Tech Brew Ride Home - Omnibus 06/09

Episode Date: June 14, 2025

Omnibus Week Of 06/09 Sign up for the ad-free feed at tech.supercast.tech 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. This is the Omnibus episode for the week of June 9th, 2025. So as I do every single year, I watched Apple's WWDC keynote so you don't have to. Here's what happened.

The gag opening video was Craig Federigi driving an F1 car around a track that was the roof of Apple's world headquarters. It was fine, obviously tied into that F1 movie with Brad Pitt that is apparently an Apple original coming out soon. after Tim Cook said hi, it was Craig that was first up, and in fact, was leading most of this keynote talking Apple Intelligence. He spoke about what they've already released the writing tools, email summary, Gen Moji, etc. When they go down the list like that, it kind of feels like, yeah, they've already released a lot of AI stuff, but then you remember, I basically never use slash notice that stuff. But then Craig said that the big thing they're doing this year is what they're calling foundation models, framework. Apple is opening up the foundational AI models that power Apple intelligence for developers to use in their apps. Finally, the models run locally on device, apparently, quote,

we couldn't be more excited about how developers can build on Apple intelligence, Craig said. But that was it in terms of talking about AI specifically, as opposed to sprinkling AI in, among other innovations. Almost immediately, it was on to the new OS visual redesigns. They say this is the biggest jump in design language since iOS 7. The key thing is that the design language is moving across all their platforms, Mac, the watch, etc. It's all going to look the same. They're calling it liquid glass, and it looks okay. I'm not a designer. It looks nice, but to my eyes, it's basically exactly what I'm used to. Again, to my dumb eyes, it looks more rounded corners, looks like translucent bubbles or, I guess, glass. That's the point. Although, as I see,

see more examples. Is it a little too transparent? I don't know. Need to see it with my own eyes. Fine. New design language. They're so excited about it, said Craig. And yes, as rumored, iOS 26, MacOS 26, et cetera. Version numbers are now named for the years. So, starting with iOS 26, the lock screen is updated, liquid glass. Again, I feel like this is maybe too transparent. The wallpaper and the lock screen is more dynamic. They're also using the neural engine to make your photos 3D on the home screen. They react to movement. It looked pretty good on the video, I have to say. The camera app has done something I've wanted for a while. In iOS 26, your immediate options when you bring up the camera app are just photo and video. If you want the other stuff like

Panorama and cinematic, it's a tap or swipe away. There's better search and photos, a new unified layer for the phone app. Also, call screening on the phone app, which can automatically answer calls from unknown numbers. An AI agent answers for you. Once they identify themselves the caller, then Apple will actually put the call through, but not until then. And then, this sounds great as well, something called hold assist. If you get put on hold when you're making a call on the music starts playing the muzac. You can basically hang up, and your iPhone will call you back when there's a live person to actually talk to. Apple leaned heavily into new features for group Chats taking a direct shot at WhatsApp. There's new polls on group chats and the ability to send

Apple Cash inside group chats, but there's also screening tools in messages, so you can also screen those spam messages you're getting. Unknown senders are now shunted into a dedicated folder that you have to check separately. Yes, pixel phone users, I know you've had this for a while. I can hear you typing out the tweets to me as I record this. And then some AI Gen Moji and emoji stuff, whatever. translation is now in FaceTime messages and the phone app. The translations run locally on device. Now, this might be something people actually notice AI stuff that people will actually use now and again. Each text can be instantly translated. FaceTime will have live subtitling. Developers can enable live translation in their apps with an API. Seriously, remember what I said a few minutes ago

about how people haven't really noticed the AI stuff on their iPhones yet? If this works as well as demoed it, people will notice this. There was a brief mention for Apple Music, music translation there as well, lyric translation. Then on to Maps. Maps will now learn your preferred routes and offer them up to you in Maps instead of having to choose from the different options for your route every time. Remember the golden age of foursquare slash swarm. They now have visited places inside Maps to keep track of where you've been in the Maps library or search. Wallet has announced 20 different car brands offering car keys to open your car and get going, unlocking with 13 more coming soon.

Digital ID is rolling out in wallet. It's your passport, but only for domestic travel. Apple Games got its own special section, a new games app. Goodbye and Good Rins to Game Center. The new games app does have a dedicated tab for Apple Arcade, but also a library tab for what you've downloaded previously. There's a Play Together tab to see what your friends are. are playing. The games app works with external controllers. Something called visual intelligence

lets you use AI to analyze photos or screenshots to query AI about them. Again, stuff Android has had for a while. I want to note that when they mention AI stuff, they keep mentioning that you can query chat GPT, but no word on other partners. Like, I don't know, Gemini or Anthropic built in as AI partners. It's interesting that we're two years into this AI stuff and they've still not added other AI partners. WatchOS. New workout stuff like Workout Buddy, which tracks your recent workouts and then paces you, I guess, shouts at you to run faster. I don't know. They called it encouraging you like a trainer, but it looked to me like Workout Buddy shames you passive aggressively. I'm sorry, did I say shame I meant? Encourages you. This seems like something I would immediately

turn off, sort of like the stand-up notifications. Hey, taking a minute to point out that as we heard, they're at least at this point only talking about AI stuff they know they're going to release this fall. But has anyone else noticed they're not mentioning the AI stuff from last year that they still haven't shipped? Just pointing that out. Ooh, you can now do a sort of wrist flick thing on your watch to immediately clear messages and alerts. That's useful. But again, what are we doing here? We're an hour in. Notes is coming to Apple Watch. It wasn't already there. I'm not even going to bother with the TVOS stuff. There's a new karaoke mode that works with Apple music on your Apple TV or TVOS to turn your iPhone into a mic, and then you can sing along

like it's a karaoke machine. Then it was on to MacOS. It's MacOS Tahoe this year, as rumored, new design, the glass design. They basically pitched it as everything we told you about on the new features and stuff from iOS. Yeah, you're getting them on your Mac 2. Why are they convinced that this new transparent glass design is a major step forward? I still don't know. You can change the color or design of your folders on macOS, personalized wallpaper and theme colors, live activities are coming to Mac, phone app is coming to the Mac, fully synced to your iPhone. Basically, this is making your Mac as live interactive as your phone is. Spotlight can launch apps from your iPhone in iPhone mirroring. Spotlight can also take actions in apps. You can send

emails from Spotlight, for example. They say this is the biggest update to Spotlight on the Mac ever. Actually, they went heavy on Spotlight combined with shortcuts. There's a lot of things here that remind me of using Superhuman for your email, but other people are saying this is like Raycast. Basically, things like you can type the keys SM, SM, like a shortcut into Spotlight, to send a message without having to click over to messages. Let me quote Nilai Patel from the Verge here. Spotlight turning into an AI-powered command line of sorts is really fascinating,

and also wild that it's totally disconnected from Siri. Like if you think AI represents a platform shift, then a magic command line that accepts natural language prompts across everything you're using on a computer is pretty huge, actually. End quote. Then stuff about Mac gaming. But not today, Tim Apple. I've been burned about taking Apple seriously,

about taking gaming seriously on the Mac too many times. Apple Vision Pro got a software update. They're moving from Vision OS 2 to Vision OS 26. It's got widgets, weather, clock, photos, widgets. Like, if you still use your Vision Pro at all, you could say put a clock on the wall above you and it gets this, stays there in place on the wall. If you're looking through your Vision Pro,

which again requires you to still be using your Vision Pro, big changes to personas. They look way, way more lifelike by pulling from your photo album. But again, that would require you to, you know, still be using your Vision Pro, but also, crucially, know a single other person in your life that owns a Vision Pro. Wait, I forgot that we never got to iPadOS, at least at this point. The big news here is that they have a new multitasking and get this, I'm sorry, Steve Jobs rolling over in your grave, a windowing system for the iPad. This is basically just turning the iPad into a Mac, which is that a

bad thing? So for the windowing, you can pull from the bottom right of any app to resell. size apps. They become Windows. Apps remember their size and position. If you close and minimize them, y'all, if you've always wanted your iPad to look like the pile of Windows on your Mac desktop, the future is here. It even has expose to see all of your windows. It has a friggin' menu bar. Files. The Files app has basically been blown up to manage and access files across apps. There's a list view. There's folders. The same customization options on Mac OS, you can choose which app to open a file with people. You can have other apps running in the background while you're in another app. So, why is the iPad not just a Mac yet? I feel like you could get to the point where you could

get your iPad to like 90% of how you use your Mac right now, so why not just put MacOS on iPads? We always word that Apple would turn the Mac into an iPad, but it looks like they've thrown in the towel and now are going in the other direction. What else? Apple claims you can now record studio quality audio from AirPods, we shall see things not mentioned at this point as we wrap up the App Store, even once? Siri? Even once? I don't remember.

