Tech Brew Ride Home - Tue. 06/11 – Did Apple Nail AI, Or Just Do AI In An Apple Way?

Episode Date: June 11, 2024

All of the drips and drabs details from yesterday’s WWDC keynote. Spotify is about to announce a higher tier of membership with some perks. Mistral raises a big new round. The Raspberry Pi IPO is a success. And what if AI could actually make every stoplight in the country more intelligent in real time? Sponsors: Vanta.com/ride Lumen.me/ride Links: Apple’s AI promise: “Your data is never stored or made accessible to Apple” (ArsTechnica) How will Apple’s new AI change your phone? I asked Tim Cook. (Washington Post) Thoughts on #WWDC24 (Chris Messina) Apple’s Push to Infuse Devices With AI Will Take Years to Pay Off (Bloomberg) Spotify to Introduce More Expensive Subscription Tier for Music Lovers (Bloomberg) Mistral secures €600mn funding as valuation soars to almost €6bn (Financial Times) Raspberry Pi shares jump more than a third on first day of trading (Financial Times) The Smart, Cheap Fix for Slow, Dumb Traffic Lights (WSJ) 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 TechMe right home for Tuesday, June 11th, 2024. I'm Brian McCalla today. All the drips and drabs details from yesterday's WWDC keynote. Spotify is about to announce a higher tier of membership with some perks.

Mistral raises a big new round. The Raspberry Pi IPO is a success. And what if AI could actually make every stoplight in the country more intelligent in real time? Here's what you missed today in the world of tech. So obviously we should start with the drips and drabs from WWDC, Apple, Apple, for example, in other WWDC sessions, has outlined two Apple intelligence AI models, a 3 billion parameter on-device language model and a larger server-based model available with private cloud compute. More details on private cloud compute. Again, they're building it with custom Apple Silicon, and Apple says independent experts can

inspect the code that runs on its servers. Quoting ours, Technica, when a bigger cloud-based model is needed to fulfill a generation, AI request in iOS, though. Federigi stressed that it will, quote, run on servers we've created, especially using Apple Silicon, which allows for the use of security tools built into the Swift programming language. The Apple Intelligence System, quote, sends only the data that is relevant to completing your task to those servers, Federigi said, rather than giving blanket access to the entirety of the contextual information the device has access to. And Apple says that minimize data is not going to be saved for future server access or used to

further train Apple's server-based models either. Your data is never stored or made accessible to Apple, Federigi said. It's used exclusively to fill your request, end quote. But you don't just have to trust Apple on this score, Federigi claimed. That's because the server code used by private cloud compute will be publicly accessible, meaning that, quote, independent experts can inspect the code that runs on these servers to verify this privacy promise. The entire system has been set up cryptographically so that Apple devices, quote, will refuse to talk to a server unless its software has been publicly logged for inspection, end quote. In an interview with the Washington Post, CEO Tim Cook said, quote, we went into this saying, these are our values and we can't veer from them.

And we took the time and the depth of thinking to come out with a product that we're proud of. We knew that we had to do things outside of the device because of the size of the language models that we're working with, so we needed to have a level of invention on the cloud. And fortunately, we were able to build it upon things that we had like Apple Silicon, end quote. So I didn't see any hot takes overnight that were super compelling to me, or at least compelling enough to tell you about, although our friend Chris Messina has collected some thoughts. If you want to read his impressions, check the second link in the show notes today. I guess I'll just go with Mark German's summary. Mark says Apple's AI features are quintessentially Apple and probably the right

approach, but many won't launch until 2025, as I feared yesterday. In fact, the new Siri is missing from the iOS 18 beta, quoting Mark German in Bloomberg. In roughly a year and a half, Apple concocted an entirely new AI platform that today remains a work in progress. The features will launch as a public beta test later this year, and it may be a rocky stretch to get to that point. Apple intelligence is still so finicky in internal testing that some features have been stripped out from the version headed to developers. Once they've been polished, the new features should be quite compelling, and they're quintessentially Apple. The technology can handle most of the AI tasks that smartphone users might want in their daily lives. That includes the ability to summarize and

rewrite text, prioritize notifications, edit photos, generate images and emoji, organize content, and transcribe audio. In Apple fashion, the tools should be well integrated enough to not feel obtrusive or clunky. Apple is probably taking the right approach to AI, even down to how it's handling privacy. Much of Monday's WWDC presentation focused on the ways Apple would protect user data. But the company may still be getting ahead of itself. Many of the features unveiled during the event won't be available until next year. Take the updated Siri. It's missing from the first developer beta version of iOS 18. Apple acknowledged during its presentation that powerful new Siri features like advanced app controls won't come until next year.

