest has gotten saturated, but maybe some people are just not finding it as useful. The new version of GPT, GPT-4, apparently prioritizes speed, but a lot of people think that maybe that speed is coming at the expense of quality. There's some other theories that perhaps OpenAI has been breaking up the machine learning model into smaller models trained in specific areas, and that way they can save some computer overhead by having smaller models because they take up less computer power, but maybe that doesn't work as well as just having the full model running. OpenAI has not confirmed whether that's the case or not. Rich, are you buying that decreased quality is the reason we might be seeing a decline in usage of these kind of tools? Maybe incrementally, but it seems like it's way more that this isn't the new shiny anymore. When you first got access, first of all, you had to get access to chat, GPT, so you're excited that you could even use it the first time you got to try it out, and then all of a sudden, it could answer all these prompts, it could do all this stuff. You know what it sounds exactly like? The first time you got the first phone with Siri, or you got your first Google Assistant, you got your first Amazon Voice Assistant. First time you talked to Eliza on a Commodore 64, for goodness sake. All of the magic with interacting with machines, and that honeymoon phase, it is sweet, it is intense, it is intoxicating, and then you start, you know, it's like, this is amazing, I can do this at all. Okay, it's not perfect, but you know, this is amazing. And then as chat GPT especially is starting to really work its way, not just into being like a tech curiosity, a bubble, but trying to integrate that more into our daily lives. I'm sure there are, you know, that's where you start seeing the gaps in its capability, what it can and cannot do. Now maybe the model does play into that. I don't want to dismiss those, you know, those concerns or something like that. And as OpenAI is moving to monetizing its APIs and that kind of stuff, speed is certainly paramount to that. But I think it just mostly has to do with we are living with this now. It is technology in the dirt. It is not the new shiny. I agree, and I do not, this is nothing but my own opinion. I do not have any data to back it up. So I have some good friends who work at OpenAI, and I'm sure they could poke holes in this if they were able to. But my guess is it isn't the chat GPT got worse or the GPT4 is worse. It's that people are now reaching the edge of what it's capable of. And as you so rightly explained just now, Rich, when you first try something, you see the amazing things it can do. Any new thing, any new piece of technology, you're always blown away by what it can do. Then you get used to that. You start taking that for granted. Well, of course I can pick up this small black rectangle and call anyone on planet Earth. That's not impressive. Any black rectangle can do that. And I think that may be happening here with these models as well where we're used to it being able to answer in complete sentences. We're used to it being able to understand the context of something we've been saying and have an ongoing conversation. Those aren't impressive anymore. And now we're trying to make it do things that it can't do. And it never could do because it's not perfect and it's not as all-powerful and dangerous as a lot of people may have feared or believed. So yeah, we're just seeing the edges of what it's capable of. That's my guess anyway. And I would say, especially for the web app, the idea is you might have found the way this fits into your workflow really well. At first you might have been trying, all right, I'm going to see if it can draft every email that I can do. And then I'm going to see if it can write up this paper that I need to write up. And you've probably found, if you're still using it, the subset of all of those things that it does well and then what it doesn't. So naturally you're just, you might be using it more focused, I would say, as opposed to the exploratory phase of all of this. And also there are more chat GPT or GPT-4, 3.5. These models are out there in all these different apps too. So you might have, there's just more of this stuff out here. So it's not that you're using LLMs or something like that less potentially. It's just that you have more options, even if a lot of them are from open AI themselves too. Yeah, still Dave says we might just be getting beyond the what is this thing curiosity and people who are probably never going to use it regularly have stopped. I think that's part of it too, well said. Well, even if chat GPT isn't seeing a decrease in quality right now, so-called AI cannibalism might lead to diminished performance in the future. This could come as the result of more and more tax from large language models being included in the training sets of future training models. Excluding this content is actually really hard right now because LLM generated text detectors aren't reliable. Looking at the tool GPT-0, ours Technica's Benj Edwards found it looks for degrees of perplexity and burstiness. Essentially these models look at the degree to which the next word or in a phrase is likely to be typical in its training set, like how unique is your completion of a given sentence or something like that and how variable your word and sentence length is. If it's all uniform or you're going run-on sentence and then staccato bang bang, the more perplexity and burstiness theoretically the more human, but it turns out people can be boring and AI is getting more complex. Yeah, the big one in this article is the US Constitution does not pass the test. It appears to be AI generated and they think that's because the Constitution is in the data sets thousands of times. And so they're saying, oh, this looks like the kind of thing that's in our data set. So it's probably coming out of our data set. And it's like, well, yeah, probably is a new data set. But this is going to be an arms race. There's always going to be a new thing that you can find that is emblematic of these AIs and then something's going to change. Maybe it's actual adversarial or maybe