Tech Brew Ride Home - Mon. 02/05 – The Most Shocking Deepfake Scam Of The AI Era

Episode Date: February 5, 2024

Let me tell you the story of the most shocking AI deepfake scam we’ve heard yet, because it’s a warning to all of us going forward. An Apple Vision Pro teardown explains why EyeSight looks so blurry. Another sign that Google is losing interest in the web. And an interesting peek behind the curtain revealing the economics and motivation of tech media. Sponsors: ShipStation.com/ride code: ride Links: Finance worker pays out $25 million after video call with deepfake ‘chief financial officer’ (CNN) Vision Pro Teardown—Why Those Fake Eyes Look So Weird (iFixIt) Google Search’s cache links are officially being retired (The Verge) TECHCRUNCH+ TERMINATION ("Securities") YouTube Video Of Vision Pro Demos Learn more about your ad choices. Visit megaphone.fm/adchoices

On April 4th, 2023, around 2 in the morning, a man was found stabbed multiple times on a sidewalk in downtown San Francisco. Hey, who did this to you? What happened next turned the story into a political firestorm. Reports have identified the victim as Bob Lee, the founder of Cash App. From Bloomberg Podcasts, this is Foundering, the Killing of Bob Lee, beginning April 16. Welcome to the TechMeme right home for Monday, February 5th, 2024. I'm Brian McCullough today. Let me tell you the story of the most shocking AI deepfake scam we've heard yet because it's a warning to all of us going forward. An Apple Vision Pro teardown explains why eyesight looks so blurry. Another sign that Google is losing interest in the web and an interesting peek behind the curtain revealing the economics and motivation of tech media. Here's what you miss today in the world of tech. Something, something. Welcome to the AI era. An employee. employee at a company in Hong Kong was tricked into paying $25 million to fraudsters who used deep fake technology to pose as the company's CFO and staff during, and this is the important

part, a video call. Quoting CNN. The elaborate scam saw the worker duped into attending a video call with what he thought were several other members of staff, but all of whom were, in fact, deep fake recreations, Hong Kong police said at a briefing on Friday. In the multi-person video conference, it turns out that everyone he saw was fake. Senior Superintendent Barron Chan Chongqing told the city's public broadcaster, R-T-H-K. Chan said the worker had grown suspicious after he received a message that was purportedly from the company's UK-based chief financial officer. Initially, the worker suspected it was a phishing email, as it talked of the need for a secret transaction to be carried out. However, the worker put aside his early doubts after the video call because other people in attendance

had looked and sounded just like colleagues he recognized, Chan said. Believing everyone else on the call was real, the worker agreed to remit a total of 200 million Hong Kong dollars, about 25.6 million U.S. dollars, the police officer added. The scam involving the fake CFO was only discovered when the employee later checked with the corporation's head office. Hong Kong police did not reveal the name or details of the company or the worker, end quote. So the thing that I find fascinating about this is, you know, you and I like to think of ourselves as sophisticated internet users, right? Don't click on links from people you don't know, scrutinize the actual email address to make sure it's coming from a reputable domain, two-factor authentication, the whole lot, right?

But we, even us, all of us, need to start thinking differently. As Gergely Oros said of this on threads, quote, As those working in tech, we owe it to educate people around us on the massive fraud potential for AI. You should absolutely be distrustful of video calls, phone calls, emails, DMs on social media. Assume it could be a deep fake even if it looks real, end quote. Exactly. Most scams work via social engineering, right? And until now, proof of life in a way in the form of, oh, that clearly has to be that person.

They look and sound like them. Look, this is their LinkedIn profile. Hey, they just FaceTimed me. Must be them. All of that is out the window now. Quoting AI fray on threads, this is the most shocking AI deep fake crime reported to date. It sounds like the kind of AI-related threat that policymakers should be far more concerned about. Case in point, there's nothing in the EU AI Act that would serve to prevent a crime of this pattern in the EU. Nothing, end quote.

Right on schedule, I Fix It has an Applevision Pro teardown, among other things they say they learned that it's not great, repairability-wise, has an unforgivable proprietary battery plug, and they think they figured out why those eyesight fake eyes look so blurry from the outside. Quote, it turns out that when the eyesight displays your eyes, it isn't just displaying a single video feed of your eyes. It's showing a bunch of videos of your eyes. Exploring inside the glass shell, we found three layers for the front-facing display, a widening layer, a lenticular layer, and the OLED display itself. Apple wanted to achieve something very specific and animated 3D-looking face with eyes. They had to make very strategic design choices and compromises to accomplish this.

