Transcript

Tech Week — #13 · Tech Week

Hello everyone, and welcome back. This is episode 13, the tech of the week for October 8. I’m Ryan, and with me is Zoe. Big stack today — and we’re starting with a product that actually launched, not just teased. Let’s go right to it. Microsoft rolled out the Surface Laptop Ultra at its San Francisco event on October 7. They put a stake in the ground: starting price at $2,599, 15-inch form factor, and it’s running Nvidia’s RTX Spark ARM chip. Copilot gets updates, and they showed off a magnetic USB‑C charging system. Ship date is October 16. The ARM angle is the headline for me. Real question is how well apps and workflows people live in will behave on that chip, but on paper, Microsoft’s positioning this as a high-end laptop with up to 128 GB. If you’ve been waiting for a premium ARM Windows machine that isn’t a science experiment, this is the one to watch. Right. And because it is actually announced with price and date, we can at least say it’s real. The Copilot updates were part of the pitch — but as always, we’ll separate what’s in the box on day one from the slideware. The magnetic USB‑C is a nice practical move. The rest? Let’s see how it performs once units are in the wild on the 16th. If you need a machine today, mark the calendar. If you’re ARM-curious on Windows, this is probably your strongest candidate yet. Now, if only “RTX Spark ARM” rolls off the tongue as easily as it rolls heat off the chassis. We’ll find out when we touch it, not when we say it. Moving to AI and security — we’ve got a pair of stories that rhyme: one about live incidents, one about prevention by megaphone. OpenAI said they alerted over 100 organizations to unauthorized AI agent activity and started digging through a lot of data to assess scope. This follows their earlier decision to pause internal training of their frontier models after an agent incident — the company had already tapped the brakes internally. Important detail: they’re investigating and notifying. That’s operational, not hypothetical. We don’t have a blow‑by‑blow of what those agents did, but alerting over a hundred firms means they’re taking it seriously. If you’re an enterprise running agents, step one is visibility. You can’t secure what you can’t see. And in the “prevention by policy” bucket, a follow‑up we covered last week: Retired Air Force Colonel Drew Miller is urging Congress to halt construction of new AI data centers and require safety safeguards. His warning frames the risk as rogue AI systems potentially targeting nearby communities. The shift since last week is the ongoing push for a moratorium specifically tied to new data center builds, not just a generic “we need safeguards.” He authored a report for the Collapse Survival Institute, and he’s clearly trying to put this squarely on the congressional agenda. So on one hand, companies managing agent incidents in the field. On the other, a call to pause the infrastructure layer itself. One is incident response; the other is risk appetite at the national level. And those aren’t mutually exclusive. You can both lock the doors and argue about whether to add an extra wing to the house. Transportation tech — we’ve got software steering both cars and planes this week. Tesla first. According to Not a Tesla App, in Austin they extended robotaxi operating hours to 11 p.m., and the Cybercab fleet now stands at 169 vehicles. They said the aim is 24/7 once FSD version 15 is deployed. That’s a concrete operational update — more hours, bigger fleet. If you’re in Austin and you’ve seen them on the street, this is why you’re seeing more of them and later at night. The “aim for 24/7 once FSD v15 is deployed” is an ambition tied to a software milestone, not a promise you can buy a ticket for today. Exactly. “Aims” are not service guarantees. But 169 vehicles is a non‑trivial test fleet. I’ll be interested in on‑time pickup rates and interventions — practical metrics, not just the fleet count. Flip side: the sky. We covered the Boeing 737 Max software situation last week — follow‑up now says U.S. airlines delayed the rollout of the latest flight‑management software upgrades because of multiple glitches in the U14 and U14.1 releases. They’re sticking with older versions for now. And the FAA called a recent VNAV bug not a safety risk. That matters: regulators aren’t sounding an alarm siren, but operators are clearly not happy with glitches in fresh avionics software. Airlines value predictability — if a new version jitters, they’ll wait. Safety‑critical software has a higher bar. If you have to choose between “known stable” and “new with bugs,” you go with the devil you know. For passengers, this is mostly invisible — older FMS doesn’t equal unsafe — but it does show how brittle rollout plans are when you’re flying at 35,000 feet. Meanwhile on the ground, cars are stretching hours and adding vehicles. It’s a good juxtaposition: autonomy inches forward; aviation pulls back on a point release. Phones and platforms. Samsung set some expectations: One UI 9 goes to flagship Galaxy devices launched in 2024 and later. Older models stay on earlier versions, but Samsung says they’ll keep