Tech Brew Ride Home - Tue. 03/11 – Biggest Changes To iOS Since 2013

Episode Date: March 11, 2025

Are we about to get the biggest overhaul of iOS since 2013? Is inference the way that everybody is going to eat Nvidia’s lunch? Exactly how much to AI search engines get it wrong? Why is the global smartwatch market shrinking? And apparently the new Mac Studios are the thing you want to get, if you can afford it! Sponsors: Take your personal data back with Incogni! Use code RIDEHOME at the link below and get 60% off an annual plan: http://incogni.com/ridehome 1password.com/ride Links: Apple Readies Dramatic Software Overhaul for iPhone, iPad and Mac (Bloomberg) Exclusive: Meta begins testing its first in-house AI training chip (Reuters) How ‘inference’ is driving competition to Nvidia’s AI chip dominance (FT) Sony is experimenting with AI-powered PlayStation characters (The Verge) AI Search Has A Citation Problem (CJR) Global Smartwatch Shipments in 2024: Market Declines for First Time, China Leads for First Time (Counterpoint Research) Apple Mac Studio (Early 2025) Review: Renewed vigor with M4 Max and M3 Ultra (Tom's Hardware) Apple Mac Studio (M3 Ultra) first look: a weekend with an $8,000 powerhouse (The Verge) 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 TechMean right home for Tuesday, March 11th, 2025. I'm Brian McCullough today. Are we about to get the biggest overhaul of iOS since 2013? Is inference the way that everybody is going to eat Nvidia's lunch? Exactly how much do AI search engines get it wrong? Why is the global smartwatch market shrinking? And apparently the new Mac studios are the thing you want to get if you can afford it. Here's what you miss today in the world of tech. I've heard of Mark German Apple Scoop Monday, but Mark German Apple Scoop Tuesday? Mark's sources says, that Apple is prepping a major interface revamp for iOS 19, iPadOS 19, and MacOS 16, loosely based on VisionOS with a consistent design and simpler navigation.

Quote, the changes are coming as part of iOS 19 and iPadOS 19, code named Luck, and MacOS 16, which is dubbed Cheer. They go well beyond a new design language and aesthetic tweaks. The software will mark the most significant upgrade to the Mac since the Big Sur operating system in 2020. For the iPhone, it will be the biggest revamp since iOS 7 in 2013. The updates are poised to be a highlight at Apple's Worldwide Developers Conference in June and could help distract from the company's tumultuous push into artificial intelligence.

Last week, Apple indefinitely delayed its AI upgrades for the Siri Digital Assistant, confirming a Bloomberg News report that the enhancements were in jeopardy. A key goal of the overhaul is to make Apple's different operating systems look similar and more consistent. Right now, the applications, icons, and Windows. styles vary across MacOS, iOS, and VisionOS. That can make it jarring to hop from one device to another. Still, Apple is stopping short of merging its operating systems, a step other tech giants have taken. The company believes it can make better Macs and iPads by keeping their operating systems separate. Another benefit for Apple is it encourages consumers to buy both devices rather than

getting by with one. VisionOS differs from iOS and Mac OS in the use of circular app icons, a simplified approach to Windows, translucent panels for navigation, and a more. more prominent use of 3D depth and shadows. But the Vision Pro's more immersive experience and use of a hand gesture interface means that some elements won't apply to the 2D world of iOS and MacOS, end quote. You know the thing where if somebody has outsized profit margins in an industry, folks come up from below to eat those margins away. But in the case of AI, it's the customers of the AI chips, specifically Nvidia chips. It's those customers themselves that want to eat Nvidia's lunch, or at least cut down on how much they're paying for lunch.

Sources tell Reuters META is testing its first in-house AI training chip, a key milestone as it moves to design more of its own silicon and reduce its reliance on Nvidia. Quote, the push to develop in-house chips is part of a long-term plan at META to bring down its mammoth infrastructure costs as the company places expensive bets on AI tools to drive growth. Meta, which also owns Instagram and WhatsApp, has forecasted total 20-25 expenses of $114 billion to $119 billion, including up to $65 billion in capital expenditure, largely driven by spending on AI infrastructure. One of the sources said META's new training chip is a dedicated accelerator, meaning it is

designed to handle only AI-specific tasks. This can make it more power-efficient than the integrated graphics processing units or GPUs generally used for AI workloads. Meta is working with Taiwan-based chip manufacturer TSM to produce the chip, this person said. the test deployment began after Meta finished its first tapeout of the chip, a significant marker of success in Silicon Development work that involves sending an initial design through a chip factory, the other source said. A typical tapeout costs tens of millions of dollars and takes roughly three to six months to complete, with no guarantee the test will succeed. A failure would