Human brains are very sensitive to faces and expressions. It's why the uncanny valley is a thing, and part of that is depth sensing. Apple needed to create a believable 3D effect. One reason why 3D renderings don't look truly 3D is because they lack a stereoscopic effect. For something to look 3D, we need to see subtly different images with each eye. eye. The Vision Pro tackles this problem with lenticular lenses. A lenticular lens displays different images when viewed from different angles. You can use this effect to simulate movement with two frames of an action, or you can create a stereoscopic 3D effect with images of the same subject from different angles. The Vision Pro has a lenticular layer on top of the exterior

OLED panel. VisionOS renders multiple face images, call them A and B, slices them up, and displays A from one angle serving your left eye, and B from another serving your right eye. This creates a 3D face via the stereoscopic effect. And those angles are tiny, and they are legion. It takes a fancy, evident, scientific microscope to really see what we mean. There are compromises to this approach, however. The horizontal resolution is dramatically reduced being divided between each of the multiple images. For example, if two images are displayed on a 2,000 pixel-wide display,

each image has only 1,000 horizontal pixels to work with. Even though we don't know the resolution of the display, nor do we know the number of images being interwoven, the resolution is necessarily reduced, and that is a major reason why eyesight eyes seem blurry. In front of the lenticular layer is another plastic lens layer with similarly lenticular ridges. This layer appears to stretch the projected face wide enough to fit the width of the Vision Pro. Removing this layer and booting the Pro showcases some very oddly pinched eyes. Additionally, the lens likely limits the effective viewing angle, limiting the effect to directly in front of the Vision Pro limits,

artifacting you might see at extreme angles, sort of like a privacy filter. The downside is that you're passing an already complex blurry image through yet another layer of lens. This makes it even blurrier and darker, end quote. Folks, I'm telling you that the underlying structure of the web, as we've understood it for 25 years, is shifting right now under our feet. Another case and point. Google has dropped the cache link from search results snippet.

They did this apparently last week, and they plan to remove the cash functionality entirely, quote, in the near future. Quoting the verge. The cash feature historically lets you view a web page as Google sees it, which is useful for a variety of different reasons, beyond just being able to see a page that's struggling to load. SEO professionals could use it to debug their sites or even keep tabs on competitors, and it can also be an enormously helpful news gathering tool, giving reporters the ability to see exactly what information a company has added, or removed from a website and a way to see details that people or companies might be trying to scrub from the web. Or, if a site is blocked in your region, Google's cache can work as a great alternative to a VPN. The removal of Google's cache links has been taking place gradually over the past couple of months and isn't complete just yet. Over at Search Engine Roundtable, Barry Schwartz spotted that the links were disappearing intermittently from search results in early December and then removed entirely

as of the end of January. In his tweet, tweet. Danny Sullivan of Google confirmed that in addition to removing the links, the cash search operator will also be going away, quote, in the near future. Although the cash links are only now being discontinued, the writing's been on the wall for a while. In early 2021, Google Developer Relations Engineer Martin Split said the cache view was, quote, basically an unmaintained legacy feature. It doesn't sound like Google has any immediate plans to replace the feature, but Sullivan says he hopes that Google could add links to the Internet Archive that could instead be, used to show how a web page has changed over time. No promises, he cautioned. We have to talk to them,

see how it all might go, involves people well beyond me, but I think it would be nice all around, end quote. So we joke all the time on the show about Google deprecating features like hot potatoes, but this is them literally deprecating the caching of the web, the web, the thing that they were built to index. We know Google eliminated the don't be evil part of their motto years ago, but remember this part of the motto, quote, Google's mission is to organize the world's information and make it universally accessible and useful, end quote. As I've been suggesting, Google strategy and motivations as a company are changing and perhaps organizing the world's information and making it universally accessible, starting with the web, it might not be a core part of Google's

mission anymore. Finally, today, we didn't get to cover this, but TechCrunch laid off about eight employees last week and announced that it is winding down TC plus. which was its paid subscription products launched back in 2019. Danny Crichton is now the editor-in-chief of Media for Lux Capital. Danny was the one who interviewed me for their securities podcast a couple months ago, which I shared as a bonus episode. But before his current job, Danny was executive editor of TechCrunch and helped launch TC Plus. This weekend in his securities newsletter, Danny outlined how and why TechCrunch had to do these layoffs, and I thought it was interesting to share some bits of this with you, because it sheds some light on the economics of the tech media that we

quote from all the time on this show. Read the whole thing for explanations of YTC Plus went away, but also listen to this and understand in a way that we've not really talked about some of the motivations behind how tech gets covered and reported. Quote, what's striking about tech crunch is its longevity. It's become the rare survivor of the graveyard of digital media, surviving successive waves of rapid change since its founding in 2005 before the launch of the iPhone and just a year after Mark Zuckerberg debuted Facebook. Its business secret was simple but hard to replicate and equally hard to scale. Its ad space was much more valuable than other digital media companies and it had a terrific and consistent events business. On ads, TechCrunch became a key battleground