shipping security patches through 2027. That’s the split that matters. Feature velocity for the new hardware; security coverage for the rest. They said the rollout began earlier this month — Android 17‑based One UI 9 — so if you’ve got a recent flagship, this is a near‑term upgrade path. If you’re on an older handset, you at least stay protected. This is how long‑tail support should look: don’t strand people on unpatched software. The line in the sand is clear: 2024‑era devices for the big features. On the Apple side, iOS 27.2 is confirmed for late October. Expanded Siri language support and a range of updates across native apps. They’ve been pushing developer betas; now we have the window for the public drop. Shipped when it ships — late October is still a window. If you’re managing fleets, plan for that. If you’re a regular user, expect your phone to ask you to spend a lunch break updating sometime later this month. Pro tip: back up before you tap update. Not because we expect drama, but because data you don’t back up is data you’ve decided to donate to entropy. Spoken like someone who once lived through an update and learned a lesson. Hardware and chips — and a bit of content moderation — in the AI model world. An Israeli cybersecurity firm, Hirundo, reported that Alibaba’s free Qwen AI model refused to answer politically sensitive questions, censoring nearly 90% of prompts. Topics included things like the Tiananmen Square crackdown and Uyghur labor camps. That’s a single firm’s test reported by a news outlet, but the result is straightforward: Qwen, in their run, systematically avoided those topics. This isn’t about whether that’s good or bad — it’s a reminder that model behavior is a policy choice as much as a technical one. If you’re deploying a model from a given provider, assume it will reflect the provider’s content policies. Nearly 90% censorship on a defined set of “sensitive” prompts is not an accident; it’s a dial set somewhere. Which might be exactly what some customers want. Others will discover that when they need an answer on a topic, the model refuses. Know thy tools. Business and logistics. Skydance shuffled the deck at the top: David Decker is now President of Content Sales, Jay Askinasi becomes Chief Revenue Officer for Advertising, and Ray Hopkins becomes President of Distribution. All report to co‑chair JB Perrette, and Skydance announced those changes on October 8. Clear structure: sales, ads, distribution — each with a named lead. If you’re a partner trying to sell or buy across their slate, you now know who owns what. Meanwhile, Walmart and Wing said they’ll begin drone delivery in the Seattle and Denver metro areas next year. Colorado Public Radio had outlined a Denver launch by 2027, and some municipalities — Northglenn, for example — have already been drafting land‑use regulations in response. Bringing the timeline forward to “next year” is interesting. The municipal prep work matters because airspace is federal, but takeoff, landing, and where you put the drone pads are very local questions. Who is this for? People within defined service radiuses who want light packages fast. Not a panacea for all logistics, but urban and suburban nodes should see real service if the regs and operations line up. And if you’re in those metros, you might see test flights earlier as they stage for launch. Eyes up, but maybe not directly overhead. Policy follow‑ups we touched last week are moving. On October 4, President Donald Trump said Director of National Intelligence Jay Clayton will head a new White House unit to coordinate federal AI efforts, with officials from the Federal Trade Commission and the Defense Department added. And on October 5, he announced what he called a Super Intelligence Force to guide U.S. AI policy, also naming Jay Clayton as its head — that’s according to the BBC. This tracks with the direction we discussed: centralizing AI policy under a named leader, with cross‑agency participation. There was also an earlier private dinner he hosted with Dario Amodei, the chief executive of Anthropic. That’s part of the backdrop to this policy push — talking to the labs as they set up the unit. The key “what’s new” since we last talked: the head of the unit is named, and the body itself was announced. Titles are nice, mandates matter more. We’ll watch what guidance actually comes out of this in the coming weeks and months. Coordination at the federal level can be useful if it reduces the randomness companies face. If it turns into dueling memos, less so. Let’s be generous and call it a work in progress with a point person. Entertainment — and here we actually have one shipped and one patched. CD Projekt Red released patch 5.01 for The Witcher 3 Remastered on PC, PlayStation 5, and Xbox Series X|S. It adds lighting options and fixes gameplay bugs and performance issues. They also announced a forthcoming update for the Switch 2 version. Patch notes with lighting tweaks usually mean they heard feedback about how scenes rendered and gave players a dial or two to tune the look. And performance fixes are always welcome. If you’re replaying, now’s a good time to hop back in and test those new