Tech News Weekly 355 Transcript

Please be advised this transcript is AI-generated and may not be word for word. Time codes refer to the approximate times in the ad-supported version of the show.

0:00:00 - Mikah Sargent
Coming up on Tech News Weekly. Emily Dreibelbis of PCMag is here and we kick off the show by talking about Social Network X releasing its first transparency report under Elon Musk. We talk about the report's numbers and the implications therein, as well as giving Emily a chance to discuss how Elon Musk's feelings, so to speak, about journalists have impacted the choice to report on the company. Then Scott Stein, editor-at-large for CNET, stops by to tell us about his time wearing Meta's new Orion AR glasses Super cool chance to talk to Scott Stein while he was actually at Meta's campus and hear about those wacky AR glasses. Lastly, but certainly not leastly, we can talk about how Microsoft has a new tool that aims to correct AI hallucinations, even though hallucinations are kind of part of what AI is doing. You'll learn more about it when we get around to that story. Stay tuned for this episode of Tech News Weekly.

0:01:18 - VO
Podcasts you love From people you trust this is Twit. Trust this is TWiT.

0:01:26 - Mikah Sargent
This is Tech News Weekly, episode 355, with Emily Dreibelbis and me, Mikah Sargent, Recorded Thursday, September 26th 2024. What it's like to wear Meta's Orion glasses. Hello and welcome to Tech News Weekly, the show where every week we talk to and about the people making and breaking that tech news. That's the sound of tech news breaking. I don't know why it's so subtle. I am Mikah Sargent and I am joined across the internet by the fabulous, the wonderful, the whipsmart, Emily Dreibelbis of PCMag. Welcome back, Emily.

0:02:04 - Emily Dreibelbis
Hi, Mikah, thank you for having me.

0:02:06 - Mikah Sargent
Thank you for being here today. So for folks who are tuning in for the first time welcome, glad you're here. This is a show where, as I said at the start, we talk about tech news, but we kick things off every week by sharing a couple of stories of the week. Both my guest and I bring a story to the table that we think is interesting, and I am looking forward to chatting about your pick this week, Emily.

0:02:34 - Emily Dreibelbis
All right. So my story of the week is about X, or Twitter, releasing its first transparency report under Musk, Elon Musk. He bought Twitter in October 2022. Last year was the first full year of his ownership. He did not release the report. Now, 2024, we get a report and it goes into how the company is moderating content, which is really interesting because Musk has made lack of content moderation a cornerstone of his leadership style and his public persona. So now we're getting a look at what the company is doing, so it's really fascinating.

0:03:09 - Mikah Sargent
Yeah, absolutely, because up to this point, I do feel like up to the point at which Elon took over, we started to see, I guess we started to see, I guess, a change in Twitter at the time, in the sense that they really did respond to, or started to respond to what people were saying about harassment and safety concerns on the platform and started to kind of iterate in public and be more transparent as a company.

And it was kind of refreshing to see, I think, that it started to build trust back into the company for the people who are using it. Because, again, this is one of these products, this type of product that requires people actually being part of it and choosing to put themselves there and does not exist without that. And so people, you know, at one point saying I don't think I want to be here because bad things happen anytime I post something To a time where you know there was more clarity about what was right, what was wrong and that we could kind of see things play out. Then Elon Musk took over the company and that rumbled, tumbled and fumbled all over the place in so many ways and it was honestly, I think, as we were both kind of talking about a little shocked to see a transparency report at all. How have they had time with the like four people who work there these days?

0:04:45 - Emily Dreibelbis
The four people the ban in Brazil, a lot going on. Yeah, I mean, you're absolutely right. Since he's taken over, he's really made a point to show how they're not moderating content. So a classic example is Donald Trump's account was banned after January 6th. Elon Musk comes in, reactivates it and then he also reactivated you know, white supremacists, white nationalists just kind of going to the extreme to prove how serious he was about this platform becoming a self-regulated kind of free-for-all. But what we see in this report is really surprising that X has actually banned almost five times the amount of accounts as it did a couple of years ago. So I'll read you the numbers on that one. So in 2021, x suspended 1.3 million accounts in the first half of the year, in the first half of this year, they suspended 5.3 million accounts. That's quite a few more.

Quite a few more. The top reason for suspension was actually it's dark. It's like child sexual exploitation stuff. There's also hateful accounts and it's dark stuff. You really would want it to be banned. So it's a huge jump and the number of accounts I don't think has jumped to that volume.

0:06:01 - Mikah Sargent
There's not five times the user ships since he took over Right Right to that volume, like there's not five times the user ships since he took over Right Right. Yeah, something that surprised me about this is when I had, you know, first read the numbers had changed. I immediately thought, okay, this is going to be his whole promise about getting rid of bots on the platform that you know the. The reason that these accounts were banned is simply because they're automated accounts that are spamming people or whatever, and so I was shocked to see that it was a different category, yes, a very dark category, but it was a different category. Is there any self-awareness in the transparency report where they are saying anything, any comments about that difference in numbers? I mean, are they themselves going? Okay, we recognize that we were supposed to be letting more be on the platform, but we're not. We're blocking more than we ever have before, or is it just like here? It is goodbye.

0:07:07 - Emily Dreibelbis
That's a good question. They don't benchmark any numbers in the report. They just say this is the total amount we suspended. They also remove 10.6 million posts. They categorize the user reports for what people wanted to be removed. So they just give the numbers from the first half of this year with no benchmark on previous years. And what we've done and other publications have done is to go through those old reports and find the difference and see how the strategy has differed over the years. So it's really surprising that they just present it as bragging kind of to me, like wow, look at all these millions of accounts we've suspended, these millions of posts we've taken off offline. And it's a totally different tone than what we get from Elon Musk himself and of course I mean he's not CEO, that would be Linda Iaccarino but just the difference between this report and what he makes it seem like is going on, between this report and what he makes it seem like is going on is very stark.

