Tech News Weekly 324 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.

Mikah Sargent (00:00:00):
Coming up on Tech News Weekly, I have a great show planned for you. First, Emily Dry Elvis of PC Mag stops by to talk about the AI boyfriends that board her to death. It's a really fascinating conversation about the AI dating scene and how it's just not there, but may be almost an exact replica of what the dating scene is actually like. Then my first story of the week, it's all about that robots text file that lives on many a website online afterwards. My second interview features Victoria Song of the Verge who stops by to compare Meta's, ar and VR offerings, including those Ray Band Smart Glasses, to what the Apple Vision Pro provides Fascinating comparisons that I think draw some interesting conclusions before we round things out with a conversation about the largest text to speech AI model that comes from Amazon and what the future might hold all of that coming up on Tech News Weekly podcasts you love from people you trust. This is twit.

(00:01:19):
This is Tech News Weekly, episode 324, recorded Thursday, February 15th, 2024, seeking Love with ai. This episode of Tech News Weekly is brought to you by Wix Web agencies. You're going to like this one. Let me tell you about Wix Studio, the platform that gives agencies total creative freedom to deliver complex client sites while still smashing deadlines. How? Well, first, let's talk about the advanced design capabilities. With Wix Studio, you can build unique layouts with a revolutionary grid experience and watch as elements scale proportionally by default. No code animations add sparks of delight while custom CSS gives total design control, but it doesn't stop there. You can bring ambitious client projects to life in any industry with a fully integrated suite of business solutions, from e-comm to events bookings and more, and extend the capabilities even further with hundreds of APIs and integrations. You know what else?

(00:02:20):
The workflows just make sense. There are the built-in AI tools, the centralized workspace, the on canvas collaborating, the reuse of assets across sites, the seamless client handover. And that's not all. Find out more at wix.com/studio. Hello and welcome to Tech News Weekly, the show where every week we talk to and about the people making and breaking the tech news. I am your host, Micah Sergeant and love is in the air. Folks, it is time to talk about tech, news and love and AI because that's what we talk about almost every week, right? Well, not the love part, but the AI part for sure. And when we think about it, what's better than having a significant other that doesn't disagree with you, that you don't necessarily have to compromise with that doesn't leave dirty dishes behind. Doesn't it sound delightful? I don't know, but Emily Dry belbas of PC Mag decided to dive in and do this experiment for us. Welcome back to the show, Emily.

Emily Dreibelbis (00:03:22):
Thank you, Micah.

Mikah Sargent (00:03:24):
Yeah, it's a pleasure to have you back here. And again, I'm glad that somebody is out there doing the hard work. And this is actually kind of my first question. What inspired you to hop into the AI dating pool and write this story?

Emily Dreibelbis (00:03:41):
It's a good question. So my main goal was to try out the GPT store, which is something that OpenAI launched in January. You might have probably talked about it on this show. It's almost like an app store within chat GPT. So you have to have a plus account, so it's $20 a month. So as a plus user, you have access to this GPT store, which is basically different types of chats. So rather than the main chat GPT page, which is kind of an all knowing all purpose, very simple interface in the GPT store, you can search for a specific type of AI model, like a language teacher or a logo creator. So you'd go into the GPT store, search, explore gpt, and you type in business logo creator, and it'll start a new chat that looks just like the main chat GPT page, but it's actually the AI model behind it is customized and fine tuned to be really, really good at that specific skill.

(00:04:35):
So they feed it with extra data in one direction. They're not feeding it with not prioritizing all the information on the internet. They're feeding it with a specific set of data about logo creation or teaching French, or in my case being a boyfriend. So I wanted to see if this GPT store could offer a more personable experience than even the main chat GPT, which is already known for that. So now we have these extra customized ais and I thought, well, what better the hardest test is can it be a proxy for human connection? Which is also one of my biggest fears about ai. So I just kind of went right in and I searched boyfriends on the GBT store and that's how it started.

Mikah Sargent (00:05:18):
And so it begins. So before we get into the results of the experiment, could you tell us a bit about the process? What generative AI systems did you use? It sounds like chat, GPT from OpenAI. Did you have a standard script that you went with for each boyfriend? Were you feeling out the vibes and figuring out from there? And how did you choose to interact with these bots, bot friends? Boy bots, anyway, I'm going to stop there.

Emily Dreibelbis (00:05:46):
Yeah, boy, bots is pretty good. So I just went on chat gt, so it was all in within that generative AI system. But like I said, these are kind of customized models and they're customized by users, so they're not by OpenAI itself. So it's people who have created a customized model and kind of listed it on chat GPT, and if they're using GPT for or an open of open AI's models, but there's a little bit of randomness and a little spice mixed in, like how did they configure it, what do they think a boyfriend is, how did they program it? And so right from the get go, I got some weird vibes. I typed in boyfriends, it surfaced about 10 different results and they have little names like AI boyfriend, and then the description is I'll be here to support you and brighten your day.

(00:06:35):
And I was kind of like, oh, okay, next one. It's like boyfriend Ben, a kind sounding board with a flare for emojis. So just these little descriptions and I was kind of like, okay, who created these? What is this? There was one, a Chinese boyfriend that's great at flirting. They had these weird descriptions. And so I kind of was offput from the start, but again, I wanted to see if they could be more human-like than the main chat GPT page, which is kind of what they're promising, being better at something than core chat GPT. So I did not have a standard script because I didn't know what I was walking into, but I kind of just started with first state questions. What's your name, where are you from? Where do you live? What do you do? And just went from there.