Also troubling, the features will only initially be available in American English leaving out a lot of people globally. That lag would probably be fine if Apple were ahead of its competitors, but Apple's biggest smartphone rival, Samsung, already offers similar capabilities powered by Google technology, and Google itself has AI features in its own pixel devices. Given all that, Apple Intelligence is unlikely to be a big sales driver for the iPhone 16 this fall. Perhaps the allure of the technology will prevent some customers from switching away to Android. phone, but it's going to be a harder sell to get someone to upgrade based on an incomplete set of features, end quote. Sources say Spotify plans to launch a higher-priced premium monthly plan later this year,

charging $5 or more for better audio, playlist and library tools, and more. Quoting Bloomberg on this new tier. Spotify will position this new option as an add-on for existing customers. That means most users will remain on their current Spotify plans, but customers who choose to upgrade will generate additional sales for the company and its business partners. The new tiers pricing will vary depending on each user's base plan, but will average out to about a 40% markup, said the person, who asked not to be identified because they weren't authorized to speak about the issue.

Among the perks for upgraded subscribers will be access to high fidelity audio. Spotify announced its hi-fi audio feature in February 2021, but delayed the rollout. Since then, competitors like Amazon Music and Apple Music began offering higher quality audio as part of their standard plans. After years of charging its customers one price for access to its service, regardless of how much people use the product, Spotify has started to create multiple price points for different kinds of users. For Spotify, the premium plan could generate hundreds of millions of dollars in additional revenue, which would then be shared with music rights holders. It's unclear how many people want to willingly increase their subscription bill for playlisting tools and higher fidelity audio, especially given high-quality audio, hasn't provided a huge lift to other services. subscribers to the new tier will also be able to instantly generate custom playlists for certain

activities, dates and times of year. The playlist will learn and adapt to each user based on their behavior and eventually develop playlists without prompting, end quote. Paris-based AI startup Mistral has raised 600 million euro. Sources say, comprising 468 million in equity and 132 million euro in debt at a 5.8 billion euro valuation. Quoting the financial times. The investment, which has tripled the company's price tag since December, is led by General Catalyst alongside several of Mistral's existing investors, including Lightspeed, Andreson Horowitz, BPI, France, and BNP Parabas. Corporate backers include Nvidia, Salesforce, Samsung, and IBM. Microsoft, which invested 15 million euro in February as part of a commercial partnership to offer

Mistral's products through its Azure Cloud computing platform, did not participate in the new round. The latest deal is the biggest yet for a startup building large general purpose AI models that is based outside Silicon Valley, as Mistral aims to grow from a European champion to a global contender with more than a billion euro and financial firepower raised. Its co-founders, Timothy LeCroy, Guillaume Lample, and Mensch, a trio of French AI researchers who previously worked at Google's DeepMind and Meta, remain majority shareholders. MENCH said that Mestrel had used a little more than a thousand of the high-powered graphics processing units chips needed to train AI systems and spent, quote, just a couple of dozen millions of euros to build products that can rival those built using much bigger budgets by some of the richest companies in the world, including OpenAI, Google, and META. We seem to be doing very different calculations of the cost needed to build large AI models to the likes of OpenAI,

Chief Sam Aldman, said Munch. Mestrel, which has been lauded by French President Emmanuel McCrone, as an example of how a new generation of European startups can compete with the biggest U.S. tech companies raised $105 million in one of Europe's largest ever seed rounds when it was just a few weeks old in June last year. Its value jumped at $2 billion in December. The capital efficiency of the company is really outstanding, said Jeanette Zhu Furstenberg, who leads General Catalyst's European business, adding that Mistrel had spent, quote, the smallest possible fraction of what its rivals had to build competitive AI models. It really is very, very impressive, how nimbly this team has operated from a financial standpoint and what they have achieved, she added.

Munch said that Mestrel had opted to raise more capital now after receiving inbound interest from investors to scale up its commercialization efforts, as well as to buy more computing resources. It remains a capital-intensive market, he said. The more computing capacity we have, the more people we can get into our team to break the barriers of artificial intelligence, end quote. The AI startup has about 60 staff, including 45 in France, 10 in the U.S., and five in the UK, with about three quarters of those working on product development and research, end quote. And hey, since we're in Europe already, a tech IPO win there.