And that was that. About 90 minutes, chock full of things, I guess. And yet, if you were to meet me on the street in about an hour after I posted this and you said to me, Brian, what was the big news? The big theme from WWDC today? My answer would be iPad is Mac now? Other than that, I honestly can't remember. And I literally just type this stuff up.

Microsoft and Asis have unveiled two new Raj Xbox ally handheld devices featuring a redesigned full-screen Xbox experience built specifically for portable gaming. This is a major step in Microsoft's broader effort to optimize Windows for handhelds and compete directly with Valve's SteamOS, which we have spoken of previously. The company has overhauled the Xbox app, game bar, and underlying Windows processes to improve performance, reduce idle power consumption, and streamline gameplay. When users boot up these devices, They'll enter a console-like Xbox interface that hides the traditional Windows desktop. The game bar now includes quick access to system settings, ASIS's command center, and Microsoft's

gaming co-pilot. Players can also switch between apps with a newly handheld-friendly task switcher using the Xbox button. Microsoft has trimmed unnecessary Windows processes, freeing up memory and extending battery life. The updated Xbox app aggregates games from Xbox GamePass, Steam, Epic Game Store, and more. A new verification system similar to Steam, decks will identify handheld optimized games. This experience debuts on the new Xbox ally models and will expand other devices next year. With SteamOS also supported on the Rog Ally, Microsoft's changes, signal an intensifying battle between Windows and Linux for the future of handheld PC gaming. But what's it like hands-on? Well, quoting IGN. So let's get the specs out of the way. The Xbox

Ally X is the top-of-the-line model sporting the AMD-Risen AI Z-2 Extreme APU, 24 gigabytes of RAM and a 1-terabyte SSD powered by an 80-watt-hour battery. The base-level Xbox Ally rocks the Ryzen Z2A chip, 16 gigabytes of RAM and 512 gigabyte SSD with a 60-watt hour battery. Both use the same 7-inch screen, which is a 1080p, 1920-by-1080 resolution iPS display with a 120-20hertz refresh rate and is capable of variable refresh rates and free sync to reduce screen tearing when performance fluctuctions. Nothing has been shared on battery life as it's still being tested and fine-tuned. While the actual gameplay experience with the base-level Xbox Ally is still in question, since I didn't get to use it, I must say that the Xbox Ally X delivers an optimal gameplay

experience. I played Gears of War reloaded for about 15 minutes running at 1080P with high settings, AMD FSR 3.1, and quality mode, 16x Anisotropic filtering, and maintained a smooth 60 frames per second the entire time. It's worth noting that the system was running in turbo mode, which increases power to get higher performance at the expense of heat and battery life. Granted, Gears of War Reloaded is a remaster of an Xbox 360-era game, which is not to diminish what seems to be a great overhaul of the original Gears of War, but there's a difference between that and trying to run, say, doom the dark ages at maximum settings. That's also to say, we've yet to see the full potential of the Xbox ally X's Z2 Extreme, end quote.

Sources are telling Bloomberg that meta is in talks for a potential multi-billion-dollar investment in Scale AI. If they pull the trigger, this will be Meta's largest external AI investment. Scale AI was last valued at around $14 billion. Quote, scale AI, whose customers include Microsoft and OpenAI, provides data labeling services to help companies train machine learning models and has become a key beneficiary of the generative AI boom. The startup was last valued at about $14 billion in 2024 in a funding round that included backing from Meta and Microsoft. Earlier this year, Bloomberg reported that Scale was in talks for a tender offer that would value it at $25 billion. Scale co-founded in 2016 by CEO Alexander Wang has been growing quickly, the startup generated

revenue of $870 million last year, and expect sales to more than double to $2 billion in 2025, Bloomberg previously reported. Scale plays a key role in making AI data available for companies. Because AI is only as good as the data that goes into it, scale uses scads of contract workers to tidy up and tag images, text, and other data that can be used for AI. training. Scale and META share an interest in defense tech. Last week, META announced a new partnership with defense contractor and rural industries to develop products for the U.S. military, including an AI powered helmet with virtual and augmented reality features. Mata has also granted approval for

U.S. government agencies and defense contractors to use its AI models. The company is already partnering with Scale on a program called Defense Lama, a version of META's Lama large language model intended for military use. Scale has increasingly been working with the U.S. government to develop for defense purposes. Earlier this year, the startup said it won a contract with the defense department to work on AI agent technology. The company called the contract a, quote, significant milestone in military advancement, end quote. Finally today, something that got a lot of chatter over the weekend was this paper Apple researchers released detailing the limitations, according to them, of top large language models and large reasoning models like opening eyes 03, especially on classic problems that have

already been solved. Basically, the whole back and forth can be summed up by this tweet storm from Lux Capital's Josh Wolf. Quote, Apple just Gary Marcus' LLM reasoning ability. Apple tested today's reasoning AIs like Claude and Deepseek, which look smart, but when complexity arises, they collapse. Not fail gracefully, collapse completely. They found LLMs don't scale reasoning like humans do. They think more up to a point than they give up early, even when they have plenty of compute left, even when handed the exact algorithm, LLMs still botched the job. Execution does not equal understanding. It's not missing creativity.

It's failing basic logic. Models overthink easy problems, exploring wrong answers after finding the right one. And when problems get harder, they think less. Wasted compute at one end, defeatism at the other. Apple's take is these models are not reasoning. They're super expensive pattern matches. Netbreak as soon as we step outside their training distribution end quote. To which the aforementioned Gary Marcus responded with a long blog post,

mentioning how these models couldn't solve relatively simple logic puzzles like the game Hanoi. The new Apple paper adds to the force of many critiques and my own about LLMs by showing that even the latest of these newfangled reasoning models, still, even after having scaled beyond O-1, failed to reason beyond the distribution reliably on a whole bunch of classic problems, like the Tower of Hanoi. For anyone hoping that reasoning or inference time compute would get LLMs back on track and take away the pain from multiple failures at getting pure scaling to yield, something worthy of the name GPT-5, this is bad news. Look, that's why we invented computers, and for that matter, calculators to compute solutions, large, tedious problems reliably. AGI shouldn't be about perfectly replicating a human. It should, as I have often said, be about combining the best of both worlds, human adaptiveness with computational brute force,

and reliability. We don't want an AGI that fails to carry the one in basic arithmetic just because sometimes humans do. And good luck getting to alignment or safety without reliability. The vision of AGI I have always had is one that combines the strengths of humans with the strength of machines, overcoming the weaknesses of humans. I am not interested in an AGI that can't do arithmetic, and I certainly wouldn't want to entrust global infrastructure or the future of humanity to such system. Whenever people ask me why I, contrary to widespread myth, actually like AI, and think that AI, though, not Gen AI, may ultimately be of great benefit to humanity, I invariably point to the advances in science and technology we might make if we could combine the causal reasoning abilities of

our best scientists with the sheer compute power of modern digital computers. We are not going to be extract the light cone of the earth or solve physics, whatever those Altman claims even mean, with systems that can't play Tower of Hanoi on a Tower of Eight Pegs. What the Apple paper shows most fundamentally, regardless of how you define AGI, is that LLMs are no substitute for good, well-specified conventional algorithms. They also can't play chess as well as conventional algorithms, can't fold proteins like special purpose neuro-symbolic hybrids, can't run databases as well as conventional databases, etc.

In the best case, not always reached, they can write Python code, supplementing their own weaknesses with outside symbolic code, but even this is not reliable. What this means for business and society is that you can't simply drop O3 or Claude into some complex problem and expect it to work reliably. At least for the next decade, LLMs, with and without inference time reasoning, will continue to have their uses, especially for coding and brainstorming and writing. And as Rao told me in a message this morning, the fact that LLMs-L-L-R-Ms don't reliably learn any single underlying algorithm is not a complete deal-killer on their use.

I think of LRMs, basically making learning to approximate the unfolding of an algorithm over increasing inference lengths. In some context, that will be perfectly fine and others not so much. But anybody who thinks LLMs are a direct route to the sort of AGI that could fundamentally transform society for the good is kidding themselves. This does not mean that the field of neural networks is dead or that deep learning is dead. LLMs are just one form of deep learning. And maybe others, especially those that play nicer with symbols will eventually thrive. Time will tell, but this particular approach has limits that are clearer by the day. I have said it before and I will say it again.