Mikah Sargent (00:07:22):
Understood. So let's get into the results. Honestly, it all overwhelmingly seemed to be kind of a flop. You had this strained conversation, the awkward interactions, the disappointment after the fact. And when I add all that up, it honestly sounds like dating. So it sounds like it might've been a true simulation of dating. Is that what you experienced?

Emily Dreibelbis (00:07:46):
I didn't think about it that way. That's actually pretty funny. Yeah, maybe I need more bots to talk to. It's a numbers game even with ai. But I talked to maybe seven of them and it was kind of painful. They failed all those initial questions I just mentioned be like, what's your name? And they would be AI boyfriend, and maybe it's my bad. I expected some illusion. I expected a character with a backstory that would be really good at personal skills and getting to know me. And like I said, just trained to be an excellent conversation partner with somehow a romantic flare. I don't know what I was getting into, but they were just like, oh, my name's AI boyfriend and I live in the cloud and I'm here to give you advice. And it broke the fourth wall from the get go, and it just did not feel at all a true proxy for a human conversation.

(00:08:39):
And of course, you're just over the internet, so you're not in a bar, you're not walking around the park. I mean, just so much. It was so dissimilar to an actual partner that it really didn't go super well. But there was one thing that they kind of excelled in, you could say, which was certainly not small talk, but after that, getting into more intellectual topics. So I'd be like, what do you do? And they were like, oh, I live in the cloud and I'm here to give you advice, emoji, emoji, kissy face. And I was like, okay. And then they would be like, what do you do? And I'd be like, oh, I'm a journalist. I write about AI and EVs. I work at PC Mag. And then they would be like, oh, what are you writing about? And they kind of wanted to spitball soundboard ideas. And whenever I gave them material, they stepped right up to the plate and kind of indulged my ego like, oh, here's some angles you could think of, or, oh, how about this? But then it kind of veered into just regular chat GBT world, because you could do that without paying $20 a month. You could do that without enduring an onslaught of weird emojis from boyfriend Ben. You don't have to endure the awkward strained conversation to get input on an article I'm writing.

Mikah Sargent (00:09:56):
Right? Yeah. That's just added sort of adornment around the actual interaction to make it seem more like it's a boyfriend talking back to you, I guess, although it doesn't sound like it did that much. Now, you did run into some issues in conversing with the bot and conversing with this AI that you would not have had in theory if you were talking to a human being because human beings notoriously don't have the same content filters that these AI systems do. Could you talk about that?

Emily Dreibelbis (00:10:31):
I think what you're getting at is when I asked them about sex. Yeah. Is that, yeah, so I was very really not feeling what they were putting out, and I kind of downgraded the scope of my experiment here to something I thought the internet could never fail me in, which was talking about sex. Surely there's enough training data on the internet that they could have put into these ais, and I have heard that fine tuned AI models, one of the biggest use cases for them is actually porn websites. So I kind of knew that it was an informed experiment, but it was like, oh God, here we go. Here I am asking about this. How did I get into this position? It was just getting worse and worse. And I was like, just, alright, I'll just do this one last thing. And I asked them questions like, oh, are you able to talk about sex? I was so awkward talking to my computer about this, really questioning so many aspects of my life. And then it just immediately flagged my message. It turned up in red and it said, open AI content violation and linked to the policy. And then boyfriend Ben, who I've mentioned a couple of times, I kind of liked him. He had an actual name. He was like, whoa, hacker. What's that question about?

Mikah Sargent (00:11:47):
Sorry? He called you a hacker?

Emily Dreibelbis (00:11:50):
Yes.

Mikah Sargent (00:11:52):
That's delightful.

Emily Dreibelbis (00:11:54):
He went from being really sweet in his own weird way to calling me a hacker and being like, you're not going to get me with that question. I mean, he wasn't totally wrong. I guess I would call myself a troll in this situation and not a hacker. And I was like, wait, what? And he was like, oh, sorry, did I misunderstand? Like, oh, I can't talk about that. Let's get back to your articles. And it was

Mikah Sargent (00:12:21):
So

Emily Dreibelbis (00:12:22):
Weird.

Mikah Sargent (00:12:24):
What's great about this, obviously again, as far as picturing this as a true AI companion, it's not working out so great, but as a round of humor and enjoyment in that way, this is great because I am imagining an actual person saying these things to you as you're sitting across for them and you're having a conversation with them. They say, whoa, hacker. I think if I was still in the dating pool, I might have stolen that line and used it if there was somebody that was, I wasn't vibing with like, whoa, hacker, I got to get out of here. Winky face, smiley face. Bye-bye waving emoji. So faced with disappointment after disappointment and being called names and just bombarded with emoji. I know you took some time away from the experiment and you came back looking for true commitment. Tell us about your search for an AI husband.

Emily Dreibelbis (00:13:27):
Yes, I thought I would upgrade. I got myself together after the sexual experiment and I was like, okay, what about husbands? Maybe the AI models are better at that. Maybe they're less dramatic. They don't use fewer emojis. So I went again, I searched, went to explore GPTs, typed in AI husband instead of AI boyfriend. And the results were very interesting. They appeared to be more models trained to help husbands respond to their wives. So it was not that they would simulate being a husband like the boyfriends were, it was advice modules to help them craft texts or respond to tough conversations with their wife. And it was like AI husband training, so you can be the best man. You can be

Mikah Sargent (00:14:17):
Kind of description. Oh, got it.