0:08:08 - Mikah Sargent
Yeah, I mean that alone was surprising to me the difference that we see here. Anything else that's really kind of stood out about the transparency report. Another thing that I was kind of caught off guard by is I'm used to transparency reports from tech companies being about how the government has requested data and what data has been handed over. Is this just simply a trust and safety transparency report, or does it also involve that kind of information?

0:08:42 - Emily Dreibelbis
It does involve that kind of information, and the Washington Post went through the requests that X has complied with this year versus previous years and found that X complied with 71% of requests, which is 20% higher than the last time X reported it, in 2021. So they're also complying with more government requests. So they're also complying with more government requests. So they're trying to paint this picture like they are, you know, and so I have some thoughts on what the motivations are here that we could get into if you want to switch to that?

Yeah, please, Okay. So this is maybe a little bit of insider baseball, but if you'll indulge me just like as a member of the press, I've been writing about Elon Musk's companies for years, so I write about AI, EVs, space, social media, that's.

0:09:50 - Mikah Sargent
Tesla.

0:09:51 - Emily Dreibelbis
SpaceX yeah, exactly All Elon all day and I have never received an email back from any of his companies Never.

0:10:02 - Mikah Sargent
Never, of course, I didn't. And they used to respond. They used to right. Yes, there were some that did right.

0:10:08 - Emily Dreibelbis
They did, but that predates me as being a journalist. So it's been really going on for a while. They just don't respond. And when Elon took over Twitter, he also ported over that strategy from Tesla to auto-respond to the press with a poop emoji. He changed it to just we'll respond later. And then last week I emailed about another story. I wrote and it just said busy, check back later. It's empty.

0:10:34 - Mikah Sargent
You do that For people who don't understand why you would keep responding. It's not, or why would you keep reaching out? In that case? It's not necessarily that you are waiting for them finally to respond. It's so that you're doing your due diligence as a journalist. Just in case they decided to claim and again correct me if I'm wrong but going there, they said, oh, we would have responded to the journalist, but they didn't reach out. You get to then say that you tried to reach out to them.

0:11:08 - Emily Dreibelbis
Yeah, I mean it's standard practice. I talk to all the companies I write about all the car companies, all the major tech companies. We email with Google, microsoft, openai. It's just standard practice. So I just do that as a good journalist, to do that due diligence in case they do respond. They don't respond.

And Musk also, like over the summer he boosted or retweeted a post that said you know, you can really ignore mainstream media. So he's made kind of like don't read the news, don't talk to the news a huge aspect of his persona which he repeatedly tweets about. So when X proactively contacted the press about this report several days ago, gave it to people under NDA. This is also an insider of how the press works. You'll get stuff in advance and then you write up the article and set it to publish at a certain time. That is like the most proactive a company can be about getting its news out, which they chose to do only for this report. And I had a friend at another publication who said she got the email and she literally thought it was spam. It is just shocking to receive an email from X, especially with a legitimate piece of news. So, yeah, I don't know what you think. I can pause there because I'm getting into the weeds.

0:12:20 - Mikah Sargent
Well, this is where I actually want to go into things a little bit. So I think what we're going to do is take a quick break. We'll come back with more about this, I'll have some other questions and then we'll see if we even have time for my story of the week. If we don't, that's fine, but if we do, we'll do it quick, because there's a lot here that I still want to talk about that I think people who watch the show will find interesting, because that's part of what Tech News Weekly specifically is about. We are talking to people who are doing the tech news and that is something that you do. So let's take a break and we will come back.

I want to tell you about Shopify, who are bringing you this episode of Tech News Weekly. It's funny. I just bought something from a site that uses Shopify this morning and it was a delightful purchasing experience. So when you think about businesses whose sales are rocketing, think like Allbirds I've got some of their shoes or Untuckit I've got a couple of their shirts. You think about an innovative product, a progressive brand and buttoned-down marketing, but an often-overlooked secret is actually the businesses behind the business making, selling and, for shoppers like me, buying simple For millions of businesses. That business is Shopify. Love that sound. Nobody does sales better than Shopify, home of the number one checkout on the planet, and the not-so-secret secret with ShopPay that boosts conversions up to 50%. That means way less carts going abandoned and way more sales being made. So I got to tell you I love that as well. When I go to a site and I see that it's using Shopify and so I go to the next page and I've got three or four different options for payment right at the top, I can quickly move through with Shop Pay or I can use, in some cases, apple Pay or PayPal, and then it auto-populates all of my information. That money's practically already out of my bank account because it makes it so simple to do so. If you're into growing your business, your commerce platform better be ready to sell wherever your customers are scrolling or strolling on the web, in your store, in their feed and everywhere in between. Businesses that sell more sell on Shopify and, goodness gracious, I love the Shop app from Shopify as well. That's what I use to track packages and, as someone who's buying things, not just having the information of oh, this is suddenly with UPS or FedEx or another carrier, but ahead of time, because it integrates with these different merchant systems. So then I know, okay, they are packaging it right now. Okay it's, you know they've pulled it from this. All of that is fantastic. Upgrade your business and get the same checkout that Allbirds uses. Sign up for your $1 per month trial period at shopify.com/twit All lowercase. Go to shopify.com/twit to upgrade your selling today. shopify.com/twit.

All right, we are back from the break and this is now an ongoing conversation about X, the social network, releasing its first transparency report under I should say, under Musk. It has released them in the past, but this is the first time that it's done so under Elon Musk. So you were kind, of, before we went to the break, talking a little bit about how this has kind of caught journalists off guard. I think that for the first time in a while, you had received a response from X or an email from X, and even one of your colleagues thought it was spam. Do you feel like this speaks to a larger change in the company, or are we working with a situation where it has to do specifically with the kind of content that this is and how it reflects positively on the company. Is there something behind it? I guess is what I'm saying a motive.