AI is not hitting a wall, but LLMs probably are, or at least a point of diminishing returns. We need new approaches and to diversify which roads are being actively explored, end quote. Well, the contours of several recent headlines are coming together in a recognizable shape all the sudden. Sources tell the New York Times that meta plans to build an AI lab dedicated to pursuing superintelligence. led by Scale AI CEO Alexander Wang. Quote, Meta has tapped Alexander Wang, 28, the founder and chief executive of the AI startup Scale AI to join the new lab, the people said, and has been in talks to invest billions of dollars in his company as part of a deal that would also bring other scale

AI employees to the company. Meta has offered seven to nine-figure compensation packages to dozens of researchers from leading AI companies such as OpenAI and Google, with some agreeing to join, according to the people. A new lab is part of a larger reorganization of Meta's AI efforts, the people said the company which owns Facebook, Instagram, and WhatsApp has recently grappled with internal management struggles over the technology, as well as employee churn and several product releases that fell flat, two of the people said. More recently, Meta's AI division has lost employees to rival

companies, according to two people familiar with the matter. The departures were the result of a grueling pace of product development in fighting among team leaders and a tight labor market. In April, CEO Mark Zuckerberg announced two new versions of Meta's Lama AI models, which he claimed performed equal or better than comparable models from OpenAI and Google, according to testing benchmarks compiled by Meta. Soon after, outside researchers found that meta's benchmarks were designed to make one product look more sophisticated than it was. Some developers were incensed at what they saw as Meta's trickery, but not as incensed as Mr. Zuckerberg, who was upset that people thought he was trying to paper over the poor performance of his latest release.

Two of the people said, Meta is now betting that Mr. Wang will help it get back into pole position in the AI race, end quote. Indeed, Bloomberg is reporting that Zuckerberg, frustrated with Meta's shortfalls in AI, is personally assembling this team of AI experts and met engineers at his homes in recent weeks. Quote, Zuckerberg has prioritized recruiting for the secretive new team referred to internally as a super-intelligence. group, according to people familiar with his plans. He has an audacious goal in mind, these people said. In his view, META can and should outstrip other tech companies in achieving what's known as artificial general intelligence or AGI, the notion that machines can perform

as well as humans at many tasks. Once meta reaches that milestone, it could weave the capability into its suite of products, not just social media and communications platforms, but also a range of AI tools, including the meta chatbot and its AI-powered ray-band glasses. Zerkerberg aims to hire around 50 people for the new team, including a new head of AI research, almost all of whom he's recruiting personally. He's rearranged desks at the company's Menlo Park headquarters, so the new staff will sit near him, the people said, asking to remain anonymous discussing private plans. Zuckerberg is building that team in tandem with a planned multi-billion dollar investment in Scale AI, which offers data services to help companies train their models and builds custom

AI applications for businesses and governments. Scale AI founder Alexander Wang is expected to join the Superintelligence Group after a deal is done. News first reported on the deal set to become meta's largest external investment to date, spokespeople for meta and scale AI declined to comment. Zuckerberg has spoken openly about making artificial intelligence a priority for his company. In the last two months, he's gone into, quote, founder mode, according to people familiar with his work, who described an increasingly hands-on management style. The latest meta-a-i releases in April proved a disappointment to Zuckerberg, who had repeatedly

told meta-insiders he wanted the best AI offering, both in terms of utilization and performance by the end of the year. His demands piled pressure on AI-focused staff working nights and weekends to achieve those goals, according to people familiar with the matter. Yet the model's performance has been questioned both internally by MetaZone leadership and externally by developers who saw them as over-promising and under-delivering, the people said. Meta later delayed plans to release its largest model yet, known as behemoth, which it had touted as superior to competing models from OpenAI Anthropic and Google. Despite those proclamations, leadership grew concerned it didn't sufficiently advance on previous models the Wall Street Journal first reported. Those missteps

pushed Zuckerberg to get more involved and led to his interest in building out the new team, according to people familiar with the matter. He ceded a WhatsApp group chat among senior leaders called Recruiting Party to discuss potential targets for the team. Members of the chat group have been engaged in discussions at all hours of the day to identify talent. Zuckerberg has been compiling his own list of recruits and likes to be the first point of contact during outreach. The CEO hopes that with the new bench, META will see improvements to its Lama models and better AI tools for voice and personalization features, the people said. Over lunches and dinners at his California homes in the past month, Zuckerberg pitched AI researchers, infrastructure

engineers, and other entrepreneurs on joining META's team, according to people familiar with the plans. He's argued that, unlike rivals who are raising large funding rounds, META's advertising business is strong enough to finance the tens of billions of dollars needed to compete in the growing AI space. He told potential recruits that meta has enough cash flow to fund a multi-gigawatt data center, which would give the company one of the most powerful server bases in the world, according to people familiar with his pitch, end quote. Interesting. Reuters says OpenAI plans to add Google Cloud to meet its growing needs for computing capacity, quote, the deal which has been under discussion for a few months was finalized in May. One of the sources

added, it underscores how massive computing demands to train and deploy AI models are reshaping the competitive dynamics in AI, and Mark's OpenAI's latest move to diversify its compute sources beyond its major supplier Microsoft, including its high-profile Stargate Data Center project. It is a win for Google's cloud unit, which will supply additional computing capacity to Open AI's existing infrastructure for training and running its AI model sources, said, who requested anonymity to discuss private matters. The move also comes as OpenAI's chat GPT poses the biggest threat to Google's dominant search business in years, with Google executives recently saying that the AI race may not be winner-take-all.

The partnership with Google is the latest of several maneuvers made by OpenAI to reduce its dependency on Microsoft, whose Azure Cloud Service had served as the ChatGPT maker's exclusive data center infrastructure provider until January. Google and OpenAI discuss an arrangement for months, but were previously blocked from signing a deal due to OpenAIs lock-in with Microsoft, a source told Reuters. Microsoft and OpenAI are also in negotiations to revise the terms of their multi-billion dollar investment, including the future equity stake Microsoft will hold an open AI. For Google, the deal comes as the tech giant is expanding the use of its in-house chip known as tensor processing units or TPUs, which were historically reserved for internal use.

That helped Google win customers, including Big Tech player Apple, as well as startups like Anthropic and Safe Intelligence, two Open AI competitors launched by former OpenAI leaders. Google's addition of OpenAI to its customer list shows how the tech giant has capitalized on its in-house AI technology from hardware to software to accelerate the growth of its cloud business. Google Cloud, whose $43 billion of sales comprised 12% of Apple's 2024 revenue, has positioned itself as a neutral arbiter of computing resources in an effort to outflank Amazon and Microsoft as the cloud provider of choice for rising legions of AI startups, whose heavy infrastructure demands generate costly bills.

Selling computing power reduces Google's own supply of chips while bolstering capacity-constrained rivals. The open-AI deal will further complicate how Alphabet CEO Sundar Pichai allocates the capacity between the competing interests of Google's enterprise and consumer business segments. Google already lacked sufficient capacity to meet its cloud customers' demand as of the last quarter. Chief Financial Officer Annette Ashkenazi told analysts in April, end quote. Waymo has suspended downtown Los Angeles operations after at least five vehicles were destroyed during protests there. They are continuing. to operate in other parts of the city, but apparently protesters called Waymo's, as if they were

customers and then, I don't know, destroyed them, quoting the LA Times. Waymo does not think the protests were directly related to its vehicles, the company said. E-scooters operated by Lyme were also set on fire. We do not believe our vehicles were intentionally targeted, but rather happened to be present during the protests. A Waymo spokesperson told CBS News, Waymo's fleet of driverless electric jaguars have become a familiar site in Los Angeles, where they have operated since November. Waymo vehicles had driven nearly two million miles in Los Angeles as of January, but have been frequent targets for vandals.

Waymo has not indicated when it plans to resume service in downtown Los Angeles, end quote. Back to Google with more data points like this. According to similar web, Google's AI tools are significantly cutting organic search traffic to news publishers, traffic to Business Insider, Huffington Post, Washington Post are all down more than 50% in three years. Quoting the journal, Business Insider cut about 21% of its staff last month. A move CEO Barbara Peng said was aimed at helping the publication, quote, endure extreme traffic drops outside of our control.