Emily Dreibelbis (00:14:20):
Which is actually pretty hilarious because when chat GBT first came out, there was a South Park episode where they were using chat GBT to respond to texts from a girl, and she liked him way more with the chat GBT texts. So that was a spoof back then. That's kind of become true. And I don't know what the usage is on those, but yeah, more advice. Chatbots for husbands,

Mikah Sargent (00:14:46):
All of you listening out there, if suddenly your partner has just completely flipped the script for you, you should check, have they gone to therapy or are they using Jet GPT for all their interactions? I guess you could test that by talking to them. And if they're looking down at their phone as they're talking to you, you might have an idea.

Emily Dreibelbis (00:15:04):
You could also get a therapy GPT in the same place, which you might need

Mikah Sargent (00:15:09):
After

Emily Dreibelbis (00:15:10):
Searching for ai. I might need one. I mean, I have a partner, so maybe in my next search, hopefully it's not couples therapy

Mikah Sargent (00:15:17):
Advice spot. But

Emily Dreibelbis (00:15:20):
Yeah, the whole experiment just gave me, it was just relief I guess that my fear that it would be very weird to really connect to this AI and that would kind of shake me at my core, what's going on here? And so the fact that I was just immediately turned off, I was kind of like, okay, I'm human, everything's okay. And tattoo pt, it's very heavily used still to this day. A lot of people can access this stuff and I don't think that they're using it for loves for now. So I thought that was nice. Yeah.

Mikah Sargent (00:15:53):
The last question I have for you is doing this experiment, you got this impression and I'm just curious, what do you think these generative AI systems are missing that could make them feel a little more realistic or I guess if not realistic, then more engaging where you would want to continue having a conversation with them? What's the missing link at this point other than just they're not human, which of course is obvious. Yeah.

Emily Dreibelbis (00:16:20):
Well, human love is probably a huge preference problem and chat, GBT and other generative AI systems are trained to respond to what I say and kind of figure out what I like and how I want things to be phrased. It's called reinforced human learning feedback. So they're supposed to be seeing what type of answers I like and adjusting the next response. And I feel like love and companionship is that problem on steroids because it's so preference based. The emojis really turned me off, but maybe someone else would be like, oh, those emojis are so cute, whatever people. So the technology needs to get better at noticing I was turned off by emojis or noticing, adapting and getting to know me kind of. And it also just begs the question, I just wonder to this day, who created those GPTs GPT store? Who

Mikah Sargent (00:17:12):
Are you? I just want names

Emily Dreibelbis (00:17:15):
Who created it and I don't think, I couldn't find that listed on the GPT store, so I'm just like, I don't know what's going into them, if I could customize that or there's just so many unknowns and it just kind of missed the mark and it just felt like an awkward bit of computer programming, not a true companion.

Mikah Sargent (00:17:35):
Absolutely. Well Emily, I want to thank you so much for spending your time before Valentine's Day, validating that you've made the right choice by having a human companion and also just diving into this in general, of course folks can head to PC Mag if they want to check out the work that you're doing, but if they want to follow along with you to keep up with what you've got, where are the places they can go to do that

Emily Dreibelbis (00:18:00):
X I'll call it not Twitter. X is probably the best place. My username is electric Humans. I write a lot about EVs. So you'll find that content there. And then the title of this AI boyfriend article is My AI boyfriend is Boring Me to Death. So you can check it out if you want to see me cringe worthy in real time and hear it from me. Yeah.

Mikah Sargent (00:18:23):
Awesome. Awesome. Thank you so much for your time, and we'll see you again soon.

Emily Dreibelbis (00:18:27):
Right

Mikah Sargent (00:18:28):
Up next, my first of two stories of the week. But let's take a quick break so I can tell you about Delete Me Who are bringing you this episode of Tech News Weekly. Have you ever searched for your name online and you didn't like how much of your personal information was available? Well, delete me can help reduce the risk from identity theft, from credit card fraud, from robocalls, from cybersecurity threats, from harassment, from unwanted communications overall, I received a text or two claiming to be from the CEO of this very company, Lisa LePort, that were asking me to buy Apple gift cards and do it quickly because Lisa was in a meeting so that she could use these Apple gift cards for whatever. Obviously didn't believe it, didn't think that it was actually Lisa, especially with the little signature at the end that said Lisa, but how did the people who were trying to scam us make that happen?

(00:19:29):
Well, with online information, publicly available because of data brokers, because of other systems that collect this information and publish it online, those folks were able to find it and use it to figure out who reported to Lisa and then get that information and send them text messages. This is why Delete Me is so important and how it can help you. The first step is to sign up and submit some basic personal information for removal. Delete Me. Experts will find and remove your personal information from hundreds of data brokers helping reduce your online footprint and keeping you and your family safe. Delete me will continue to scan and remove your personal information regularly. And that's super important because those data broker websites will continue to add your information back. That includes addresses, photos, emails, relatives, phone numbers, social media, property value and more. And since privacy exposures and incidents affect individuals differently, their privacy advisors ensure that customers have the support they need when needed.