it's just the way people are using it. Maybe it's AI cannibalism where the training set starts to get polluted with things from AIs that are publishing on the web. And yeah, it's just never going to be perfect at detecting this. My guess is this will slowly get better as we learn these lessons though, don't you think? I think so. And I think we'll start getting away from the... I know a lot of these tools like GPT-0, like hedges, everything. It says it's likely or it's not likely or something like that. But I do think as we come more familiar with these technologies, I also think we'll become more comfortable with indicators of confidence when it comes to something like this, just like we've been talking about. Yes, yes. With these LLMs. Gradients. Yeah, for the gradients of how quote unquote accurate is the statement. We might see that as a feature in LLMs down the road. I also think these detectors really need to build in that nuance as well. So it doesn't become, especially in academic use, where there's academic integrity concerns as well. Yeah. Also, it strikes me that perplexity and burstiness really describe me in junior high. All right. One last thing to mention here today, Corsair. You probably think of them as the memory makers if you're old like me, or maybe you know them for what they've become more well known for recently, gaming hardware. They own Elgato. So maybe you think of them as the folks who own the folks who make the stream deck, but they just acquired Drop, an online retailer specializing in mechanical keyboards and keyboard customizations. If you are in the clicky keyboard customization world, this is rocking that world. Corsair plans to keep Drop operating as a separate brand. So don't get too frightened. But they do say they want to do some specialized Corsair and Elgato products through Drop right alongside their Marvel and Lord of the Rings items. Yeah, I kind of see this as they're extending all of their IP partnerships, right, so that they can just add these to Drops. But yeah, this is really, you know, the hotness for, hey, we have a limited dish. And we're going to have 50 of these keycap sets, you know, and, you know, they fit the certain profile and they're going to look great on, you know, all those hotswappable MX Cherry Brown switches or whatever the new hotness is. I can't even keep up on all the different types of switches, but this is kind of the default place you start looking to customize your mechanical keyboard journey, then you get into your subreddits and then you get into all the different community stuff where you really can go wild. You're printing your own. You're, you know, you're putting together your own keyboard and stuff like that. But Drop is kind of that gateway into this weird world of mechanical keyboard. So it's a heavy move by Corsair. You know, they, I was, I've been surprised how well they've expanded into kind of gaming peripherals and Elgato and stuff like that. And this makes total sense more into, I guess, lifestyle stuff too. Yeah, and it's a deep rabbit hole. I mean, I, I have an IBM clicking keyboard from my old IBM PS to 32 86. I feel like you know more about mechanical keyboards than I do because you've, you've dabbled, but I know you're, you're just on the edge of the deep well of, of, of this. Yeah, I like cold, but yeah, I can't even get into all the types of custom switches that are out there. Like that's, that's where you get into it, where you're like, it has like only 55 grams of activation force. Yeah, yeah, yeah. You know, yeah, that's. Keyboard fans. What do you think of, of Corsair get-and-drop? Let us know. Feedbackanddailytechnewshow.com. Let's check out that mailbag. Well, Mike and Austin rode in on a blockchain solution partially made possible by DTNS, or at least inspired by. He says, hi, cool with a K. I mean, little Tommy Tinkertown on episode 45 59. Tinker. Tinker. It's very important you pronounce that correctly. Oh man. This is what happens when I try to read alliteration fast. So please keep sending those in. I'm episode 45 59. You were responding to an email from Lauren from Montreal who wanted to know if there's a blockchain solution for tracking who contributed to the creation of digital media. I'm building videates that exactly for video content. Not only can you see everyone who helped in making the videos you watch like on chain versions of the credits of a movie. You also can contribute. But you can contribute to them monthly similar to Patreon, except instead of your money going to a single owner of the video, it automatically gets split appropriately between all the people who helped create each video you watched that month. And he says three years ago in episode 38 14, you explained how the Swiss nonprofit DFINITY was building a decentralized serverless app architecture. And after watching that episode doing a little research of his own, I chose their product, the internet computer to build this on. Look at this. Hey, that's fantastic. I'm so glad to hear that, Mike. And thanks for writing in and let us know about this. Mike also offered to let us do some stuff on the platform. So I will get back to you about that, Mike, and don't let me forget. I'd like to try it out. Sounds fantastic. Thank you so much for sending us that email. Tommy Tinker Town. Tommy Tinker. Tinker with an R. No L. That's an R. So thank you so much for sending us that email. I'm sure you can stick around for the extended show. Good day. Internet. We're going to be talking about. Nostalgic. Glee. We're going to be talking about a search engine called wibby. We're going to be talking about the fact that everyone loves the nineties again. Apparently so much that they only want to see web 1.0 style sites. Stick around. And patrons, you can catch the show live Monday through Friday. And you can find out more through the World Wide Web at DailyTechNewShow.com. It is a website. We'll be back tomorrow with and talk, we'll be back tomorrow talking about replacing human subjects with chatbots and scientific studies with Dr. Nicky Atkins. This show is part of the Frog Pants Network. Get more at frogpants.com. The Diamond Club hopes you have enjoyed this program.