0:16:30 - Emily Dreibelbis
I think there is a motive, and the most obvious one is that advertiser revenue has really plummeted and continues to plummet and X is having some financial troubles and the advertisers have said that they don't feel comfortable putting their you know shoe ad or their Shopify ad right next to, like a white nationalist post, like it might appear in a feed that way. So the lack of content moderation has driven advertisers away. So it does seem like now X is trying to change the narrative, trying to turn that ship around and put out materials and say, hey, we actually really are moderating a ton of content. So that's what they're trying to do. It is interesting that strategy and the narrative is so different from Musk but again, he's not CEO, but it does seem like they are trying to dig themselves out of a little bit of a ditch that that persona has gotten them in dig themselves out of a little bit of a ditch that that persona has gotten them in.

0:17:36 - Mikah Sargent
See, and that's interesting to me because I was so. Just the other day I was talking with a friend who was asking about X, formerly known as Twitter, and why people were saying X excuse me, formerly known as Twitter, and for a while AP and other organizations had kind of suggested that that's how you should refer to it, and so I wanted to see if things had been updated. So I went to my handy dandy AP style book account and saw that they now just refer to it as X. And you know what? I'm not going to just say this from memory, I'm actually going to go there and read what it said because I found it interesting and a little shady. I don't think that it was trying to be shady, as the AP Stylebook is not about that, at least not on its actual style book, maybe more on its social media accounts. But it says X is the entry, that's all it says. It says a social network, formerly called Twitter, on which users share text, photos, videos and links with their followers in short messages. It goes on to talk about when Elon purchased it and then, in the third paragraph, it starts out, though its cultural relevance has declined.

X is still used by influential people, including journalists, policymakers and celebrities. It is not necessarily reflective of the general population. It should not be a substitute for traditional interviews and reporting. Right, we know that last part, but I thought that first part, though its cultural relevance has declined. This makes me want, like I want to. I wish that I could talk to the people and get true and clear answers, because it almost makes me think that there are only so many quote unquote edgelords in the world. And once you have spent your time trying to bring in all the edgelords and then you realize that they make up a subset of the population, realize that they make up a subset of the population and also, more importantly, they result in companies not wanting to advertise on your platform, then you're between a rock and a hard place, although it's not that hard of a decision, but it can be. And so you got to kind of go oh right, we actually need to appeal to the general masses, because we've absolutely fumbled the ball here, because once it was such cultural relevance, because we've absolutely fumbled the ball here, because once it was such cultural relevance and we've broken.

This is where I get a little annoyed because I just think about how, right here it says Mikah Sargent and then it says at Mikah Sargent, this used to mean go to Twitter, and that was me. It still does mean that, but it could also mean go to Threads, and that's me. Go to Mastodon and type that in and hope that you find me, go to Blue Sky, and I might be there at this. I'm not there at this. What does this mean anymore? And that's so disappointing to me and I wonder if, internally, they're having that realization that they need to try to reclaim the Twitter of yore. But how do you distance yourself from someone who is more of a walking ego than many other people? I don't know.

0:21:06 - Emily Dreibelbis
Yeah, it's tough. I wonder what time frame the AP is talking about there, because I do think Twitter has missed the boat on courting the younger generations, which are going. I think Instagram or TikTok are more popular and now Musk is kind of narrowing it even more and trying to say, okay, this is maybe a hub for people, People who love Twitter under him are more right-leaning, Maybe there's a big crypto community. These are the communities he's trying to foster. So it's almost becoming more of like a niche site. I do think it still has relevance on both sides of the political spectrum, but it's User ship is down and it's not really growing. Money is down and it's not really growing, and Twitter blew the ability to buy a checkmark, did not turn that ship around. Most users do not pay a monthly $8 a month to use Twitter X. I say X in my reporting. I've adopted it for the record.

0:22:03 - Mikah Sargent
I mean I would exactly with AP saying it. Then I'm like okay, I do what AP says, except for the more than versus over I still say more than. But anyway, sorry, go ahead.

0:22:15 - Emily Dreibelbis
Right, I agree with your rant.

My rant on this as a member of the press is like you want to disparage publications, create this broad stroke mainstream media whatever the heck that means, that broad term to just take down people who work every day to bring unbiased reporting, to do interviews outside of X, to get large swaths of population's opinion on things and represent the wider world.

That's what we do. We try to educate people and he spends every day trying to cut down on those people. But now, when he needs money, when his platform needs the audience that those publications have, now they reach out and they say, hey, write about this because it's going to get us money. So now you need us and you've weakened us by creating this whole against the press thing, by creating this whole against the press thing, and now you just expect us to write about it. So that's why in my article I did provide this other context and why they might be pushing out the report. But it's just a frustrating relationship and if he is going to like throw stones and throw rocks at the press, I don't see why. You know, we shouldn't call out some of these finer nuances of just how odd this report is and what it could mean for the company.

0:23:32 - Mikah Sargent
Wow, yeah, that must be incredibly frustrating, I think, particularly because there is a level of, there is a level of there's still a level of handcuffs involved, because the company does have just enough relevance that, even if you would want to not cover it, it needs to be covered to a certain extent.

0:24:01 - Emily Dreibelbis
Yeah, also the content moderation practices can on X, do need to be closely monitored because they have had real world impacts. Like you know, january 6th I brought up the Trump example. They did ban that account because he was quote unquote inciting violence through those posts. It's in the public interest to report on what is happening on this platform in terms of content moderation. So we do. We always take the higher road. It's just like why can't everyone do that?