Organic search traffic to its websites declined by 55% between April 2022 and April 2025, according to data from similar web. At a company-wide meeting earlier this year, Nicholas Thompson, chief executive of the Atlantic, said the publication, should assume traffic from Google would drop toward zero, and the company needed to evolve its business model. Google's introduction last year of AI overviews, which summarized search results at the top of the page, denned traffic to features like vacation guides and health tips, as well as to product review sites. Its U.S. rollout last month of AI mode, an effort to

compete directly with the likes of ChatGBT, BT, is expected to deliver a stronger blow. AI mode responds to user queries and a chatbot-style conversation with far fewer links. Google is shifting from being a search engine to an answer engine, Thompson said in an interview with the Wall Street Journal, we have to develop new strategies. The rapid development of click-free answers in search is, quote, a serious threat to journalism that should not be underestimated, said William Lewis, the Washington Post publisher and chief executive. Lewis is a former CEO of the journals publisher Dow Jones. The Washington Post is, quote, moving with urgency to connect with previously overlooked audiences and pursue new revenue sources and prepare for a post-search era,

he said. At the New York Times, the share of traffic coming from organic search to the paper's desktop and mobile websites slid to 36.5% in April 2025 from almost 44% three years earlier, according to similar web. When Dot Dash merged with Meredith in 2021, Google Search accounted for around 60% of the company's traffic. Today, it is about one-third. Overall, traffic is growing, thanks to efforts including newsletters and the My Recipe Locker. An Apple executive said in federal court last month that Google searches in Safari, the iPhone maker's browser, had recently fallen for the first time in two decades, end quote. A couple of drips and drabs from WWDC, Apple says MacOS 26 Tahoe will be the last MacOS release that supports Intel-based Macs, with MacOS versions

from 2026 only available for Apple's silicon models, quoting 9 to 5 Mac. Of course, Intel Max will continue to get critical security updates for some time thereafter, but users should not expect to be able to update to get new features from macOS 27 onwards, as no Intel Mac will be supported on MacOS 27. In some ways, Apple has already stopped supporting some non-Apple Silicon models of its lineup, MacOS Tahoe, does not work with any Intel MacBook Air or Mac Mini, for instance, but Tahoe does still support some Intel Macs. That includes compatibility with the 2019-16-inch MacBook Pro, the 2020-I-Mac-Mac-Mac-Bac-Bac-Bac-Bac-Based on Apple. Apple's warning, you can expect that macOS 27 will drop support for all of these legacy machines,

and therefore, macOS 26 will be the last compatible version. These devices will continue to receive security updates for another three years, however. Going forward, the minimum support hardware generations will be from 2020 onwards, as that is when Apple began the Apple Silicon transition with the M1, M1 Pro, and M1 Mac's MacBook Pros followed in 2021, end quote. Finally today, obviously my knee-jerk reaction to Apple's new liquid glass, design paradigm is a bit premature, as I've not gotten to actually use it on a device yet, but this piece in Wired suggests I'm not alone in being a little bit concerned.

Quote, after the WWDC 2025 keynote concluded on Monday, many design-focused developers Wired spoke with were impressed by the major update, but had lingering questions about how this translucent look could impact readability for users. It's hard to read. Some of it, says Alan, you, a product designer, currently building the workplace messaging app output, mainly because I think they made it too transparent. You suggest bumping up the blurring or adjusting the backgrounds to make on-screen designs more readable. Similar to the first beta for iOS 7, what we've seen so far is rough on the edges and potentially veers into distracting or challenging to read, especially

for users with visual impairments, says Josh Puckett, co-founder of iteration, which helps start-ups with designs. Still, Puckett is optimistic, based on Apple's past accessibility features that readability will improve over time. Beyond readability concerns, the first impression from some designers is that this new look could be unnecessarily distracting for users. From a technical perspective, it's a very impressive effect. I applaud the team and effort it must have taken to mimic refraction and dispersion of light to such a high degree, says Adam Whitcroft, a designer at owner.com, which makes apps and websites for restaurants. But sadly, I haven't seen a single example of where it's pulled off in a way that's complementary to the broader context it's

presented in. Whitcroft points to the dispersion and refraction of layers beneath the apps as visually distracting, especially as the user interface is changing layouts. If you've designed a UI that draws the attention of the eye away from the wider context, you've gone down the wrong path, he says. Puckett's initial reaction to the revamp is more positive than Whitcroft's. He thinks the shift away from flatness is the right design move. I'm excited that Apple is reintroducing feelings to their digital surfaces, creating interfaces that shimmer, bend, and breathe. Puckett hopes this ignites a larger design trend of more expressive experimental software. Apple is doing a great job in trying to pull us forward somewhere. It's very brave to do this. I just don't know if the direction is the right

place. He says, if anyone can do it, though, Apple can do it. I'm just scrambling to make our designs work, end quote. OpenAI has debuted O3 Pro for chat GPT Pro and team users and in its API, costing $20 per 1 million input and $80 per 1 million output tokens. Enterprise, and EDU will get access next week. Quoting TechCrunch. O3 Pro is also live in OpenAI's developer APIs as of this afternoon. In expert evaluations, reviewers consistently prefer O3 pro over O3 in every tested category and especially in key domains like science, education, programming, business, and writing help.

Open AI writes in a change log. Reviewers also rated O3 Pro consistently higher for clarity, comprehensiveness, instruction following, and accuracy. O3 Pro has access to tools, according to OpenAI, allowing it to search the web, analyze files, reason about visual inputs, use Python, personalize its responses, leveraging memory, and more. As a drawback, the model's responses typically take longer than O1 Pro to complete, according to OpenAI. O3 Pro has other limitations to temporary chats with the model and chat GPT are disabled for now,

while OpenAI resolves a technical issue, O3Pro can't generate images, and canvas, OpenAI's AI-powered workspace feature isn't supported by O3 Pro, end quote. But what is it like to use? Our friends over at Latent Space have a hands-on saying O3 Pro is much smarter than O3 and amazing at using tools. But the model requires a lot of context to run well, and without enough context, it tends to overthink. The key I discovered was to not chat with it. Instead, treat it like a report generator. Give it context, give it a goal, and let it rip.

And that's exactly how I use O3 today. But therein lies to the problem with evaluating O3 Pro. It's smarter, much smarter. But in order to see that, you need to give it a lot more context, and I'm running out of context. There was no simple test or question I could ask it that blew me away. But then I took a different approach. My co-founder Alexis and I took the time to assemble a history of all our past planning meetings at Rain Drop, all of our goals, even record voice memos, and then asked O3 Pro to come up with a plan. We were blown away. It spit out the exact kind of concrete plan and analysis. I've always wanted an LLM to create, complete with target metrics, timelines, what to prioritize, and strict instructions on what to absolutely cut. The plan O3 gave us was plausible, reasonable, but the plan O3 Pro gave us was specific and rooted enough that it actually changed how we are thinking about our future. This is hard to capture in an e-val. O3 Pro feels very, very different from Opus and Gemini 2.5 Pro, where Claude Opus feels big, but never truly showed me a clear sign of its bigness.

O3 Pro's takes are just better. It feels like a completely different playing field. OpenAI is really driving down this vertical RL path, deep research, codex, not just teaching models how to use tools, but how to reason about when to use them. The best ways to prompt reasoning models haven't changed. My guide on how to prompt 01 still stands. Context is everything. It's like feeding cookies to the cookie monster. It's a way of bootstrapping LLM memory, but actually targeted so it works well. And the system prompt really matters. Models have actually become really malleable, so LLM harnesses that teach a model about its environment and its goals have outsized impact. It's this harness, a combination of model, tools, memory, and other methods that makes AI products actually good,

what makes things like cursor just work most of the time, end quote. OpenAI also announced an 80% price drop for its O3 model and a flex mode for synchronous processing that charges $5 for input and $20 for output per million. In an essay on his personal blog, Sam Altman said, Intelligence and Energy, the limitors on human progress will be abundant in the 2030s, and superintelligence should be widely distributed. Quote, 2025 has seen the arrival of agents that can do real cognitive work. Writing computer code will never be the same. 26 will likely see the arrival of systems that can figure out novel insights.

2027 may see the arrival of robots that can do tasks in the real world. A lot more people will be able to create software. and art, but the world wants a lot more of both, and experts will probably still be better than novices as long as they embrace the new tools. Generally speaking, the ability for one person to get much more done in 2030 than they could in 2020 will be a striking change, and when many people will figure out how to benefit from. In the most important ways, the 2030s will not be wildly different. People will still love their families, express their creativity, play games, and swim in lakes. but in still very important ways the 2030s are likely going to be wildly different from any time that has come before.