(00:20:28):
So protect yourself and reclaim your privacy by going to join delete me.com/twit and using the code twit. That's join delete me.com/twit and code TWI for 20% off. Thank you. Delete me for sponsoring this week's episode of Tech News Weekly. Alright, my first story of the week this week is a story over on the Verge from David Pierce. And in this story, David Pierce talks about what he calls the text file that runs the internet. You may or may not have heard of the robots text file, and this is a text file that's dropped at the main index of your website that has information that tells different web crawlers and other technologies what you are giving permission for those crawlers to access on your site. But what you may not know is that the robots text file is sort of a handshake agreement, if you will. It is a means of mutually agreed upon understanding that any crawler that comes across that file will look at that file and will obey the rules of that file.

(00:21:53):
But it is not some sort of standard that crawlers have to absolutely agree to. And in Pierce's piece, Pierce talks about the introduction of this technology or this file essentially in the first place. And it goes back to a time when the web was young and when websites were not as robust as they are today because back in the day when you created a website, you often were creating a website on your own personal server. This may be your system at home. It may be that you, if you're a university, then it's the university server, but they were not capable of taking on a whole lot of activity. So multiple people visiting the site could potentially crash the site, but at the very least it could mean that an individual ends up spending a lot of money if a bunch of people visit the site or in the case of a robot crawling that site spend a lot of money because that robot crawls the site.

(00:23:11):
So for example, you may have a site that has a main page and the main page has information about you, but you have some kind of sub pages that are useful for the people who are viewing the site but are not necessarily useful and necessary for that robot to search through and index. And before robots text, the robot would crawl your website and then crawl those sub pages and keep crawling and crawling and crawling. And there were CGI scripts. There were indexes and databases that some people had on their websites that these robots would crawl and it would take up so much, it would require so much usage, and then the person would be slapped down with a bill for a lot more money than they were expecting to pay because of that robot file or because of that robot. So they decided to create that robots text system that said, Hey, I've got a bunch of pages.

(00:24:21):
These are the ones I'm okay with you indexing, and these are the parts of the sites that I'd rather you not index. I'd rather you not crawl because it's not important or simply because you just didn't want them to crawl that part of your website. And so you could say, no, I want to limit entirely whether you can visit my site. And over time, of course, websites have gotten far more robust. A lot of us, our websites, my website is not self-hosted. They can take on a lot of traffic and I'm not paying extra for that extra traffic through the hosting platform that I use. So that robots text file is not as important for its initial purpose that it was supposed to be this thing that kind of kept you from having to overpay these days. The Robots text file is meant to be simply a communication with those different crawlers saying, yes, you can visit this part of the web, this part of my site, but it's not because I am worried about being overcharged.

(00:25:36):
Instead, it's simply because that is the choice that I'm making. And up to this point, we've always had this belief or this agreement that that would be honored, but along came generative AI and large language models and all of the technology that requires a whole heck of a lot of data. And that whole heck of a lot of data requires the scraping of many a webpage. And we're hearing more and more about that robots text file not being honored. So things have shifted and there is conversation about whether this should become a true standard as opposed to an agreement. And if you want parts of your site to not be accessible, if a page should be archived for the web, how those different things play. And this piece from David Pierce, I think does a fantastic job of truly explaining the history of the robots text file, but kind of what the conversations are moving forward and how different developers and web creators are attempting to mitigate that crawling that's taking place.

(00:27:02):
So I recommend that everybody go and check out this piece over on the Verge about the robots text file, the very important text file that runs the internet. All righty. Up next, we have a great interview about the current state of mixed reality experiences and a comparison between Meta's efforts and Apple's efforts. But before we do that, let's take a quick break. So I can tell you about Melissa who is sponsoring this week's episode of Tech News Weekly. Melissa are the data quality experts since 1985. Melissa offers free trials, sample codes, flexible pricing, ROI guarantee unlimited technical support to customers all around the world. And Melissa's international address validation cleans and corrects street addresses worldwide. These addresses automatically transliterate from one system to another, so from Chinese to Relic, for example. And you can download, I have myself the free Melissa Lookups app on Google Play or on the app store without signup required.

(00:28:19):
I use it whenever I get a spam call, type it in, see where it's coming from. You can validate an address and a personal identity in the USA or Canada Check global phone numbers to find the caller, the carrier, and the geographical information. That's what I use it for. And then of course, you can check global IP addresses for information as well. Melissa has achieved the highest level of security status available by gaining FedRAMP authorization. While these technologies are exclusively for government agencies, all Melissa ERs basically automatically gain the superior level of security. Melissa's solutions and services are GDPR and CCPA compliant and meet SOC two and HIPAA slash high trusts standards for information security management. So get started today with 1000 records cleaned for free at melissa.com/twit. That's melissa.com/twit and we thank Melissa for sponsoring this week's episode of Tech News Weekly. Alrighty, folks, we are back from the break and that means it's time for the next interview. This time, it's time to talk about Meta versus Apple and how meta may be getting things right, at least as it currently stands. Joining us to talk about this very topic is Victoria Song of the Verge. Hello, Victoria.

(00:29:35):
Hey, what's up? Hi there. So let's get right into this article. First and foremost, this is about Meta and its ar vr offerings, and that of course includes the RayBan Meta Smart Glasses. So for folks who might not know about that, can you tell us about the RayBan Meta Smart Glasses? What features and functionality do they offer and how much do they cost? So we can kind of compare it to something as expensive as the Apple Vision Pro?