0:24:29 - Mikah Sargent
Right, Emily Dreibelbis, I want to thank you so much for taking the time to join me today to get into the weeds a little bit and talk about things from the journalist's perspective regarding the first transparency report under Elon Musk, although sort of under, uh, yakarino at the same time. Um, thank you for being here today. Of course, folks can head to PCMagcom to check out the work that you're doing. You're a prolific writer there. Uh, if they want to follow along with what you're doing, where should they go to do so?

0:25:03 - Emily Dreibelbis
So I'm on primarily X and TikTok. Um, it's the same handle in both places, which is electric, underscore humans. I write a lot about electric vehicles and all things tech, including this stuff, so you can find me there written form on X videos on TikTok.

0:25:20 - Mikah Sargent
Awesome. Thank you for your time today. We appreciate it.

0:25:23 - Emily Dreibelbis
Thank you, we're not going to get to your story. Are we up? Yeah, that's okay, oh man. Well, hopefully you can talk about it later.

0:25:26 - Mikah Sargent
Thank you, we're not going to get to your story, is that, are we?

0:25:27 - Emily Dreibelbis
up. That's yeah, that's okay. Oh man Well hopefully you can talk about it later, next time. Yeah.

0:25:30 - Mikah Sargent
Yeah, next time Exactly, bye.

0:25:32 - Emily Dreibelbis
Emily Take care. Thank you, bye.

0:25:35 - Mikah Sargent
Alrighty, folks, we're going to take another break and we will be back soon with our next interview. I want to tell you about ThreatLocker, who are bringing you this episode of Tech News Weekly. Here's a question Do zero-day exploits and supply chain attacks? Have you never making it to bed? Tossing and turning, stressing out? Yeah, I would not be surprised if that was the case. Worry no more.

You can harden your security with ThreatLocker. Worldwide companies like JetBlue trust ThreatLocker to secure their data and keep their business operations flying high. Just think about using a proactive deny-by-default approach to cybersecurity, blocking every action, process and user unless authorized by your team. See, threatlocker helps you do this and provides a full audit of every action for risk management and compliance. Its 24-7 US-based support team fully supports onboarding and beyond, so you can stop the exploitation of trusted applications within your organization and keep your business secure and protected from ransomware. Organizations across any industry can benefit from ThreatLocker's ring fencing by isolating critical and trusted applications from unintended uses or weaponization and limiting attackers' lateral movement within their network.

We learned about ring fencing talking to the folks at ThreatLocker and it was really cool. They almost kind of were so subtle about it. It almost felt like a bit of an undersell, because we found out how ring fencing had such an impact. It turns out that Threat Locker's ring fencing was able to foil a number of attacks that were not stopped by traditional EDR. The 2020 cyber attack on SolarWinds Orion was foiled by ring fencing. Yeah, super cool to hear.

Oh, and for those of you on Mac, worry not, because ThreatLocker works for Macs as well. So get unprecedented visibility and control of your cybersecurity quickly, easily and cost-effectively. Threatlocker's Zero Trust Endpoint Protection Platform offers a unified approach to protecting users, devices and networks against the exploitation of zero-day vulnerabilities. Get a free 30-day trial and learn more about how ThreatLocker can help mitigate unknown threats and ensure compliance. Visit threatlocker.com. That's threatlocker.com, and we thank ThreatLocker for sponsoring this episode of Tech News Weekly. All right up next, we have editor-at-large of CNET, Scott Stein, who incredibly got an opportunity to wear Meta's Orion AR glasses. Welcome back to the show, Scott.

0:28:23 - Scott Stein
Hey, thanks, good to be here. I'm here at Meta right now.

0:28:25 - Mikah Sargent
Wow, yes, on campus, and it was really cool hearing about your experience. I was hoping that you could start by describing the physical appearance and the kind of overall feel of Meta's Orion AR glasses, given that these are mostly well, they are going to be free developers and seem to be something that the company is using internally right now. Was there still a kind of a premium feel to them, and how do they compare to any other AR devices that you've tried, or even things that come close to AR in terms of size and comfort?

0:29:00 - Scott Stein
Yeah, they definitely had a little more of a developer kit feel, but not in terms of build. I mean, they were super solid and, as they showed me in another model, that was transparent. The tech inside is completely jam-packed in there and you got the sense that they were very dense but they were also light. So, uh, that was making a big deal that they're weighing about 100 grams and they felt uh, normal in their weight. I wouldn't call them normal in glasses, everyday glasses, but certainly a call them normal in glasses, everyday glasses, but certainly a lot less than any mixed reality headset or something by comparison like snap spectacles that they just had a, you know, a full AR pair of developer glasses.

So these rested really comfortable in my face. They were kind of balanced, with thick arms, thicker than my glasses, and I wear thick glasses. You know the pictures actually I thought look kind of flattering on me, but they had a kind of a slight tint to the lenses. So I don't know if that was the design of the lenses. Do auto tint, but they are, um, they were not using that there, but it had just a little bit of a slightly smoky feel at Silicon carbide lenses. That's a new type of build material and they felt uh, they felt comfortable. I kind of forgot about them in terms of how they felt and they had like a matte finish, a lot of cameras studded throughout. All the processing for the apps is done on this wireless puck that's kept on the side.

0:30:21 - Mikah Sargent
I want to let everybody know, as Scott said at the beginning of the call here, he's actually reporting to us live from Metta's campus. So if you happen to hear any other people in the background, that is simply because we got a boots on the ground journalist today. It's like one that's gone out into a storm wearing the raincoat and everything. So we appreciate you taking the time to join us from there. Now you mentioned them being flattering. It's funny that you said that, because that was one of the first things I thought. I thought those look really good on Scott. They seem to fit you quite well and they didn't immediately kind of stand out as oh, these are, you know, chock full of tech, and so I was kind of impressed by that. It just felt like almost a fashion choice that was being made. Now you did mention that they've got a 70-degree field of view. How does that compare in well comparison to other AR devices and did it have an impact on your experience overall?