Raspberry Pi jumped by about 39% on its London trading debut, hitting a 542 million pound market cap after pricing shares at 2 pounds 80 and expects to ship 8.4 million PCs in 2024, quoting the financial times again. demand for the stock, which will go on sale to retail investors from Friday, is a shot in the arm for a London market struggling to attract interest from high-growth technology companies, which mostly prefer to list in New York. The listing caps more than a decade of growth for Raspberry Pi, which started out selling single-board computers for enthusiasts and hobbyists. The company made a profit of almost 40 million last year. Listing Raspberry Pi shares in the UK was, quote, not a patriotic decision, said Chief Executive Ebyn Upton. We did take a look at New York, but we realized that for a company of our scale, the London market is probably a better home.

Upton said there was too much gloom surrounding the U.K. stock market. Most of the stories that people tell about the differences between the U.S. and the U.K., particularly this sort of magical, multiple arbitrage, don't seem to be real. Upton said the money would enable Raspberry Pi to make its supply chains more robust and fund new engineering projects. The London Stock Exchange will be pleased to see that it may have a palpable hit on its hands, even if the deal is not particularly big and the free float is relatively limited, said Russ Mould, investment director at A.J. Bell. Raspberry Pi began in 2012 and has since sold more than 60 million single board computers and computer modules. It expects to sell 8.4 million units in 2024 up from 7.4 million in

2023, according to a registration document. The Raspberry Pi Foundation, a charity founded in 2008 to promote computer science education for young people, held almost 73% of shares in the computer maker before the IPO. It said it would remain a shareholder. The low-cost computer makers' revenues rose to $265.8 million in 2023 with pre-tax profits, almost doubling to $38.2 million, according to filings, end quote. Finally today, what if stoplights could get better with AI? Quoting the journal, Google is rapidly expanding a one-year-old product called Green Light that will create green waves. That is one green light after another. In 14 cities where the system is already deployed, said Juliet Rothenberg, product lead of Climate AI, part of Google Research. And research from the University of Michigan

suggests these kinds of systems could actually save traffic agencies money, making it a no-brainer to adopt them. The secret sauce is that none of these solutions requires new hardware. Instead, they use data gathered directly from new internet-connected vehicles or from navigation apps on their driver's phones to help municipalities adjust the timing of their traffic lights, making them more responsive to real-world traffic patterns. Instead of installing cameras at an intersection, if we know the trajectory of the vehicle, then the vehicle itself becomes the traffic sensor, says Henry Liu, who leads the research team at the University of Michigan.

If this approach could be rolled out nationwide, its proponents say it could make a significant dent in the amount of time we all spend idling at the country's 300,000-plus traffic signals. Data from the cities where Google's Greenlight is already an operation, including Abu Dhabi, Hamburg, Seattle, and Kolkata suggests the system yields a 30% reduction in stop-and-go traffic at intersections. A 2021 study by Traffic Analytics Company Enrics found that on short trips, American spend an average of 10% of their travel time stopped at lights. Reducing that time could help the climate and save our lungs, since research indicates that pollution at intersections is 29 times higher than on open roads,

on account of all that idling. Google uses its own data gathered from users of Google Maps to optimize traffic light timing. Separately, researchers at the University of Michigan used telemetry data from connected General Motors vehicles to help the Detroit suburb of Birmingham, Michigan evaluate the timing of 34 traffic signals in 2022. The resulting changes the city made to the timing of those signals weren't huge, but they had a big effect on traffic flow during rush hour and other important times, says Daniel DeNu, Director of Traffic Safety at Oakland, County. One benefit of this system is that it cuts out the need for the expensive and time-consuming studies normally required to change the timing of a traffic signal, says Lou. Such studies involve

counting the number of cars at a given intersection throughout a day and usually cost around $5,000 every time they are conducted. The cost of such studies is one reason the timing of traffic signals at most intersections is only updated once every five years, and there are some intersections where timing hasn't been updated in decades, he adds. At least half the traffic signals we encounter in our daily lives function, much as they did 30 or more years ago. They are fixed time signals as opposed to the newer dynamic kind, which use either cameras or sensors under the road to detect how many cars are present and adjust their timing accordingly. Today's fixed traffic signals generally have one pattern for peak times, such as rush hour on weekdays and another for off-peak,

like late night or on the weekends, says De Nune. In a perfect world, every traffic signal would be dynamic and connected to a central control system so that an entire city or County could be optimized from moment to moment in much the way that traffic on the internet is constantly optimized to make our connections as fast as possible. Google's Greenlight currently has a waiting list, but the project is rapidly expanding to new cities which Google will announce later this year, says Rothenberg. But Lou says any tech startup or transit agency could do what his team did since they made their methodology available to the public, and anyone is free to use it. All that is required is getting the right kind of real-time data from auto manufacturers or

navigation app makers, end quote. That would be AI making life better that I could definitely get behind. Nothing more for you today. Talk to you tomorrow.