require meta to diagnose the problem and repeat the tapeout step. Meta executives have said they want to start using their own chips by 2026 for training or the compute intensive process of feeding the AI system reams of data to teach it how to perform. As with the inference chip, the goal for the training chip is to start with recommendation systems and later use it for generative AI products like chatbot meta AI, the executive said, end quote. Meanwhile, others are specifically targeting AI inference to challenge Nvidia. Barclays sees inference CAPEX surpassing training in two years reaching $208.2 billion in 2026, quoting the FT. Invidia's Challenges are seizing a new opportunity to crack its dominance of artificial intelligence chips

after Chinese startup DeepSeek accelerated a shift in AI's computing requirements. Deepseeks are one and other so-called reasoning models such as OpenAIs O3 and Anthropics Clod 3.7 consume more computing resources than previous AI systems at the point when a user makes their request, a process called inference. That has flipped the focus of demand for AI computing, which until recently was centered on training or creating a model. Inference is expected to become a greater portion of the technology's needs. as demand grows among individuals and businesses for applications that go beyond today's popular

chatbots, such as ChatGPT or XAI's GROC. It is here that Invidia's competitors, which range from AI chipmaker's startups such as Cerebrus and GROC to custom accelerator processors from big tech companies, including Google, Amazon, Microsoft, and Meta are focusing their efforts to disrupt the world's most valuable semiconductor company. Training makes AI and inference uses AI, said Andrew Feldman, chief executive of Cerebris, and the usage of AI has gone through the roof. opportunity right now to make a chip that is vastly better for inference than for training is larger than it has been previously. InVIDIA dominates the market for huge computing clusters such as Elon Musk's XAI facility in Memphis or OpenAI's Stargate project with SoftBank,

but its investors are looking for reassurance that it can continue to outsell its rivals in far smaller data centers under construction that will focus on inference. Vipul Ved Prakash, chief executive and co-founder of Together AI, a cloud provider focused on AI that was valued at $3.3 billion last month in a round led by General Catalyst said inference was a big focus for his business. I believe running inference at scale will be the biggest workload on the internet at some point, he said. Analysts at Morgan Stanley have estimated more than 75% of power and computational demand for data centers in the U.S. will be for inference in the coming years, though they warned of significant uncertainty over exactly how the transition will play out.

Still, that means hundreds of billions of dollars' worth of investments could flow towards inference facilities in the next few years if usage of AI continues to grow at its current pace. Analysts at Parkley's estimate capital expenditure for inference in frontier AI, referring to the largest and most advanced systems, will exceed that of training over the next two years, jumping from $122.6 billion in 2025 to $208.2 billion in 2026, end quote. Meanwhile, a video has leaked suggesting that Sony is working on a prototype AI-powered version of a PlayStation game character that can interact with. players through voice prompts, quoting The Verge. An anonymous tipster has shared an internal video from

Sony's PlayStation Group with The Verge that demonstrates an AI-powered version of Alloy from Horizon Forbidden West. After we publish this story, the video was pulled from YouTube due to a copyright claim from Muso, a copyrights enforcement company which advertises Sony Interactive Entertainment, aka PlayStation, as a client. The video is narrated by Sharwin Rego Barjadol, a director of software engineering at Sony Interactive Entertainment, who works on video game technology, AI, computer vision, and face technology for Sony's PlayStation Studios Advanced Technology Group. We watched Rago Bartigal demonstrate an AI-powered version of Alloy that can hold a conversation with a player through voice prompts during gameplay, end quote.

In a demonstration, you see Alloy interacting with users through AI-generated voice and facial animations in both a tech demo and within Horizon Forbidden West. Rago Bartigal emphasizes this is merely a prototype developed in collaboration with guerrilla games to showcase the technology within Sony. The demo utilizes OpenAI's whisper for converting speech to text with GPT4 and Lama 3 handling conversations and decision-making processes. For voice synthesis, Sony employs its proprietary emotional voice synthesis or EVS system, while facial animations are created using Sony's mockingbird technology. Although shown running on a PC, Sony has apparently also tested components of this technology directly on PS5 consoles

with minimal performance impact, according to Ragobarajal. The technology was initially demonstrated internally a year prior, following an enhanced version shown privately at Sony's Technology Exchange Fair in Tokyo last November. This is just a glimpse of what is possible, says Ragobartagall in the video. A study of eight AI search engines found they provided incorrect citations of news articles and more than 60% of queries. GROC 3, for example, answered 94% of the queries incorrectly. Quoting Columbia Journalism Review.