0:31:19 - Scott Stein
Everybody's been asking me that question and a lot of these numbers like field of view, or Meta talked about pixels per degree for the resolution, as opposed to your standard pixel resolution. That pixels per degree is a common metric in ar, which is. Nobody knows what that really means in terms of their everyday life, but the way I was describing it I kind of look for like how my hands feel when I'm doing this. You know, kind of felt like you can't see what this feels like to me exactly, but it kind of felt like this. I realized I hold my hands. It doesn't communicate, but if you hold your hands up like that to your face a couple of inches apart, it's like it goes close to the edges of my glasses, but not completely to the edges of my glasses, but it was enough that the apps they showed that were pop-up were totally filling my area and were fine and I didn't see any cutoff. You know, the snap spectacles were more like a this thing and um, you know, while you can see my glasses as I look at you that way, I can't see the whole world that way. It felt like a narrow window that was taller. So you know, meta was not showing expansive landscapes of, uh, of holographic things across the room. So, you know, kind of like the early apple watch faces, you weren't as aware of theraphic things across the room. So, you know, kind of like the early Apple Watch faces, you weren't as aware of the borders, of the limit of the field of view.

But I never really thought about it, which was a big achievement because you know I've never seen that in a normal AR headset. That's a big move. So I hope that describes it. The resolution is less than you would get in a vr headset 13 pixels per degree, which is actually pretty low res. Um, it didn't feel that way in their demos because a lot of things they showed were kind of simpler icons to click and retro looking games with big icons, and so you know it wasn't. I watched a video of the jets you know highlights from the week before because they knew who I was but it looked fine and doable. But they're working on higher resolution solutions.

0:33:15 - Mikah Sargent
Got it. Yeah, In the article you do talk about and this was something that stood out to us as we were watching the event they talk about the neural input, and this is a wireless neural wristband input. Tell us how this works? Is there a cord that goes from your wrist all the way up to your brainstem? What are the advantages to it in comparison to other, more traditional input methods?

0:33:43 - Scott Stein
And this is easily the most fascinating thing to me, because for years now I've been following Meta's ambitions in this space. They acquired a company called Control Labs years ago that I never got to try their you know their EMG wristband that uses that neural input tech and that's what they adapted over time to become this wristband. I went to Meta's Reality Research labs in 2022 and, you know, saw other people. I saw mark zuckerberg using an early version, but it's funny, neural you know the name probably brain computer interface conjures a lot of different things, and this is not a brain computer interface, but it's. But it uses um inputs, as they say, it's a electromyography, so it's um skin, electrical skin sensors, and I'm not a neuroscientist or you know, so like I, um, I'm I'm still learning more but it uses electrical skin, um, uh impulses to measure muscle movement and you know, or, or equate, equated elements of muscle movement, and so what that's doing is turning those like little impulses into kind of like the way uh, an accelerometer on your watch or you know, optical heart rate can pull out different reads. This is using a different type of metric to pull out a rich bunch of information that turns things into micro gestures. You know it can tell how you're pinching your fingers moving. You know I did these thumb gestures and you know to Meta it can do a lot more. The possibilities for this could include typing on desk surfaces in the future, you know, possibly even having other train moves, and so I think they're starting with this.

But you know it was very small. The band really felt like a Fitbit like someone else has mentioned a Fitbit strap I think that was now how I'm thinking about it kind of like a ribbed. It had, um, inner electrodes that just felt like little contact points that were just gentle and they made it snug, but as snug as you would wear, like a fitbit band type thing. You didn't want falling off if you snapped it. I had a magnetic clasp and then all I had to do is move my hand around and the gestures were actually pretty familiar if you've been following this space.

Apple vision pro and its little gestures how you rest your hands and you can keep them down below was sort of the trick that was happening here, because they don't have to be in view of the cameras, but you can even put your hand off to the side now, like totally out of view. And you know, I just want to see that emerge, not just for the current headsets and glasses but for all products. So I just kind of kept my eyes and hands on that and thought about that even more than the glasses.

0:36:22 - Mikah Sargent
Absolutely Now. You did get the opportunity you've talked about a few of them to experience several demos with the Orion glasses. Which ones stood out to you the most and what made them stand out?

0:36:34 - Scott Stein
Yeah, you know it was funny. A lot of the demos across the room didn't quite surprise me, because I've seen variations of those over the years with HoloLens and Apple Vision Pro, but I think that what they showed was that you could bring up things sitting in a room and start to forget that you were wearing glasses to do it. When I brought up the menu to scroll with apps with my eyes, it has eye tracking and much like the Apple Vision Pro, you look around a selected app and you tap, but this time all that tapping measurement is just on the wrist and it felt even more casual and so they would have an incoming call where I chatted with somebody. I watched the video, I played the most compelling demos with the games. Actually, there was one where I flew a Starship using my head movements the 3D game and then looked at things and then use my fingers to shoot, which you know that those games are kind of on the Apple Vision Pro right now. But again, the Vision Pro is bigger and it's this big headset and so now I'm doing this with glasses and forgetting that I have glasses on. So that was the cool part.

And then I also played like a two person game. That was Pong. That was the cool part. And then I also played like a two-person game. That was pong. That was the most like star wars holochess thing. We stood across the table and moved our hands to control the paddle and I walked around. I could see this like 3d kind of future pong retro thing across the table. That was a super cool demo. But it needed a qr code to position the table. Right now I I want to see when it can start recognizing the room seamlessly. They didn't show off a lot of that, but that was cool too Because again, clever demo in that I wasn't able to see the limits of the field of view, but at the distance I was standing it looked good and that is also rare. Like snaps, I had to keep positioning things right to see the virtual objects.