We do not know how far beyond human-level intelligence we can go, but we are about to find out, end quote. Maybe you're someone who's tired of iOS news. Well, I've got something for you. Google has rolled out Android 16 to pixel phones with live lock screen notifications and debuts features including more AI editing and Google Photos in a broad update, quoting CNET. Rolling out first to most supported pixel phones, Android 16 lands months earlier than recent major versions with a focus on productivity features and enhancements. These include live updates for tracking deliveries and rideshare cars, streamline notifications, the ability to use desktop-style windows when your phones connected to a monitor, new device security features, native controls for hearing aids, and more. If you're waiting for your

sushi to arrive or watching your ride share slowly inch across 10, Android 16 now offers live updates that are pinned to your lock screen so you don't have to keep opening an app to stay in the know. Also, notifications from the same app are now automatically grouped, reducing clutter in your lock screen from notification-heavy apps like Gmail and ring. A single toggle now bundles Google's toughest defenses blocking dodgy websites, scam calls, and even USB attacks while the phone is locked. Originally built for high-risk users like journalists and politicians, advanced protection is free and available for everyone on. on Android 16. Tablet owners running Android 16 will get a desktop windowing mode later this year, built-in partnership with Samsung. It lets you float, snap, or resize multiple app windows on your device like you would on a desktop, and Android 16 lays the groundwork for that, end quote.

Nintendo says it has sold 3.5 million Switch 2 consoles globally in its first four days, a record for any of its hardware. Nintendo aims to sell 15 million units by March. quoting Bloomberg. The Japanese company has already sold more of the device than the roughly 2.7 million the original switch managed during its first month in 2017. The numbers released by the company Wednesday bode well for its target to sell 15 million units by March next year. They also reinforce analysts' projections that Nintendo may be able to sell far more if it can pump up supply. The release of the new switch was regarded as a watershed moment for the industry, steering business decisions by partners and competitors for years to come. At a time,

of thinning margins and exploding development budgets, a popular new console may galvanize the sector and provide a counterbalance to the increasing dominance of a handful of marquee live service games. There is very little margin for execution error for Ninti at this stage. The big issue right now is its high pricing, which is not a big deal in its first year when diehards want to get their hands on the console, said Amir Azurveda, a Japan equity strategist at asymmetric advisors. the pressure will be on it for it to reduce production costs to get the price down, and tariffs may complicate that, end quote. Catching up with runway demand is the first major challenge Nintendo faces.

The console is manufactured mainly in China by partners including Foxcon, exposing Nintendo to potential disruptions from a global trade war. President Shuntaro Furukawa has apologized after customers came away from lotteries for the switch to empty-handed. The Kyoto-based company has asked its partners to speak production of the console. It's also secured agreements from Japanese online marketplace operators such as recutin to discourage resellers from taking advantage of the hardware scarcity. The pace is good, said Hideki Yasuda, an analyst with Toyo's securities. The key will be to maintain assembly capacity

and increase production going forward, end quote. Elon Musk says Tesla's Robotaxy Service will tentatively begin on June 22nd in Austin, Texas, with the first autonomous deluxe. of a car to a customer's home on June 28th. Quoting CNBC, In a post on X, Musk indicated that he's flying from Los Angeles to Austin for the kickoff, which he previously said would occur sometime in June. When a commenter asked when public rides will start, Musk said the current plan is for June 22nd,

and that the first driverless trip from the Tesla factory to a customer's house will take place on his birthday, June 28th. We are being super paranoid about safety, so the date could shift. Earlier on Tuesday, Musk shared a video on X showing that Tesla was testing driverless vehicles on the roads of Austin without a human safety supervisor behind the wheel. The eight-second clip showed the latest version of the model YSUV painted black with a white robotaxy graffiti-style logo painted on it, navigating an intersection and pausing to allow pedestrians to traverse a crosswalk.

Musk recently told CNBC's David Faber that Tesla will start with a very small rollout, including about 10 to 20 of its robotaxies with a new unsupervised version of the company's FSD or full self-driving technology installed. The test will involve the Model Y, not the futuristic-looking cybercab that Tesla plans to produce next year. Musk and Tesla will geofense the service limiting where the Model Y taxis can initially operate, and that employees will remotely monitor the fleet, end quote. I'm starting to think that big things might be going on under the hood at Google. According to the information, Google has offered voluntary buyouts to U.S. staff across several

key businesses and units, including the one housing its core search team and much of the ads organization. The move comes as Google's search business faces unprecedented threats from chat GPT and following its loss in a U.S. antitrust case that could hurt the company's ability to distribute its search product. Nick Fox, who runs the knowledge group, said in an email to his team that the buyouts were aimed at U.S. employees. Google also offered buyouts on Tuesday to U.S. employees in CORE, which is the engineering team working on Google's underlying technical infrastructure, as well as research, marketing and communications divisions, a Google spokesperson said.

Earlier this year, Google also offered buyouts to the legal and finance team, the spokesperson said. All of these offers haven't been previously reported. The buyouts could be a prelude to layoffs. In January, Google offered a similar voluntary buyout to members of the 20,000 person platform and devices team. which oversees Android, Chrome, and Pixel. By April, Google laid off hundreds of employees in that unit, as the information previously reported, including 10% in its Privacy Sandbox Division, which was supposed to help usher in a more privacy-safe vision of digital advertising. Google's piecemeal cuts this year differ from its approach in 2023 when it cut 12,000 employees or 6% of staff. Google recently made a small number of cuts in its ads organization. At the end of May, Google laid off employees of its ads safety division, according to a person with knowledge of the men. matter. Earlier in May, Google laid off around 200 employees in its global business organization, which handles advertising sales and other partnerships, but sits within chief business officer Philip Schindler's team. Google's search is still Google's biggest business, contributing 56%

to Google's total revenue in the first quarter, but it's growing slower than YouTube and Google's cloud business, and those two divisions notably didn't offer voluntary buyouts to workers on Tuesday, end quote. Finally today from the interesting gadget file is this the future of video calls. HP has unveiled Dimension, the first hardware built with Google's 3D video conferencing tech, featuring a 65-inch light-field display, launching later this year for $25,000 a pop. Quoting The Verge, HP has become the first company to preview hardware built with Google Beam, the 3D video communication technology, formerly known as Project Starline.

It's launching the HP Dimension, a device that features a 65-inch light-field display with six high-speed cameras inside the bezel to create a true-to-life 3D video of your caller. Google first announced Project Starlines rebrand last month. At the time, Google said that it would let third-party manufacturers, starting with HP, use its design framework to build devices with Beam. The HP dimension is meant for enterprise use. It costs $24,99, and that doesn't even include the software. needed to hold video calls as users will need to purchase a Google Beam license separately to gain access to Zoom or Google Meet. HP and Google haven't said how much a license will cost yet. As my colleague Alex Heath pointed out in a demo of Google Beam, you don't need a headset,

you don't need glasses or any kind of special equipment to see a colleague as if they're in the same room. Bo Wilder, HP's head of future customer experiences said you don't need a special room either, but having a white background is best for an optimal experience. We're not trying to put a caricature in a small box across the table from you, Wilder said. We want you to walk into the room and instantly make eye contact without even thinking about it. HP Dimension users will still be able to hold calls with people from other video conferencing platforms and devices, but their colleagues won't be able to see their images projected in 3D and vice versa. Along with a series of built-in cameras,

the HP dimension features adaptive lighting that adjusts to the environment, allowing users to see realistic shadows on facial features and natural skin tones. The HP Dimension pairs its light field display with spatial audio that Wilder said never separates the voice from the body. It comes with HP's new Poly Studio A2 table mics, as well as four speakers located behind its curved, acoustically transparent midwall for a direct path to the ears. The ultimate goal of Google Beam and it's manifested on HP Dimension is to feel like you're there. Andrew Narker, general manager of Google Beam said during the briefing, you feel just like you're there at the table working together. it's all meant to bring us together and ultimately feel like we're completely physically present. The HP dimension will launch in the U.S., Canada, the UK, France, Germany, and Japan later this year.