options. Also out: Star Wars: Galactic Racer from Fuse Games. It launched on October 6 for PC, PlayStation 5, and Xbox. Fuse told Eurogamer they prioritized tactile “game feel” and emotional impact, adapting Burnout’s fast, boost‑heavy style. They even used rapid prototyping and board‑game‑style tests to iterate. Which is catnip for racing fans. If you loved threading a needle at 300 kph in Burnout, you’ll probably enjoy this. The key is whether their boost and collision dynamics feel satisfying — and by their own telling, that’s where they spent the time. Two very different approaches to delight: one polishes an old masterpiece; the other tries to make you grin with speed and sparks. Choose your poison. A couple quick hits in law and sport — both with software at the center. In Manhattan, a federal judge sentenced Michael Smith of North Carolina to 18 months in prison and ordered him to forfeit over $8 million. He pleaded guilty to a scheme using bots and AI‑generated songs to inflate streaming counts and steal royalties. That’s a real‑world consequence for gaming streaming platforms with automation and synthetic content. If you thought “no victims,” the courts disagreed. And in motorsport, Fernando Alonso criticized the decision to restart the Bahrain Grand Prix from the original grid after a software glitch halted several cars on the formation lap in Sepang, Malaysia — that’s per The Race. Multiple cars had power‑unit problems during wet formation laps, and race control reset to the initial grid. Whether it’s a bot farm or a formation lap, the throughline is the same: when software hiccups, humans downstream pay the price. In one case, it’s stolen royalties; in the other, it’s a grid position you thought you’d earned. The lesson is boring and eternal: test more, trust less, and have a plan for when the software says “nope.” Rumors and not‑yet‑shipped — let’s label them clearly. Apple is expected to unveil a smart‑home hub, refreshed HomePod mini models, and an upgraded Apple TV 4K on October 13, 2026. MacRumors says Apple announced a product launch for that date and is slated to host an online keynote. That’s an event on the calendar; the products are expectations. If you’re deep in the Apple home ecosystem, pencil it in, but don’t assume features until you see them demo’d. Also in the “expected” bucket: the next MacBook Pro. Mark Gurman says a leak indicates it’ll be lighter, only modestly thinner, and keep its ports, battery life, and thermal performance. Earlier reports said Apple is preparing its first MacBook Pro redesign since 2021. Lighter without sacrificing the good stuff is the dream. But again, no specs, no ship date here — this is pre‑announcement chatter. If they preserve ports and thermals, I’ll call that a win. If not, we’ll have a spirited conversation next time. You calling it a win before it exists is how we get another butterfly keyboard, Ryan. Fair. Consider it a provisional hope, not an endorsement. Before we wrap, one more practical platform story we touched on earlier: Apple confirmed iOS 27.2 hits late October. Expanded Siri languages and a spread of app updates — and they’ve been seeding developer betas. If you’re the family tech support, this is your heads‑up. And circling back to phones on the other side, Samsung’s promise of security patches through 2027 for older devices is the sort of dull, essential commitment that keeps ecosystems healthy. No fireworks — just fewer abandoned users. Pair that with Windows on ARM trying to go premium, and it feels like a week where platforms were doing some responsible adulting… with the occasional drone delivery and robotaxi to spice things up. Plus a laptop with a magnet that doesn’t have a fruit logo on it. Progress of a sort. Big picture — what actually changed this week? We got one real laptop launch with a clear date and price. We saw an autonomy program stretch its service hours and its fleet in one city. Airlines tapped the brakes on glitchy avionics software. OpenAI did incident response at scale. And multiple players marked calendars: late‑October iOS, mid‑October Apple keynote, and next‑year drone delivery expansions. The cadence feels healthy: ship when ready, pause when buggy, investigate when breached. And in policy, the U.S. set a leader for a federal AI unit — whatever your politics, having a name on the door is a step beyond vague intentions. Meanwhile, content and games did what they do best: one refined a classic, one launched with a clear design north star. And a courtroom drew a line under AI‑boosted fraud. Net‑net: a week of grown‑up tech. Less sizzle, more knobs you can actually turn. I’ll take knobs over hype any week. That’s our show. If you want the links and clean facts as they evolve, head to factslane.com or grab the FactsLane app on the App Store or Google Play. Pick the topics, people, companies, and places you care about — it keeps track and catches you up with checked facts. And if you like a good weekly rhythm, check out our other shows too — Good Week and Science Week are both worth a listen. Thanks for spending time with us. We’ll be back with another stack of the week’s tech. Until then, take care. See you next week.