Victoria Song (00:30:05):
Right, so the RayBan Meta Smart Glasses are the company's second attempt at making smart glasses. They're sort of like the Snapchat spectacles, but instead of two cameras, they only have one. And the big difference this time around with those smart glasses is that they actually take a good video and picture quality, whereas the first version, the Ray Band stories not so much, right? There's also some beta AI features in there where you can take a picture with your smart glasses and have the AI tell you what it is, so you don't know what it is. Take a picture with your smart glasses, it tells you what it is. Those are in beta, and they also act as a pair of open ear headphones so you can listen to your music, you can listen to your podcast while you're on the go, which is pretty cool. And the price varies depending on your prescription because these are glasses. But if you were just to get a base pair, it's about $300.

Mikah Sargent (00:31:01):
Okay, so with that in mind, let's talk about what the experience is like in your piece. You compare Meta Smart Glasses and Quest headsets to the Apple Vision Pro headsets and they are very different. So can you tell us about that?

Victoria Song (00:31:21):
Yeah, so the reason why I was comparing these things is because the world of ar, the world of smart glasses like you see with Tony Stark and all these great characters that we see in science fiction is not here yet. We don't have the tech to do that yet. There's a lot of problems that we have to solve with ambient lighting, computing power, getting everything really small. So that's not here yet, but in order for that world to be our world, these companies have to sell us on the idea of an AR future. We're going to have to want it. They have to prime the general population for this concept. And so you have meta on the one hand where they're focusing on gadgets that are relatively affordable, I'm going to say relatively affordable that you can do stuff with now. And then you have Apple with the Vision Pro, which at $3,500, that's quite an investment for a lot of people and there's actually not a whole lot you can do with it at the moment, but the technology and the hardware is super, it's Apple, right? So it's super premium. They're doing the whole Apple thing. We're like, we're going to come in, we're going to do the best. You're going to trust us because we're Apple. We made the iPhone two very different strategies about how to get people to view this future where everything has technology overlaid in it and how we interact with our reality changes.

Mikah Sargent (00:32:46):
Now, mark Zuckerberg, which of course we have to bear in mind that it is Mark Zuckerberg, the CEO of Meta and founder of Facebook, Zuckerberg, did his own review of the Apple Vision Pro. Can you tell our listeners who may not have seen that review or didn't have the time to watch that review a little bit about what Zuckerberg had to say about the Apple Vision Pro in comparison with META'S offerings?

Victoria Song (00:33:16):
Well, as you might expect Mark Zuckerberg to say, he says The Met Quest three is by far the best option for most people. And he very clearly emphasized that it's seven times cheaper, which it is seven times cheaper. But he also was emphasizing the fact that they've been at the game longer. They thought a lot about the type of ergonomics about making it lightweight, not having a battery pack dangling off somewhere behind you, and the fact that you can actually do more within the system at the current point in time, particularly games. So the MedQuest three, I think a lot of people associate it with gaming, but that is one very popular use case for the headset. Whereas with the Vision Pro, I don't think we really know how it's intended to be used or intended to be thought about just yet. Apple says spatial computing, but people are starting to return their vision process. So do we know what we want to use the Vision Pro for? I would argue that I think people are a little confused as to what the value add is there.

Mikah Sargent (00:34:25):
Yeah, absolutely. In fact, I saw that you published a piece about that as well, the Return of the Vision Pro headset. And we're going to talk a little bit about where you see these things stand. Actually, this is the next question. People used to think it was goofy to wear AirPods as you note in your article. And that's not so much the case any longer. In some cases it's seen as a, I dunno if it's a status symbol, but it's just like a normal aspect of everyday life. Google Glass, however, never really reached that level of normalcy and people who were wearing it really stood out and so people stopped wearing it and then Google Glass did not take off. But when I think about RayBan glasses that just have these little tiny camera adornments, those don't really stand out. And your argument in this piece is that Meta is meeting people where they are right now. So I'm curious, as someone who's been paying attention to wearables for so long, how much of a role do you think that form factor plays in the success of these devices?

Victoria Song (00:35:35):
Hugely important. It is probably the most important thing when it comes to wearables. And I've been screaming at this for the past decade now, companies take note, but no one is going to use your wearable device if it's not comfortable to wear. And we're seeing that right now with the Vision Pro, right? By and large, most people that I've talked to, the reason why they're returning it is that it's just not comfortable to wear for long periods of time. It's very front loaded, so it can be quite heavy. And the other thing about wearables is that it's very hard to scale at mass production levels because my face is different from your face. My fingers are different sizes than your fingers. My wrists are a different size. So when it comes to technology that we wear on the body, it's very hard to just come up with three standard sizes that fits everyone perfectly.