0:38:28 - Mikah Sargent
Let's then talk about image quality. How do you feel it compares to other AR options that are out there, which are few and far between, and VR options that you've used? Resolution, even in the demos, seemed pretty low. I thought it was kind of clever that Mark and the team continued to refer to them as holograms, because I think we're used to when we talk about holograms it's that sci-fi sort of flickering soft edges situation and that really kind of settles you into an expectation versus what we see with other companies, where it's like the best resolution you've ever experienced.

0:39:09 - Scott Stein
Well, one thing I didn't comment on in my story that I should really add later but now occurs to me this morning, is that, while the resolution may have been lower, one thing that was very common in earlier AR glasses was a bit of a rainbow haze that would kind of appear, certainly with HoloLens 2 over things, so that your hologram was even kind of uneven in color tone. It looked like you were looking at it through kind of like slightly rainbow streaked glasses that you had to clean. But that was just the way it was. There was none of that here and that was really great. So it was a very clear, very clear view of, yes, a slightly lower resolution to my eyes experience you know.

Again, I didn't think about the lenses. That was the cool part, and it's even cooler this morning is that I didn't think about the lenses. I thought about the, what I was being shown, and sometimes in those previous demos you kept thinking too much about the lenses. So, um, they did show me, uh, an early build of a higher resolution version of orion as well at the end, which indicated they were thinking actively about that and knew it was an issue, one with 26 pixels per degree, so twice the resolution, but still approaching what you would get in a VR headset, and that I watched a Jurassic Park movie movie clip, I think, and, um, it looked fine. But again, when you're looking at ar stuff, it's always semi-transparent too, although these can auto dim, so I wasn't thinking about it quite that way.

You know, compared to word x-reals glasses recently, which are more for display use and they have a thicker design, but they use, um, uh, micro-oled displays and they look really good for watching movies, it's kind of equivalent to the Apple vision pro. You know good enough that I would say, okay, I would watch a film in this. I didn't feel like I would want to watch a film in Orion, but I don't think it's being designed for that. According to the team making it, the I don't think it's being designed for that. According to the team making it, the achievement now is about the wide field of view, and then the next step looks like visual fidelity. So maybe in time we start to also see that.

0:41:16 - Mikah Sargent
Nice. Yeah, the article does mention AI integration. That's a big word, or not a big word, but a buzzword. How does AI enhance the AR experience in devices like Orion, where the main sort of draw seems to be the newness, that is, ar in the first place at this level. What does AI add there?

0:41:44 - Scott Stein
Yeah, that part feels a little more TBD, and I think that's not just the case here, I think it's the case for the whole industry. You know, apple has not introduced Apple Intelligence yet for the Vision Pro. Meta still hasn't really done any AI developments of big significance for the Quest headset. The Meta glasses are where they're pushing some AI developments, the Ray-Bans but here I was a little disappointed that they didn't show something really transformative. What they did show were a little more Ray-Ban-like things. One of them I looked at ingredients on a table and asked for a recipe and this is all scripted demonstration, not an on-the-fly one and when I asked it, no surprise, it gave me the recipe, but it also indicated the objects with little dots to code them. But what I wanted, besides just showing a recipe I could flip through, is I wanted, you know, instructions and arrows, and you know, I wanted something more interactive or like how to play a musical instrument which they brought up, but I didn't see an example of that. I also got to generate some uh, you know, on the fly, uh, you know, emoji or art, you know, generated art requests, which is a common demo now in ai. You know, give me a dinosaur on a cactus, you know, holding a pina colada or whatever, and that worked. But that's no different than anything you can do at this point in a lot of different places. It's just showing it on that display.

Um, I think the advantage to AI in the future should be live capture of everything you're looking at, giving you advice, giving you warnings, giving you heads up context. I think every company is still working on that. Meta at the Dev Conference was talking about an interesting thing that I did not get to try with Ray-Bbans, where there's a live video feed going on with the ray-bans and that I'm wearing right now and it can start giving you ai feedback on the live feed and that's obviously that could drain battery life. But to me, that's the, that's the ar glasses future, that's the tony starkey type of thing. They didn't show that, but I think the idea here was it felt like a moonshot, you know, moon landing. We had a team, a room full of people, showing this could work, but it was like the first Oculus demos, you know it can work. Now I need to see what else it can do and what they can do with it, which are, you know, ai is for meta and everyone is at the top of the list.

0:44:05 - Mikah Sargent
The last thing that I'll ask you, based on your experience using this Orion. This, I think, is arguably the most important question what do you think are the biggest challenges that Meta has to overcome before Orion can actually become something that consumers will be able to purchase and use?

0:44:27 - Scott Stein
Meta talked about some of their goals for it, which had to do with price, because apparently for other reports I didn't ask the question, but the glasses cost about $10,000 to make. So you know they are not something that will show up even for developers right now, I don't think, much less consumers, and it's going to stay like as an internal Met meta product to tasks. And they also need to be smaller still, even though they look cool to me at times, and they need to have higher resolution displays. And also, here's the big thing too software AI apps reasons to use it. And I think the other thing that was kind of AI apps reasons to use it. And I think the other thing that was kind of lurking in the corner with this to me is that Meta never made a phone.