Overall, the chatbots often failed to retrieve the correct articles. Collectively, they provided incorrect answers to more than 60% of queries. Across different platforms, the level of inaccuracy varied, with perplexity answering 37% of the queries incorrectly, while Grok 3 had a much higher error rate answering 94% of the queries incorrectly. Most of the tools we tested presented inaccurate answers with alarming confidence, rarely using qualifying phrases such as it appears, it's possible, might, etc., or acknowledging knowledge gaps with statements like I couldn't locate the exact article. ChatGPT, for instance,

incorrectly identified 134 articles but signaled a lack of confidence in just 15 times out of its 200 responses and never declined to provide an answer. With the exception of co-pilot, which declined more questions than it answered, all of the tools were consistently more likely to provide an incorrect answer than to acknowledge limitations, end quote. Five of the eight tested chatbots, chatGPT, perplexity, Perplexity Proplexity Pro, ProPy and Gemini have publicly disclosed their crawler names, allowing publishers to block them if desired, while three others, Deepseek, GROC 2, and GROC3 have not revealed their crawler's identities. The researchers anticipated that chatbots would accurately respond to queries about

publishers that had granted crawler access and refused to answer questions about websites that had blocked their crawlers. However, their observations contradicted this thesis as the actual behavior of these systems differed significantly from this theoretical compliance model. What's up with this? Global smartwatch shipments fell 7% year-on-year in 2024, which makes this the market's first ever decline. Apple fell 19% while Huawei grew 35%, Jami grew 135%, and Samsung grew 3%. But is the smartwatch industry saturated? Quoting Counterpoint Research, commenting on Apple's performance senior research analyst Anishka Jane said Apple Watch witnessed a decline in momentum on its 10th anniversary, despite the launch of the S-10 series. The biggest driver of the decline was North America,

where the absence of the Ultra-3 and minimal feature upgrades in the S-10 lineup led consumers to hold back purchases. Additionally, patent disputes limited shipments in the first half of the year, the slowdown of the existing Apple Watch SE lineup and the lack of new SE models also contributed to the decline. Commenting on the different smartwatch segments, research analysts Balbier Singh said only the kids' smartwatch segment witness growth, with EMU remaining the market leader driven by its affordable feature-rich offerings. The kids' smartwatch segment is gaining traction as parents are concerned for their children's safety, and they desire to track and stay constantly connected with their children. With the decline of basic smartwatch's other brands such as noise, boat, and Google's Fitbit

have started diversifying their portfolio to include kids' smartwatches. This is a good sign for the market's diversification and indicates further growth in the future, end quote. Got our kids' Apple Watch SEs as a way of delaying getting them smartphones. All of their friends have them too. Finally today, the reviews for the new MacBook's air with the M4 chip are out. But I'm going to skip that because it's the same MacBook, just better chip. I'm going to instead end today with a quick review of the Mac Studio, because apparently the gains in performance there are notable.

Quoting Tom's hardware, the Mac Studio is a fascinating product for less than the cost of just a Zeon or Threadripper workstation processor, you can get a fully functioning MacOS machine with some serious horsepower under the hood. The M3 Ultra dominated most of our CPU tests, except for the single thread geekbench test that favored the M4 Max, and the Blender CPU test, which favored the Zeon and Threadripper with their huge advantages and core counts. The M3 Studio blew through our handbrake transcode test in record time,

aced our file transfer test, and held its own on the Blender CPU, benchmark tests against the Intel and AMD competition. However, when we switch to the GPU in the Blender test, the Macs Studio blew the doors off the competition, with the M4 Max trailing not too far behind. The new Macs Studio achieves these feats silently, as even under full load, I could only hear the dual internal fans if my ear was up against the exhaust vent. But for all its pluses, the Mac Studio still has some downsides. There's no way to upgrade your memory or SSD after purchase, so you'll need to forecast what your needs will be a few years down the road regarding hardware resources.

Also, memory and SSD upgrade prices are far pricer than what you find on the PC side. But if you're firmly in the Apple Camp or even a PC stalwart, that doesn't mind expanding his or her horizons to the macOS realm, the Mac Studio remains a compelling choice in a compact package, end quote. And quoting the Verges, whose review unit was a specked-out $8,000 beast, Quote, like its predecessors, the 2025 Mac Studio delivers breaknet performance while running shockingly quiet, as my former colleague Monica Chin wrote two years ago. That remains just as true today as it was then. In a few days of testing, this thing has laughed at my lightroom edits,

made quick work of Adobe AIs, noise reduction, and other effects, and I've never so much as heard any fan noise. The M3 Ultra Chip is overkill for many. If you need this level of power, you already know exactly how you'll get the most from it. It's for visual effects, artists, and animators. It's for professionals doing ambitious audio and video production work. Are you regularly crunching big medical data sets? Maybe you can use all those cores and memory to their fullest potential. And AI development continues to flourish. The kidded out configurations with 256 gigabytes or 512 gigabytes of memory could prove appealing to anyone interested in running sophisticated LLM models locally on their machine. I've only had our

Mac Studio review unit for a few days, so for now I'm providing the usual benchmarks and setting it up for some of these LLM test cases to gauge what it's capable of. In the weeks ahead, I'll also be looking to friends and experts and other fields that can fully appreciate the studio's capabilities to see what they think of its performance. If you've got ideas or tests, you'd like to see us run, though, feel free to share them in the comments, end quote. That kind of makes my first-generation Mac Studio with the M1 Ultra Chip have a sad. but hey, it's still performing great.

And remember, I future-proof this thing. Hopefully it can get me into the 2030s. We'll see. Talk to you tomorrow.