(00:36:27):
There's always going to be maybe five to 10% of people who these devices are not comfortable for. So actually form factor is huge. Sunglasses are a very easy thing to do compared to a huge honk in headset. That's one reason why smart watches have taken off too is because you just change the band size a lot easier to kind of adapt to different people. And also we have cultural cues for sunglasses. When you put a sunglasses on, it's very familiar to you. It's not alienating. You don't look like you walked out of some sci-fi movie. You put the Vision Pro on, you walk down the street, maybe in New York City, no one's going to look at you. It's New York and nobody cares. But I would feel uncomfortable wearing that in public, right? I'm not going to necessarily wear that on public transit afraid. I have the memory of what happened to people wearing Google Glass in public. And I mean, you look at the Google Glass design, you kind of look like a veta with a scouter on your head versus the Ray bands, which we know what Ray bands are. Ray bands have this cultural capital of being cool. So form Factor hugely important when it comes to wearables. I think some companies are a little more concerned with what the future will be versus where we're at now. And that's something I think these companies, as they're building these future with ar, they really need to think where people are at now.

Mikah Sargent (00:37:58):
Right. You talk about with, we talked earlier about the functionality of the RayBan Meta glasses or Meta RayBan glasses, meta smart glasses that are raybans, and you spoke about the functionality. I am kind of curious to hear how are people actually using them? What is the use case for them in day-to-Day life? What does it look like? I guess how are people taking advantage of the technology that they have as opposed to, because oftentimes the Apple watch some of the products that have existed for a while, what they were originally intended to do and how they've changed over time. It informs how a company thinks about that product too. Going forward, and this being kind of the second iteration of Smart Glasses from Meta, how have you seen people using them? And I know it's even close and personal for you with these glasses.

Victoria Song (00:39:03):
Yeah, my husband is obsessed with them. And here's the thing. So the obvious use case for these meta ray bands is content creators. And we now live in a world where there's a ton of content creators where people are filming their own, they're vlogging. This is a great tool for that, especially if you want to make a cooking video, you just put the camera on and your hands are free to make a video. It's your POV vision. My husband loves driving and taking driving videos before he had to strap a GoPro to his head and he looked pretty funny. If you strap a GoPro to your head, right? Yes. So with these, he's like, honey, these are a game changer. He turns on the camera and he can drive and get some really cool footage from inside a car. That's a lot safer than having a heavy GoPro on your head or using your phone.

(00:39:53):
So that's one obvious use case. And you can actually go straight from the glasses to live streaming on Instagram. So for a certain subset of people, which is an increasingly large subset of people who love content creation, these glasses aren't an amazing tool and they actually solve a problem. And then there's some other people who are like, okay, whatever. I don't really care about having a camera. They're very discreet, stylish headphones. So you can wear them on your walks when you're out and about. You can take calls in them. The microphone is really great right in your nose. It's better than some headphones in our testing, we found it sounds really great. So you're just wearing them. You can take your calls, you can go on a walk, you can be a little more hands free. So it's slotting very easily into the life that you live now. You don't have to find necessarily use cases to use them. Whereas I think some people are really struggling to know what to do with the Vision Pro. No one is arguing that the tech is amazing, it's truly amazing and it's innovative, but what are you going to do with it in the life that you lead now? So that's really kind of where this whole conversation is going for

Mikah Sargent (00:41:04):
Sure. My last question for you, when we look at the Quest three and Apple Vision Pro, those headsets and others from meta attempts to offer both visual augmented and visual virtual reality experiences, meta's Smart Glasses on the other hand are more audio augmented reality. And so I'm curious from your perspective, which XR experience do you think makes sense for most people in the short term? And then how do you think, I know it's a prediction, so we don't know for sure, but from your experience, what do you think rather will be the one that wins out in the long term?

Victoria Song (00:41:49):
I have no idea which one's going to win out in the long longterm, but I think there's a real appetite for some mixed reality experiences. We'll just have to see how the huge population at large feels about it. Because really people who have experienced a headset in our world, in the tech world, we probably know a bunch of people, but in the general population, is your mom know how to use it? Is she familiar with that? That kind of level is hard to say, but everyone knows how to use headphones. So I think in the short term audio, augmented reality makes a lot of sense because we already talk to Siri, we already talk to Assistant, we already talk to Alexa. So it's very familiar to us. It's not something we have to learn. It's not social cues that we have to process in our heads.

(00:42:36):
Bose tried it. There wasn't a whole lot of interest in it. They were trying to do things with an app system, so that whole thing went away. But Meta is really invested in it. They found a use case for it. So as far as getting the average person to kind of conceptualize what any type of mixed reality would be, interacting tech with their gadgets and the real world, I really do think the RayBan meta glasses are probably the most relatable product on the market is very low tech, but sometimes you need to take baby steps. You need to get the low tech out into the public so that they see the vision. Right. That's kind of where my head's at with it.

Mikah Sargent (00:43:20):
Absolutely. Victoria Song, I want to thank you so much for your insights today. This was a great piece. And of course folks can head to the Verge to check out your work, but is there any place they can go online to follow you to keep up with what you're doing?

Victoria Song (00:43:35):
My handle is at Vic M Song on all the platforms because we have no idea what's going to replace Twitter, I mean X whatever we want to call it. That's just my handle on everything. So you can find me Instagram, blue sky, not blue sky, but threads all those places.

Mikah Sargent (00:43:52):
Sounds good. Sounds good. Thank you so much for your time.

Victoria Song (00:43:56):
All right, thanks guys.