Meta has always been moving around the idea of phones. Ar glasses, ray-bans work with phones now and I see a lot of companies Xreal, snap, meta almost seem like they're tackling elements of the same problem. How do you create a set of AR glasses that can do things on the go but don't have to interface with phones that are locking them out of key features and compatibility? Right now, apple is not opening up access for AR glasses in those ways and Google's not on Android. Maybe Google will, but this is almost the next step to me. I still think that you need to have these things work with phones, much like smartwatches, but we have to get to a new level of compatibility and ease. We don't think about watches and phones having issues working anymore, but with glasses, lots of question marks.

So I think you know that compute pocket. Sure you can carry it around. It's not even just the size and weirdness of having that, but really that should be the phone, in my opinion. So I think that you know years to come, when they do make these. They talk about a years long roadmap. Maybe we will be there with phones, you know, or maybe they'll have a little thing you stick in your pocket. So it feels vestigial now. It feels like a vestigial tale. Yeah.

0:46:37 - Mikah Sargent
Yeah Well, Scott, I want to thank you so very much for taking the time to you know head to campus and join us here to talk about this. It's so cool that you had the opportunity to check it out and that you're willing to come on the show and talk about your time with it. Of course, folks can head to cnetcom to keep up with what you're doing. Is there anywhere else? They should go to follow along and, you know, be ahead of the curve to make sure that they see the stuff you're putting out there.

0:47:12 - Scott Stein
It's so hard. Now I always get that question. I don't even know anymore. I mean, I've kind of left Twitter, although sometimes I'll post on there. I'm also on threads, definitely. You know which cross posts, the Mastodon, you know, and I think those are probably the best ways to keep up with me at the moment. So, yeah, Awesome.

0:47:31 - Mikah Sargent
Thank you very much. We appreciate it and we'll see you again soon.

0:47:35 - Scott Stein
Okay, yeah, thanks.

0:47:37 - Mikah Sargent
All righty folks Coming up. I have a quick little story of the week I want to tell you about, but we are going to take a quick break so I can remind you all to join Club TWiT at twit.tv/clubtwit for just $7 a month. Oh and, by the way, if you are watching this live with video, there's a little QR code in the top corner of the screen I'm not going to say right or left, the top corner of the screen that you can scan with your device to head right to Club TWiT twit.tv/clubtwit Seven bucks a month, and when you join the club you get access to some pretty awesome things. You will be able to gain access to our shows completely ad-free. It's just the content. You also gain access to the Twit+ bonus feed that has extra content you won't find anywhere else Behind the scenes, before the show, after the show, special Club TWiT events get published there. Discord server, which is a fun place to go to chat with your fellow Club TWiT members and also those of us here at Twit, and access to the video versions of our Club TWiT shows. So if all of that sounds good to you and I'm sure it does you should absolutely head to twit.tv/clubtwit. It's so cool seeing folks joining all the time, joining the chat. We've got some really awesome regulars who have just popped up recently and it's been really cool seeing the community building and building and building. So we would love to have you hop on board with the others who are there by joining Club TWiT.

All right back from that little break, and I want to tell you I originally was going to have on Kyle Wiggers from TechCrunch to talk about a story that Kyle had recently written about a new tool that Microsoft has released. Unfortunately, I was unable to make it today, but I still want to go on with this story because I think it's really interesting, and what I love about this piece is that Kyle didn't just talk about a new tool from Microsoft that involves AI, but also talked about the kind of implications of the tool and spoke to some experts about their thoughts on a tool. So Microsoft just recently announced a tool called Correction. Microsoft just recently announced a tool called Correction. This was on September 24th that the tool was announced, and it is a way for people using AI to attempt to correct AI responses, and so it's attempting to limit what we've begun to call hallucinations the times when AI lies. That's the other term for it. Lying does suggest that it has awareness of the fact that it's doing so, but regardless, let's get past that and let's just talk about it.

It's saying things that are incorrect. What correction does is it gives it's kind of a cool tool, because think about using something like GPT-4.0 from OpenAI or Lama from Meta, which was just the latest update was just announced on stage at Meta's Connect conference and with both of those models, right they will. You can query them and you can ask a question and it will come up with an answer. And let's say you are asking it something about a company's earnings report. Okay, so you say, how much money did Microsoft make on Surface devices according to its latest earnings report? And if you were to just ask OpenAI's GPT-4.0 or Meta's Lama, you may get an incorrect answer. You may get an answer that is. You know, I don't have that information. It all depends.

What correction does is it works along the way, because this is an API. This is not me going to chat GPT, talking to chat GPT, which is GPT-4.0, and getting a response. This is for people who are building tools, building apps, building experiences with GPT-4.0, with Lama, and so you would be doing this in kind of a personal system. So you can imagine an analyst, for example, using this tool and trying to get breakdowns of things. So you know, surface devices in the last 10 years. How much did Microsoft make off them? Blah, blah, blah, blah.

Correction works by kind of combining these large language models with small language models and using this to look at the information that's coming out, get the prompt or not the prompt, but the output and then compare it to what they call grounding documents. And these grounding documents, which are facts. That's basically what we're talking about here. We're talking about here an earnings reports transcript, for example. It will compare the response from GPT-4-0 or LAMA to the grounding documents and then say, oh, this said that it was 40,000 or whatever. This says that it's actually 42.7 thousand and will correct it and then provide the answer. So it's just a fact checker along the way. But here's the thing.

Wiggers spoke to multiple experts in the field and kind of talked to them about what are the issues here, what are, what are some of the concerns that we have? And one thing that rang true for me that I thought was kind of interesting is that trying to prevent hallucinations is a little bit of an issue. Because hallucinations are of an issue? Because hallucinations are kind of what AI is doing in the first place. You know, it is taking its understanding of the statistical value of words and where they come in a line and what should come next, and so it's kind of all hallucination. It is all just, uh, that that is why hallucination happens, because it's just looking at what is most likely to follow in, uh, you know, in the response. So by trying to, basically I will, I will argue at the same time that this feels a little bit like a pedantry argument in the sense that they're saying you can't really prevent hallucinations because hallucinations are also the way that you get the right answer, right answer.