between tech companies for the hearts and minds of its audience of early stage founders who might accidentally be locking in purchase decisions worth hundreds of millions of dollars as they built their startups. Profitably for TechCrunch, its ads were bought to lock out competitors in massive multi-billion dollar markets. TechCrunch's beat coverage is on new startups disrupting the old incumbents, and those incumbents wanted to be sure that no one forgot they were still in the fight. That gave TechCrunch key leverage on its ad space that few other digital media companies could match against a culture-focused site like HuffPost, a system, brand for a while via AOL and then Oath and then Verizon Media. TechCrunch's ad space could be

five to ten times more valuable on a CPM basis, the standard ad metric of a thousand page views. Matching those revenues was a structural advantage in terms of traffic. As one of the most venerable sites covering tech on the web, major announcements from Elon Musk, Tesla, Apple, Facebook, and other big technology companies drove heavy traffic to TechCrunch. Most of this was relayed via Google search and Google News, and at times, more than 90% of the site's traffic came from just those two sources. Critically, this coverage was eminently affordable. Writing up an article on the latest ravings of Elon Musk might take about 15 minutes. There usually wasn't that much to say other than his statement, after all, but that one article could drive 100,000 page

views or more. That was the secret treasure that funded the real in-depth reporting. cheap coverage of a big tech company coupled with the lucre of comparatively extraordinary ad revenue. For the business side, TechCrunch's focus on startups required second-order thinking. Startup-related articles got a fraction of the readership of an article on Apple, since no one is searching on Google for the name of a startup they have never heard of before. So why bother? Indeed, many of TechCrunch's now-dead competitors didn't bother. The key insight, though, is that these articles attract the startup CEOs and founders,

and founders, and it is precisely this demographic that is so valuable for advertisers. Startup coverage was a form of service journalism and one that happened to create a perpetual revenue machine. The other half of TechCrunch's business is events, and namely Disrupt. Disrupt attracts around 10,000 attendees per year, and the fruits of that service journalism on startups kept on giving. Disrupt offered founder pass packages that were quite affordable even for the youngest companies, while charging eye-watering sums for business executives from legacy technology companies. the economic price discrimination is and was brilliant. Make sure the cool kids are there and then charge the so-called grownups to be around them. TechCrunch's events were more profitable

compared to the industry norm, since most of them were local to San Francisco. None of the speakers were paid. panels could be constructed by accomplished beatwriters who already knew who should be there, greatly reducing the number of event planners required, and the site itself became the best advertising medium to sell tickets and sponsorships vastly reducing marketing costs. The unique economics for TechCrunch around advertising and events funded the organization well, but they have an obvious flaw. They don't really scale. There isn't an infinite universe of big tech companies or venture-backed bubble companies willing to spend lavish sums on ad space. As for events, they rarely get better with ever more attendees, and it's hard to replicate

the scarce thrill of a flagship event multiple times per year. If TechCrunch's business leverage was graphed as a parabola, it was holding steady at the optimal maximum for years, not growing growing, not shrinking, but as sustainable as digital media companies can be this century. Extra Crunch as a subscription offering was meant to be the stabilizing third leg of the revenue tripod. Ads and events are both heavily influenced by global macro factors, but subscription offered a route to more predictable long-term revenues. When we were conceiving the new product in late 2017 and early 2018, the idea was simple, offer compelling analysis of successful startups from both business and product lenses. That's where the idea of an EC1,

came from, named for the SEC Form S1 filings for IPOs, TechCrunch would cover the founding stories of companies, but also the intricacies of their revenue models and unique product eccentricities, such as how Patreon handles creator relations or how StockX built an e-commerce authentication team by hiring sneakerheads who obsessively knew every detail of every product on the marketplace. The most important challenge of modern media is balancing an audience's desire for certain types of stories with a human reporter's ability to deliver them. Unlike a tech company building an app or a cloud service, this is not an easy product to iterate. If you want to improve coverage of the automotive industry, an editor must seek out and develop a reporter

who loves cars and understands how they get manufactured. What points of competition exists between companies? What auto economics are and how they are changing and what disruption might look like for an industry in the years ahead. Passion plus perspective plus precision is asking a lot of one person or even a small band of reporters. Even if you can find that talent, then the challenge becomes one. of compensation. If someone understands the venture industry well enough, for example, then they can almost certainly get a job at a VC firm and make a multiple of their media salary. Reporting on cloud infrastructure, they can triple their salary working at Amazon Web Services

without the daily doom of media layoffs looming over their overworked typing hands. TechCrunch Plus eventually succumbed to its middling status, essential for the health of the business, but unable to grow enough. Just enough for writers to keep its editorial calendar sustained with analysis, but never enough to allow the editorial to truly flourish, end quote. If you're anything like me, then you spent all weekend looking at user-uploaded demos of using the Apple Vision Pro in real life on social media all weekend. If you're not like me, and you didn't do that, the final bottom link in the show notes today is to a quick YouTube video I threw together, showing you my favorite user demos that I saw this weekend. Talk to you tomorrow.