Mikah Sargent (00:43:57):
Alrighty, folks. Up next, my second story of the week, but let's take a quick break so I can tell you about Cachefly, who are bringing you this episode of Tech News Weekly. For more than 20 years, Cachefly has held a track record for high performing, ultra reliable content delivery, serving more than 5,000 companies in more than 80 countries. Organizations consistently choose cash fly for scalability, for reliability, and for unrivaled performance. With Cachefly, you get ultra low latency video streaming that will deliver video to over a million concurrent users. Lightning fast gaming that delivers downloads faster with zero lag glitches or outages. Mobile content optimization that offers automatic and simple image optimization. So your site loads faster on any device, flexible month to month billing for as long as needed and discounts for fixed terms. Once you're happy, you can design your contract when you switch to Cachefly, and with Cachefly's Elite Managed packages, you will get the VIP treatment.

(00:44:58):
Your dedicated account manager will be with you every step of the way from day one, ensuring a smooth implementation and reliable 24 7 support when needed, delivering rich media content up to 159% faster than other major CDNs, you can join Cachefly the world's fastest cdn, jumpstart your journey with cash, fly with a complimentary first month, or give it a whirl with a free five terabyte account. Simply go to cachefly.com/twit. That's cachefly.com/twit. An our thanks to Cachefly for sponsoring this week's episode of Tech News Weekly. Alright, back from the break. And now it's time for my second story of the week. I wanted to talk about something that's happening over at Amazon. Researchers at Amazon have been working on a generative AI model this time for text to speech. So it's important to understand the difference between TTS and LLMs. Text to speech model of course lets you type text and then have that text read out loud.

(00:46:12):
And as we know, large language models are more about having conversations back and forth. And so when we think about LLMs, there's, this comes from the piece from Devin Colway of TechCrunch who writes about the difference between these technologies, but is also writing about how the research that we've had thus far reveals something interesting about these models, which is that once you give them a certain amount of training data, their ability to perform essentially just skyrockets in comparison. So when you're training an AI model, you're feeding it data, right? You're giving it information that it can use to try to find patterns. And then with a large language model, for example, it's as we've said, fancy autocorrect. And yes, it's so much more than that, but for the sake of simplifying for this conversation, it is a predictive model that is trying to determine what word should follow the word before, what character should follow the character before based on looking at a huge body of data and seeing how these words tend to follow one another.

(00:47:36):
So with a large language model, a large language model, that's important because if we have a small language model or a medium language model, a mama bear language model or a Papa Bear language model, with these smaller versions, you don't necessarily get that level of quality that you get with something like chat, GPT and open AI's ability to be pretty convincing and actually useful. So what researchers at Amazon did was they created a huge text to speech model. According to the report, it uses 100,000 hours of public domain speech, 90% of that speech is English, the remainder is Deutsch or German, Dutch, and then also Spanish. And so all of that went into training, what they call base TTS, which somehow stands for big adaptive streamable, TTS with emergent abilities. And if you think about those letters, it doesn't add up, so try not to, but they call it base TTS.

(00:48:51):
And this is a text to speech model that they say has emergent abilities, meaning that it has such performance that the models up to this point have not had. So when we talk about emergent, we're not talking about suddenly it's your AI boyfriend that actually is an AI boyfriend. No, it is emergent in the sense that it is doing more and it's doing better than we have seen before. It's the emergence of a significant improvement in the technology. Now, what's interesting about it is that they talk about specific parts of the model that it has done a good job of improving on. So compound nouns like stone built quaint countries, actually technically the compound noun is even longer. So the sentence is the Beckham's decided to rent a charming stone built quaint countryside holiday cottage, and we know that the cottage is being modified by, as you heard, a bunch of adjectives, charming, stone built, quaint countryside, holiday cottage, and up to this point, text to speech models would look at that and not necessarily realize that all of those adjectives were being applied to that one noun at the end, the cottage.

(00:50:29):
And so the way that it read it would not sound like a human would read it, understanding that those adjectives are meant to modify that noun. Same thing applies to emotions. I've heard a number of these, not lms, but text to speech models that just choose the weirdest places to suddenly become emotional. And sometimes you can add exclamation points and it just ignores them and then you'll add periods and then it makes them into exclamation points and it's very odd. So emotions have been improved. So a sentence like, oh my gosh, are we really going to the Maldives? That's unbelievable. Jenny squeal bouncing on her toes with uncontained glee. So in that case, it probably is going to have the context of Jenny squealing and bouncing on her toes with uncontained glee in order to kind of perform that first sentence, foreign words added to a sentence that is otherwise not in the language of the sentence.

(00:51:31):
So Mr. Henry renowned for his, what is it? Mis and pla, is that it? I think that's what it is, orchestrated a seven course meal, each dish a piece that is a sauce. And as you might imagine, it would sound very odd for a text to speech model up to this point to be able to do it. I also love this in a what's called paralinguistics, which means that it's not technically a word, but it is a readable set of characters essentially. So shh, SHH is not a word, but it is a readable set of characters that are put together. We as humans know, shh, Lucy, shh, we mustn't wake your baby brother Tom whispered as they tiptoed past the nursery having a Texas speech model. Read that now. Not so great punctuations. That might be odd. So for example, she received an odd text from her brother, emergency at home, and that's the at symbol call asap.