Companies like Salesforce, Deloitte, and NEC Corporation have already committed to bringing Google Beam into their offices, end quote. Gotta catch you up on something that sort of fell through the cracks because of when it happened. Meta has apparently formally agreed to take a 49% stake in scale AI for $14.8 billion, and the deal that gives cash to scales shareholders and makes Alexander Wang a top meta executive. So sort of an acchia hire, right? Quote, meta would put Wang in charge of a new superintelligence lab along with other top-scale technical employees. The New York Times and Bloomberg reported Tuesday morning that will put Wang 28 in competition

with some of his customers and friends, including OpenAI CEO Sam Altman. The deal likely would further enrich Wang, who became the youngest self-made billionaire in the U.S. several years ago. The deal which hasn't been finalized appears to be a rich one for Scales shareholders with big paydays for some of Scales' biggest investors such as a sell, index ventures, founders fund, and Green Oaks, as well as current and former employees. Scale shareholders also would maintain their existing holdings in scale, which would now be valued at $28 billion, including the cash invested up from $13.8 billion last year, end quote. But back to the Accu-hire angle of this. The information also

reports that this deal was motivated by Mark Zuckerberg's desire to find new leadership for meta's AI efforts and his personal relationship with Alexander Wang. Quote, Meta Platform's CEO Mark Zuckerberg has tried to fix a seemingly existential problem in recent months. His company's large language model Lama was falling behind competitors and struggling to perform as well as models from Open AI, particularly on complex tasks. Increasingly, Zuckerberg turned to an unlikely person for advice. Alexander Wang, CEO of Scale AI, whose startup did the relatively low-grade work of hiring human experts to improve artificial intelligence models for an array of customers, including meta. Wang was essentially a business guy with technical chops, not the kind of

elite researchers Zuckerberg recently had been lavishing with attention and offers of gigantic paydays. And Scale AI's humble work sometimes made it a punching bag among Silicon Valley insiders who didn't put it in the same class as cutting-edge AI startups. But Wang, 28, provided useful advice, and Zuckerberg began to reference direct feedback from Wang about potential solutions to meta's AI problems in meetings with other advisors. Zuckerberg felt Wang had good perspective from working with other research labs. That work means Wang knows what data those labs are interested in and how those labs might be trying to improve their own models. Now Wang's kinship with Zuckerberg has thrust Wang into a seat at the table of a tech giant with one of the largest piles of money

to devote to AI. The deal, which is expected to be announced as early as Thursday, may still have to pass scrutiny from regulators despite the fact that META is taking a minority stake in scale. Some officials in Donald Trump's administration were upset by reports about the deal and have privately called it an issue of national security due to scale's contracts with the Department of Defense, according to a person who spoke with them. The scale deal, which values the startup at about $28 billion, is the second largest investment META has ever made in another company after its $22 billion purchase of WhatsApp. It comes since META's cash pile pushed past $70 billion, even as it spends heavily on new data centers.

But Zuckerberg's decision to do the deal is occurring despite complaints from leaders within Meta's generative AI group about Scale AI over the past year. Like other Scale AI customers, Meta relies on the company's experts who often have doctorates in fields like math and biology to annotate data and write ideal responses that it then uses to improve how AI models perform specific tasks. But Meta's AI leaders complained that the startup's data labeling contractors often return low-quality data. They were also upset about Meta's generative AI group exceeding its budget with scale, which led to the heads of

some other teams at Meta to ask their staff not to spend money with scale. The complaints about Scale AI weren't enough to keep Meta's leaders from pursuing a deal, though. In doing so, Zuckerberg mostly wanted to find new leadership for Meta's AI efforts, he told people, even after overhauling the structure of its generative AI group in February and again in May. Zuckerberg had spent time with other prospects for jump-starting Meta's AI efforts since last year, In recent months, he tried to recruit Karekowiczkoglu, Google's chief AI scientist. He also reached out to Mira Muradi, OpenAI's former chief technology officer, who now runs her own AI startup. In the end, Zuckerberg came back to Wang. It helped that Wang got along well with Meta's chief product

officer Chris Cox and other senior leaders at the company. Despite the fact that scale AI didn't work on the most sophisticated part of developing AI models, Zuckerberg saw him as having a deep understanding of where the industry was heading, and Wang had developed a reputation as a credible salesperson, building scale AI's revenue to about $870 million last year. Zuckerberg and other meta leaders aren't the only tech luminaries Wang has impressed. Wang has spent years developing mentor relationships with an older generation of tech leaders, including ex-Google CEO Eric Schmidt. He even shared an apartment with OpenAI CEO Sam Altman.

He also became an occasional confidant of Stripe CEO Patrick Collison, who now sits on Meta's board of directors. In an interview last year, Collison said Wang, quote, has described to me on several occasions a somewhat non-consensus perspective on what's going to happen in some part of the industry, end quote. The piece is much longer, going in depth into the infighting and problems inside meta's AI initiatives. Again, the word is that this deal will formally be announced today, perhaps, so consider this all a placeholder for that if it does indeed come to pass. Disney and NBC Universal are suing Mid-Jurney in California, accusing it of direct and secondary copyright infringement. They say talks with Mid-Journey failed to resolve their concerns, quoting Axios.

It's the first legal action that major Hollywood studios have taken against a generative AI company. The complaint filed in a U.S. District Court in Central California accuses Mid-Journey of both direct and secondary copyright infringement by using the studio's intellectual property to train their large language model and by displaying AI-generated images of their copyrighted characters. The filing shows dozens of visual examples that it claims show how Mid Journey's image generation tool produces replicas of their copyright-protected characters such as NBCU's minions characters and Disney characters from movies such as The Lion King and Aladdin. Disney and NBCU claim that they tried to talk to Mid-Journey about the issue before taking legal action, but unlike other generative AI platforms that they say agreed to implement measures to stop the

theft of their IP, Mid-Journey did not take the issue seriously. Mid Journey, quote, continued to release new versions of its image service, which, according to Mid Journey's founder and CEO, have even higher quality infringing images, the complaint reads. Mid Journey, quote, is focused on its own bottom line and ignored plaintiff's demands, it continues. It's notable that Disney and NBCU, which own two of the largest Hollywood IP libraries, have teamed up to sue Mid Journey. The lawsuit suggests Hollywood Heavywates will try to focus their copyright fight on platforms that create and distribute replicas of their copyrighted content, rather than the users. of those platforms, and quote. The Wikimedia Foundation has paused an experiment that showed Wikipedia users AI-generated summaries at the top of some articles following an editor backlash, quoting 404 Media.

Just because Google has rolled out its AI summaries doesn't mean we need to one-up them. I sincerely beg you not to test this on mobile or anywhere else, one editor said in response to Wikimedia Foundation's announcement that it will launch a two-week trial of the summaries on the mobile version of Wikipedia. This would do immediate and irreversible harm to our readers and to our reputation as a decently trustworthy and serious source. Wikipedia has in some ways become a byword for sober boringness, which is excellent. Let's not insult our reader's intelligence and join the stampede to roll out flashy AI summaries, which is what these are, although here the word machine generated is used instead, end quote. Two other editors simply commented, yuck.

A page detailing the AI-generated summaries project called Simple Article Summaries explains that it was proposed after a discussion at Wikimedia's 2024 conference, Wikimania, where, quote, Wikimedians discussed ways that AI slash machine-generated remixing of the already created content can be used to make Wikipedia more accessible and easier to learn from. Editors who participated in the discussion thought that these summaries could improve the learning experience on Wikipedia, where some article summaries can be quite dense and filled with technical jargon, but that AI features needed to be clearly labeled as such, and that users needed an easy way to flag issues with machine-generated remixed content once it was published or generated automatically. In one experiment where summaries were enabled for users who have the Wikipedia browser extension installed, the generated summary showed up at the top of the article which users had to click to expand to read.

That summary was also flagged with a yellow unverified label. Wikimedia announced that it was going to run the generated summaries experiment on June 2nd and was immediately met with dozens of replies from editors who said, very bad idea, strongest possible oppose, absolutely not, etc. A day later, Wikimedia announced that it would pause the launch of the experiment, but indicated that it's still interested in AI generated summaries, end quote. I cannot express enough how fascinating it is to see stable coins go mainstream in the world of finance. If you look at this through the original lens of Satoshi's stated desire for crypto to take over traditional finance and money,

this is that coming true in a way. Example number whatever. According to Visa, stable coin transaction volumes hit $752 billion in May up from $409 billion in May of last year, and wallets that regularly send and receive payments hit a record of 46 million. Quoting the FT. Investment Bank Standard Chartered has forecast there could be some $2 trillion of stable coins in circulation by the end of 2008 from around $250 billion now. Others are even more bullish.

Payments, giants, Stripe, and Visa are deepening their investments in the industry. Japan's Sony Bank is testing its own token for payments, and there is speculation that banks and big Silicon Valley technology, companies will also join the fray. Uber, the ride-hailing app, is considering using them to make cross-border payments and reduce currency costs. Capturing the mood among investors, the market valuation of Circle, which runs the world's second largest stablecoin, nearly quadrupled in value in its first three days as a listed company in New York to $25 billion. But stablecoins exist in a gray area somewhere between a payments network, a bank deposit, and a security. Issuers have liabilities like a bank,

but don't make commercial loans. They are tradable and invest in assets like money market funds, but U.S. regulators have ruled that they are not securities if the coins can be fully redeemed on demand and do not pass the investment income to holders. A stable coin is supposed to keep a fixed value against the asset to which it is pegged, but often deviates from it by more than a couple of a percent. Few see an immediate need for stable coins in countries with relatively advanced banking and payment systems where credit and debit cards can be used to help pay for daily expenses, Most of the traction is focused on cross-border payments where the system is most broken, says Michael Shalov, chief executive of fireblocks, an infrastructure provider for digital assets.