But I think that the big thing is one obviously, while it may reduce errors, it may also introduce new errors. So when you're using a smaller model that is trained in the same way or similarly to the model that you're using to get the answer, then that model also has the ability to, and the potential opportunity to, hallucinate. So you may have another situation where the thing hallucinates and you get a different answer. That is also wrong. But I think the biggest part of this that Mike Cook, a research fellow at Queen Mary University, talks about and that I think stands out is if you have something like correction working behind the scenes and you are, let's say, you're the tech, you're the CTO or something right under that, and you have a team of analysts at this analyst firm and you say we're using large language models, we're using AI, generative AI, to help you with your job, and then everybody is going, okay, that's cool, but doesn't this thing have errors? Should we really be using that? You're right, it does, so make sure you check its answers every time, and then later on, you say by the way, we just enabled on the server Microsoft's new correction tool, so now it's going to check your work against the grounding documents that you upload. Here's how you do that. You have the analysts maybe checking in less with. Is this correct? Is this accurate? Because there's this false sense of security that's created, where now, my work is being double-checked. I don't need to triple-check it, it's still AI that's doing it, though, and it could also result in what Cook calls the compounding trust issue, because there are already trust and explainability issues involving AI. So, while analysts you know there may be some analysts that are on board. They're like, oh, it makes my job easier. There may also be the analysts who are like we don't want to use that at all because we don't get it. And then you come out with this new technology and you say, oh well, this is going to check your answers. And then they're going. But how does it check my answer? Wait, but it's the same. It's AI that's checking the AI. There are issues there as well.

Lastly, you know there's also the argument that a correction tool when you have an Okay, think about this. We have seen that companies have really thrown their weight behind generative AI. Right, you've seen many of the big tech companies doing this and many smaller companies doing this. They're all throwing their weight behind generative AI. They're all putting generative AI into their stuff, and one of the biggest complaints about generative AI is still that generative AI has the potential to be inaccurate. So if a big part of your business is selling generative AI products to people and the biggest concern that people have about generative AI is that it might be inaccurate, is it not in your best interest to introduce features that will also correct the AI correct? I put in quotes because maybe it does, maybe it doesn't and then sell those to people who are already buying your products in the first place. So now they're using your AI and they're using your correction tool, and then you start to ramp up the cost on the correction tool as it gets better, or it's using more tokens every single month because it's not only doing the prompting and response, but it's also doing the correction prompting and response.

There's that business motivation concern involved as well. I should mention Google does have a similar tool. It's part of the Vertex AI feature, whereas Microsoft is Azure's AI content safety API. That's what correction is a part of, and Microsoft's tool is a little bit more unique in that it works with multiple AI systems, whereas Google's is currently kind of an internal thing. So if you're using Google's own AI systems, then you're able to use this as a correction tool.

The last kind of issue that stands out here is, of course, an issue that we hear about a lot, and that is the issue of transparency. We were just talking about that at the beginning of the show. As it stands, we don't know how the small and large models were trained exactly and if it differs from the training of the large language models that you're using to get your answer in the first place, and if that training is, you know, part of an accurate data set, etc. Etc. Etc, etc, etc. So that alone is a hurdle to overcome for a company making a decision whether they should buy into this technology. And then, of course, one other thing that goes along with this is and I think that this is part of the process but the fact that it has to rely on grounding documents.

If we're talking about basic facts that a human being can tell fact from fiction, I don't need a grounding document to know that. I can't think of anything off the top of my head, but an example would have been when Google was turning up search results for should I eat a rock, and the response was like yes, you should eat a rock. You should eat a rock once every month, blah, blah, blah. It was obviously coming from training that provided a false answer. I don't need grounding documents to tell me that I shouldn't eat a rock, but the technology that is involved here requires that in some cases, to ground it in reality, as opposed to sort of the statistical possibility of what comes next, statistical possibility of what comes next given a set of training data. So we've got the potential for more errors, we've got the potential for people to trust AI even more and not verify, and we've got the business implications in terms of how much money can be made off of this and whether the technology requires grounding documents or does not. All of that plays into kind of the potential downsides of a correction tool, but at the same time, it's worth noting that something like this is a good thing, or could be a good thing, because we do need to. If this is the way things are going, if generative AI is the future, then we do need to have tools in place that help with this. We just have to have the education going in at the same time. That explains that this is not a trust-don't-verify situation. It is a trust-but-verify situation, so I recommend that everyone go over to TechCrunch. We'll, of course, link it in the show notes, kyle's article. Microsoft Claim new tool can correct AI hallucinations, but experts advise caution. It is a really great piece and, like I said, I love that it was about more than just talking about this tool that Microsoft released, but also taking the time to try to break it down and say what could be the potential issues here versus what is good about it. Folks, that is going to bring us to the end of this episode of Tech News Weekly.

The show publishes every Thursday at twittv slash TNW. That is where you go to subscribe to the show in audio and video formats. I mentioned Club TWiT during the show twit.tv/clubtwit $7 a month. We'd love to have you join the club and see you in the Discord. That'd be great as well. If you'd like to follow me online, I'm at Mikah Sargent on many a social media network, or you can go to chihuahua.coffee. That's C-H-I-H-U-A-H-U-A.coffee where I've got links to the places I'm most active online. Check out iOS today and Hands-On Mac later today, which we'll publish, and you can also, on Sunday, tune in to watch Hands-On Tech, the show where I answer tech questions and review devices and other gadgets and gizmos. Thank you so much for your time and I will catch you again next week for another episode of Tech News Weekly. Bye-bye, we'll see you next time.