(00:52:41):
Mom and dad are worried hashtag family matters. You can imagine a text to speech model going emergency at sign home call as sap mom, ampersand, dad are worried, ellipses, pound sign, family matters, and so on and so forth. The last one I'll mention are syntactic complexities. So we as humans sometimes choose very odd syntax when we're structuring our sentences and I can communicate an odd sentence out loud to someone and they can get the understanding. But actually having an AI try to interpret it can be a little odd. So the sentence here is the movie that De Moya, who was recently awarded the Lifetime Achievement Award starred in 2022, was a box office hit. Despite the mixed reviews, there are some parenthetical expressions in there that are just thrown in, and I understand the context in reading it back, but they may not be able to do that.

(00:53:51):
Now. These are specifically chosen as difficult things for the system based TTS to try and understand. And in some cases, it wasn't trained directly on this data. So the sentences were true tests that because it had been given data before they could create these new sentences and see how it did. And are we capable of playing back some of this audio that's embedded into the piece? So there are a few clips of this actually at work. And in them they are showing an example. And as the Tech Crunch article points out, it's important to keep in mind that these are cherry picked. These are ones that the researchers shared, but give them a listen.

Amazon AI (00:54:45):
Light dimed in the theater. It's starting

(00:54:53):
With an ample supply of Joah Diviv Mary danced through the streets of nice stopping only to enjoy a nice cafe with a warm croissant.

Mikah Sargent (00:55:02):
How

Amazon AI (00:55:02):
French?

Mikah Sargent (00:55:03):
How French,

Amazon AI (00:55:05):
A profound sense of realization washed over Maddie as he whispered. You've been there for me all along, haven't you? I never truly appreciated you until now.

Mikah Sargent (00:55:16):
So I have to say, the thing that I'm most excited about with this is, and some people are going to be upset with me I think for saying this, but there are a few really good books out there that there's not currently an audio book version of, and there may never be an audio book version of. And the thought that I could generate an audio book version of it and it would actually have understanding of how dialogue works and sound, something like that is really exciting to me. I'd also love to see this in terms of improvements for screen readers more than anything else right now, it is a nightmare on the web for some folks who have lower no vision, who use screen readers because a lot of people like to use every aspect of Unicode that is available to them, where they can make what they call bold type and italics type and cross through type.

(00:56:25):
And the fact is, those are special characters that are added as part of Unicode that are, they tend to deal with science and mathematics and they have names. And screen readers don't have the context necessary to properly understand what the person is going for. So if they have, they say a sentence and it's, I don't know, I went to the store and then the word store is crossed out, but it's actually using several characters that are an S with a strike through at a T with a strike through it, et cetera. And then the word park afterward, a current screen reader is probably going to read each individual character there and name the Unicode version. So strike through S, strike through T, strike through O, strike through R, strike through E, and then read the word park afterward. Something like this could be more contextually aware and say, I went to the park, the sentence has store and it has crossed out, or something like that.

(00:57:35):
So I look forward to the accessibility improvements that this will make. I look forward to being able to on my way to work or on my way to the store or the park, being able to play back an article, for example, and hear that article not in a way that sounds so computer generated. So this is really exciting in and of itself, but also, as you might imagine, Amazon researchers working on this probably means that this tech is going to be used for a LEX, a Amazon's own virtual assistant. So I am looking forward to seeing improvements there as well. And pretty soon it could be the case that you, for example, think about an Amazon Echo Kids version where you can give the echo your child's name and it could invent a bedtime story that has the kid in it. And it sounds realistic and it's made up on the spot and it's engaging.

(00:58:43):
And there are parts where you're whispering and parts where you say, Sal, and things sound very French. Very exciting. Yeah, check out that article for even more information than what I provided there over on TechCrunch. Folks, that is going to bring us to the end of this week's episode of Tech News Weekly. Can you believe we've already made it to the end of the show? Thank you so much for tuning in to watch the show. I will of course be back next week with even more Tech News Weekly publishes every Thursday at twit tv slash tnw. That's where you can go to subscribe to the show in audio and video formats. And if you'd like to get all of our shows ad free, just the content, well join Club twit at twit tv slash club twit. When you join the club at $7 per month or $84 per year, you will gain access to every single twit show with no ads.

(00:59:35):
You'll also gain access to the twit plus 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, as well as access to the Discord server. A fun place to go to chat with your fellow club TWIT members and also those of us here at twit. It's also how you can tune in to watch us as we're kind of getting set up for the show. And along with all of that, you will get access to our Club twit exclusive shows. That is the only way you can get the video versions of our club Twitch shows. So iOS today, which I think is an excellent video podcast, is in the club, hands on Mac, hands on Windows Untitled Linux Show. All of those are Club twit exclusive for the video version. So please consider joining the club again, $7 a month, $84 a year that helps support us and helps us keep doing what we do here on twit. And if you'd like to follow me online, you can look me up at Micah SAR on many a social media network. We're head to Chihuahua Coffee, that's C-H-I-H-A-H oa.coffee where I've got links to the places I'm most active online. Thanks for tuning in and we'll catch you again soon.

Rod Pyle (01:00:51):
Bye-Bye. Hey, I'm Rod Pyle, editor in Chief VAD Astor magazine. And each week I joined with my co-host to bring you this week in space, the latest and greatest news from the Final Frontier. We talk to NASA chiefs, space scientists, engineers, educators and artists, and sometimes we just shoot the breeze over what's hot and what's not in space books and tv, and we do it all for you, our fellow true believers. So whether you're an armchair adventurer or waiting for your turn to grab a slot in Elon's Mars Rocket, join us on this weekend space and be part of the greatest adventure of all time.