You're shortening settlement from three days to ten seconds. It's going to go into every payment system, end quote. But large-scale use may take some time, even though issuers are subject to anti-money laundering laws, stable coins remain the principal crypto asset for illicit transactions. A 2024 U.N. report named the Tethercoin as the preferred choice for Asian crime syndicates. Blockchain analysis company chain alias estimated that criminal activity associated with cryptocurrencies hit $51 billion last year with stable coins accounting for 63% of that, end quote. Finally today, I know my usual phrase is blowing up the cable bundle just to reconstitute it, but what this next segment presupposes is, what if we blew up the TV commercial just to resuscitate it?

sources and documents reveal that Amazon Prime Video's ad load has increased to four to six minutes per hour, up from two to three and a half minutes when ads were first introduced in January of 2024. That's still not as bad as the old must-see TV ad load, but still give them time. Quoting Ad Week. According to six ad buyers and documents reviewed by Adweek, the current ad load on Prime Video now ranges from four to six minutes per hour. And while that could bring down CPMs, cost per millo, that's how ads are sold. buyers will be watching whether this impacts user experience. The increase, which Amazon had telegraphed to investors, but has not publicly acknowledged to consumers, gives the company significantly more

inventory to sell across its rapidly expanding streaming business. They told us the ad load would be increasing, said Kendra Tang, programmatic supervisor at Rain the Growth Agency. That's been confirmed recently when we noticed more avails in the system, she said. The uptick in commercials is the latest sign of maturation from the streaming service, which in recent months has debuted a slew of new products designed to make it more appealing to marketers. Its show-level data, private auction deals, and forthcoming contextual offerings have all sought to separate Prime Video from its competitive set. When Prime Video launched its ad tier, Amazon made commercials the default for all Prime subscribers, prompting backlash

from some consumers, but giving the company an immediate footprint of over 150 million monthly ad-supported viewers. Still, to ease that transition, Amazon kept its ad-load light. They had to make the ad load palatable, said Doug Palladino of PMG. But by late 2024, Amazon had already told investors it would, quote, ramp up the volume in 2025. According to buyers, the shift reflects a broader effort to right-sized Prime Video's inventory relative to the rest of the market. This is a lot of them coming back to equilibrium, Paladino said. They have more subscribers than any other ad-supported streamer, but they weren't watching enough for that to matter. More ad load helps bring that back into balance, end quote. And it's new. And it's new.

level, Prime Video's ad load begins to mirror broader industry standards. Netflix still offers the lightest ad experience while services like Hulu, Tubi, and Paramount Plus carry heavier ad loads. Prime Video is now firmly in that middle tier, Palladino said, noting its ad volume now matches other premium platforms. Critically, streamers' ad load generally still pales in comparison to what viewers experience on linear television, which typically ranges from 13 to 16 minutes per hour. By increasing its ad load, Amazon has created more inventory to sell. More inventory to sell. More inventory typically leads to lower CPMs, and while buyers haven't yet seen major drops, they expect them to come soon. Currently, buyers said that Amazon Prime Video CPMs fall somewhere in the middle compared

with other streamers. That's the upside here, said David Nirenberg, senior vice president of Digital at Intermedia Advertising. A biddable environment plus greater supply should allow buyers to find impressions at more efficient rates. It's a good thing if they can scale without degrading user experience. If Prime Video drops 10 to 20% in price, it could move speed. their way, Palladino said. We're in a tariff-heavy retail environment where clients need to do more with less. A cheaper premium platform helps. The increased inventory should also make it easier for ad buyers to target niche categories at scale, according to Nirenberg. That would help advertisers take further advantage of the granular data capabilities it offers through its demand side platform,

end quote. Something something. AI chatbots are the new vector for getting at or exposing your data. TechCrunch points out that the public public feed of the meta-AI app is filled with private and sensitive information, suggesting that a lot of users might not be aware they are sharing their chats publicly. Quote, it sounds like the start of a 21st century horror film. Your browser history has been public all along, and you had no idea. That's basically what it feels like right now on the new standalone meta-AI app where swathes of people are publishing their ostensibly private conversations with the chatbot.

When you ask the AI a question, you have the option of hitting a share button, which then directs you to a screen, showing a preview of the post, which you can then publish. But some users appear blissfully unaware that they are sharing these text conversations, audio clips, and images publicly with the world. When I woke up this morning, I did not expect to hear an audio recording of a man in a Southern accent asking, Hey, hey, meta, do some farts stink more than other farts? Flatulence-related inquiries are the least of meta's problems on the meta-a-a-a-a-a-a-a-a-a-a-a-a-a. AI app I have seen people ask for help with tax evasion if their family members would be arrested for their proximity to white-collar crimes, or how to write a character reference letter for an

employee facing legal troubles with that person's first and last name included. Others, like security expert Rachel Toback found examples of people's home addresses and sensitive court details among other private information. When reached by TechCrunch, a meta-spokesperson did not comment on the record. Whether you admit to committing a crime or having a weird rash, this is a privacy nightmare. Meta does not indicate to users what their privacy settings are as they post or where they are even posting to. So if you log into Meta AI with Instagram and your Instagram account is public, then so too are your searches about how to meet big booty women. Much of this could have been avoided if Meta didn't ship an app with the bunker's idea that people would want to see each other's conversations with MetaI, or if anyone at Meta could have foreseen that this kind of feature would be problematic. There's a reason why Google has never tried to turn its search engine into a social media feed

or why AOL's publication of synonymized user's searches in 2006 went so badly. It's a recipe for disaster. According to app figures and app intelligence firm, the meta AI app has only been downloaded 6.5 million times since it debuted on April 29th, end quote. So it is official meta yesterday announced the $14.3 billion investment in scale AI and the hiring of CEO Alexander Wang to help oversee its AI efforts. A source says, Scale AI's post-money valuation is now $29 billion. Alexander Wang is departing day to day at Scale AI as CEO to join Meta

and named Chief Strategy Officer Jason Drogey as interim CEO. Wang will remain a board director at scale. Bloomberg says that Meta's scale AI deal dwarfs rivals, quote, reverse aqua hires in scope and ambition. showing its fear of losing in AI outweighs any sort of concerns over regulatory backlash. Quote, Meta's massive gambit for scale AI shows just how far it will go to avoid falling behind in the AI arms race. Despite watching Microsoft, Amazon, and Google face months of regulatory probes for similar deals, charging ahead at a time when it's already in a lawsuit with the FTC, which claims it is a monopoly,

shows how Meta's worries over seeding ground to AI rivals outweigh the fear of regulatory backlash. Meta could have done this in a quieter way, maybe take a smaller stake, said Daniel Comeham, an associate professor of management at Columbia Business School, this deal reflects meta's impatience and fear that they're falling behind, end quote. Combe said the strategy is indicative that AI products have hit a plateau and that tech giants need to look beyond their own labs to gain an edge. A lot of products have reached a ceiling, and they all look more or less the same, he said. So with this in mind, is it worth the risk? end quote. Mark Gurman says Apple has set an internal release target of spring 2026 for that delayed

upgrade of Siri as a part of an iOS 26.4 software update, quoting Bloomberg. The long promise changes will allow Siri to tap into customers' personal data and on-screen activities in order to better fulfill queries. Apple's dot-four updates known as E on the company's internal software development schedule are typically released in March. That was the case with iOS 8.5.5. 18.4 this year and iOS 17.4 in 2024. But an exact date hasn't been set internally for the software beyond a spring time frame said the people who asked not to be identified because the work is private. The technology in the works also includes a system called app intents that allows Siri to more precisely control applications and in-app actions across Apple devices. If the latest release timing

sticks, Apple will have gone nearly two years between announcing the new Siri and delivering it to consumers. It's been an especially high-profile delay because the capabilities were part of the iPhone 16 marketing last year, despite the new Siri not being close to ready. Internally, Apple's AI and marketing teams have pointed fingers at each other. The engineering side has blamed marketing for overhyping features, while marketing maintains it operated on timelines provided to them by the company's AI teams, according to people with knowledge of the matter. There also remains a debate over how much AI functionality Apple should be building itself and how much it should push off to partners like OpenAI. And the company has held internal discussions about buying smaller AI-related startups, end quote.

So maybe another acu-hire like the scale AI one coming down the road. Chime shares jumped 37% in their NASDAQ debut on Thursday, closing at $37.11 and giving the company a market cap of $13.5 billion. Quoting CNBC, Chime's IPO from a valuation perspective represents a big step down from where venture investors like Sequoia Capital valued the company in its last fundraising round in 2021 when private tech markets were raging. The valuation at that time was $25 billion. Still, Chimes' offering is the latest sign that the FinTech IPO market is opening up after a multi-year freeze brought on by rising interest rates and valuation resets. Recent debuts from E. Toro and Crypto Company Circle have rekindled optimism in the sector with both stocks seeing strong initial pops. Chime reported

$518.7 million in revenue for the most recent quarter, a 32% increase from a year earlier. Net income narrowed slightly to $12.9 million down from $15.9 million in the same period last year. CEO Chris Britt said Chime has built a loyal user base by serving Americans earning $100,000 a year or less, a group often overlooked by traditional banks. Two-thirds of our customer base use us as their direct deposit account and primary account relationship, Britt told CNBC's David Faber. We help our members avoid fees, get access to short-term liquidity, build their credit, and build their savings, and it's that combination of services that really resonates and matters most to the everyday consumer. The average chime customer completes more than 55 transactions per month

using the chime card and interacts with the app four to five times a day. Active member growth rose 23% in the first quarter from a year earlier, Britt said, with 8.6 million monthly active users and an increasing number turning to chime to serve as their primary banking relationship. Customer acquisition doesn't come cheap, chime disclosed in its prospectus that it spent $1.4 billion on marketing between 2022 and 2024. Brits said the retention rate is above 90% once users set up direct deposit, end quote. By the way, if you follow my socials, you'll know that just for fun, I've started keeping a running IPO meter where zero is the IPO markets are closed, and 10 is, woo-hoo, we're partying like it's 1999 again. In other words, the meter keeps track of

my sense of how open the public markets are for IPOs from tech companies. Check my pinned tweet on my account at Brian MCC. I think I'll keep the meter where it is right now, roughly 3.6 on that scale to 10. But if Chimes shares rise over the next few weeks, maybe I'll nudge it up a bit. Sometimes the news has a tendency to pile on things I say almost immediately. What did I say this week about stablecoins taking over the world? Well, Shopify has announced it's partnered with Coinbase to enable USDC payments for its merchants through Coinbase's Base Layer 2 network, starting with a limited group of merchants. But, well, you don't get much more trad-fi than this. The journal also says that Walmart, Amazon Expedia, and several large airlines have recently explored issuing or using

their own stable coins in the U.S. to potentially save fees from cash and card transactions. Quote, a move to launch a payment system by Walmart or Amazon that bypasses the traditional payment system would send shivers through the nation's banks. With vast networks of customers and employees, troves of data and far lighter regulations, retail and technology companies have long been viewed as particular threats to banks including regional and community lenders. The retailer's final decisions could depend on a bill called the Genius Act, which would begin to establish a regulatory framework for stablecoins. The bill recently passed another procedural hurdle, but still needs to clear the Senate and the House.

Stablecoins could allow merchants to circumvent traditional payment rails, which cost them billions of dollars in fees each year, including the interchange fee they pay when customers make purchases using their cards. Payments can take days to settle, delaying the time it takes for merchants To receive the proceeds from sales, stablecoins offer the possibility for a quicker process. They could be of particular interest to merchants with suppliers who are located abroad. Merchants have long tried to launch payment alternatives to get around the card-based system that is dominated by Visa and MasterCard, though most of those have failed to gain traction. Amazon's efforts are still in the early stages, a person familiar with the discussion said,

and some of the talks have centered on having the company's own coin for online purchases. The companies have also weighed how to use outside stablecoin, some of the people said, even if they decide not to pursue their own. That could be through a consortium of merchants led by one stablecoin issuer. For example, megabanks have been considering a stablecoin consortium of their own. The Wall Street Journal has reported, end quote. And we talk about how AI adoption right now is a lot about efficiency, which is really all about money. Let me tell you about the ad that betting platform Kalshi aired during the NBA finals.

A 30-second ad, fully AI-generated, that costs, about $2,000 to produce. Quoting the verge, the AI-generated ad highlights various things people are betting on, like whether the Oklahoma City Thunder or Indiana Pacers will win the NBA finals, how many hurricanes will occur this year, and whether the price of eggs will go up this month. It flashes between scenes of an elderly man wearing a cowboy hat while carrying a chihuahua, someone swimming in a pool of eggs, and an alien chugging beer. In a post on X, PJ Ascenturo, who identifies himself as an AI filmmaker, says Kalshi hired him to create the ad using Google's text-to-video generator, V-O-3.

My colleague Alison Johnson recently called V-O-3 a slop-monger's dream. Quote, this took about three to 400 generations to get 15 usable clips. As Satoro writes, one person, two to three days, that's a 95% cost reduction versus traditional ads, end quote. Aseturo outlines his process for. of creating the ad, which he says involved writing a script and then asking Gemini to generate a shot list with prompts for V-O-3. I always tell it to return five prompts at a time any more than that, and the quality starts to slip, Asituro writes. After generating the prompts, Asituro says he paced them into V-O-3 and puts together the ads using a video editing app like Capcut or Adobe Premiere

Pro, end quote. It is time for the weekend long read suggestions, and what have we been talking about this week. We've been talking about stable coins. Actually, though, I've been sitting on this first particular long read for about a week before all this news broke. What if stable coins are the thing that could trigger the next great financial crisis? Why? Because look at the underlying assets stable coins are backed up with. Quoting the FT. With stable coins, the promise is that a dollar is a dollar. They are meant to be backed one for one with reserves of equal value. Holders do not receive interest, but the operators often do to the tune of billions of dollars a year or any adjustment for inflation. But they do get to shoot something that smells a bit like real money

around the cryptosphere with great ease. Back in 2021, warnings were emerging about the risk this poses to normal markets. Rating agency Fitch pointed out that if a stable coin were to fold for any reason, it could be forced to sell all of its holdings, the dollar assets held in reserve, upsetting the underlying markets. Last month, a working paper from the Bank for International Settlements, the Central Bank for Central Banks, cranked up the volume on that warning. In it, Rashad Ahmed and Inaki Aldousoro calculate that when stable coins, of which tether is by far the biggest and most impactful, draw in funds and churn them into reserves, that has a marked impact on the value of short-term U.S. government debt.

That is a reassuring sign that stable coin operators are indeed buying reserves to match their inflows, still, this is a substantial and little-understood market force. According to the researchers, large inflows of over $3.5 billion over five days can place enough upward pressure on the price of short-term U.S. government debt to pull down yields by up to 0.025 percentage points over 10 days. That does not sound like much, but the paper says it is, quote, comparable to that of a small-scale quantitative easing on long-term yields in the same ballpark as central bank efforts to stimulate a flagging economy. So that covers when stablecoin money comes in, but what happens when it goes out is more important. The impact on short-term government debt prices is two to three times larger.

When money comes in, stablecoin operators can exercise some discretion over precisely how and when to buy reserves. When they face redemptions, they have to act faster. We might cheer the assent of stablecoins as a side effect of the relentless crypto boosting from the Trump family as it helps on the margins to lower borrowing costs, although it might be preferable for people to cut out the intermediary and buy short-term debt and enjoy the interest payments themselves. But if anything were to go wrong in crypto in future, hardly a wild theoretical exercise, we might all feel the ripple effects. And either way, it all adds an additional layer of complication for central banks. If the stable coin sector continues to grow rapidly, it may eventually affect the pass-through of monetary policy to Treasury yields.

The researchers say, adding that the opacity of reserve holdings disclosure by Tether complicates efforts to model its possible impacts. The financial stability risks embedded in all this at a time when the U.S. is seeking to foster greater growth in stablecoins are obvious. Stablecoin operators hold more short-term U.S. debt securities than large foreign investors, such as China. Between them, they bought more than $40 billion worth of Treasury bills in 2024, end quote. I want to say that last bit again. Stable coins now. hold more short-term U.S. Treasuries than China does. Finally today, from the shortcut, a quick Nintendo Switch 2 review.

They say it's got great build quality, better haptics, and improved performance for Switch 1 games, but has a weak battery life and only a few launch titles. From their conclusion, quote, The Switch 2 may disappoint some who were hoping for another left-field approach from Nintendo, but honestly, Nintendo would have been foolish to throw away such a winning formula. Instead, it has delivered the ultimate version of the Switch, one that will no doubt be imitated but never beaten by rival companies in the years to come. This is the best console Nintendo has ever made, period. Now it just needs the same suite of compelling software we saw on Switch to truly earn that honor.

It's off to a strong start with the excellent Mario Kart world, but Nintendo needs to keep that momentum going, end quote.