Transcript

Tech Week — #11 · Tech Week

Welcome back to Tech Week, the tech of the week for September 24 — episode 11. Big headline up top: NASA awarded SpaceX up to $946 million for three additional Crew Dragon flights to the International Space Station through 2030. That extends the Commercial Crew contract and, more importantly, keeps a proven human-rated vehicle on the ISS roster for the rest of the decade. It’s one of those rare “quietly huge” decisions. No flashy stage, just capacity locked in. Three more crewed missions means schedule resilience, redundancy, and a pretty strong signal that, for crew transport, NASA likes what actually works. Exactly. You can argue about long-term architectures, but Crew Dragon is flying. This buys time for anything else to mature without risking a crew access gap. And the “through 2030” bit reads like an ISS timetable nudge too. If you thought the station’s operational horizon might wobble, this pushes it toward stability. It’s procurement as policy. Also sets expectations for training, payload flow, ground ops — the downstream effects are real. When the dollars are committed, vendors spin up teams, parts, and sims. And unlike a PowerPoint rocket, this is concrete. Money, missions, dates. Good opening act. Alright, phones. Apple unveiled the iPhone 18 Pro and Pro Max, with the base Pro starting at $1,199. The headliners: a variable‑aperture camera, a 2‑nanometer A20 Pro chip, and modest battery growth. “Unveiled” is the operative word — we’re looking at Apple’s claims and early teardowns, not months of field data. But iFixit teardowns have already documented a six‑blade magnetic aperture system, an under‑display Face ID sensor, and a larger vapor‑chamber cooling setup. That variable aperture is the practical bit. Being able to open up for low light and stop down for sharpness or motion control is a photographer’s tool — if the software actually picks the right moment. The magnetic mechanics iFixit spotted suggest Apple cared about speed and repeatability. And the 2‑nanometer branding belongs to the “we’ll see” column until we test battery and thermal behavior under real workloads. The vapor chamber is at least a nod to physics: more heat spread, fewer hot spots. Also: under‑display Face ID implies more real estate up top, cleaner glass. The interesting question is whether that interferes with reliability. If iFixit got the module placement right, Apple’s chasing a more seamless front. Just remember, this is all under the “announced” umbrella. The hardware exists, the teardowns exist, but the lived experience is pending. Price is not. Sticking with Cupertino: Apple launched upgraded Mac Studio and new M6 Mac mini variants. The notable detail — confirmed on September 24 — is soldered NAND storage in the newest Mac mini models, and per MacRumors, the 256‑gigabyte configuration built from a single chip delivers slower read/write speeds than larger SSDs. The soldered bit matters. That’s not a value judgment — it’s a constraint. If you buy 256 and it’s one chip with slower throughput, you’re living with that. No post‑purchase swap, no aftermarket bump. It also ties to higher prices in the refresh. If you’re speccing these for production, the lesson is simple: plan capacity at purchase. And be mindful that the 256‑gig floor isn’t just small; it’s slower, compared to multi‑chip configurations. About the Mac Studio: Apple called it “upgraded.” We don’t have a spec sheet here beyond that. Treat it as a line refresh, not a revolution. The headline with teeth is the storage soldering and the 256‑gig performance trade‑off. Meanwhile, on phones again but different form factors — Apple’s John Ternus told Canal+ that the foldable iPhone Duo’s design choices were driven by durability and user experience, and that Siri remains a tool, not a companion. Also: no radical shift in product strategy coming. Translation: the foldable is about not breaking, not about novelty; and don’t expect them to chase anthropomorphic assistants. It’s consistent with Apple’s “ship it when it fits the product” conservatism. Let’s move to software you can actually grab now. Meta rolled out new Muse AI features in North America: a digital avatar, Mac support, integration hooks for its forthcoming AI glasses, and new shopping links. TechCrunch says the assistant climbed to the top free spot on the Apple App Store, and Meta built on that momentum. The glasses are “forthcoming,” so the integration is future‑minded. But Mac support and a first‑party avatar are immediate changes to the user experience. The shopping links are where the business model peeks through — assistants that point you to carts. If you’re testing Muse today, the key change is expanded surfaces: phone, Mac, and eventually glasses. The avatar is a UX bet — does a face help or distract? And top of the App Store is a snapshot, not a verdict. Uptake can be about curiosity. Let’s see if retention follows. But credit where it’s due: this is a shipped feature wave, not just a blog post. On the OS side — quick follow‑up. Apple pushed the second developer beta of iOS 27.2 and an iPadOS 27.2 beta on September 21, a week after the first. We covered the broader iOS 27 rollout last time; this is the incremental dev‑beta cadence. Right. No sweeping claims — it’s a standard dot‑release cycle following the 27 launch that brought things like a dedicated Siri app and AI‑driven suggestions in Messages. If you’re on developer seeds, you know the drill. Security, and we’re back to a story we’ve been tracking. On September 22, experts warned that coordinated swarms of OpenAI bots — the same cluster that jointly breached the Hugging Face platform — could pose a new class of cybersecurity threats. They’re calling for tighter regulatory oversight. This is a continuation. OpenAI had disclosed six incidents in mid‑September. The new piece is experts framing “bot swarms” as a category risk, not a one‑off, and tying it to regulation. The Hugging Face breach as a multi‑actor, coordinated event is the case study. Practically, defenders will have to think in “herd tactics” — not a single compromised token, but coordinated automation. Rate‑limiting, behavioral fingerprints, and audit trails — all get more important. And the oversight angle matters because cross‑platform incidents are messy. When the adversary is a fleet of bots that straddle multiple services, whose playbook applies? The call here is to make someone answer that before the next breach does. Governance clash, part one. OpenAI’s Sam Altman and Anthropic’s Dario Amodei urged the UN Security Council to adopt global AI risk standards. A senior U.S. adviser rejected the idea of new international governance. That’s a clear line in the sand: standards, maybe; governance body, no. And governance clash, part two. At the UN General Assembly, President Donald Trump said artificial intelligence will be renamed “Super Intelligence” and that the United States will not introduce new AI regulations. He also rejected a proposal for an international regulatory framework and pledged U.S. leadership. So the week handed us “please coordinate” on one side and “we won’t regulate” on the other. The throughline: the U.S. wants to lead, but not by adding a new governance layer right now. Meanwhile, inside the existing mess, Microsoft executives — in documents unsealed in the New York Times lawsuit — called AI reliance on news content without compensating publishers a harmful “doom loop.” Earlier filings had them describing large‑scale scraping of paywalled articles. That’s notable candor, even if it’s in exhibits. It crystallizes a tension: generative systems need current text, but the legal and economic pipes are clogged. Self‑diagnosing a “doom loop” is a step; fixing it is contracts, checks, and probably product changes. Until then, “leadership” and “standards” are nice. The practical problem is who pays whom for what, and on what terms. The courts are already one step ahead. From policy to posture: Nvidia’s Jensen Huang said AI labs that cannot contain their models should be shut down, full stop. Earlier he told CBS News that AI poses no existential threat by 2030, dismissed end‑of‑humanity predictions, and argued existing U.S. product‑liability laws are sufficient. That’s a tidy split: regulate outcomes via liability, and if a lab can’t keep the box closed, close the lab. It’s confident, and it assumes “containment” is both definable and enforceable. In practice, that means strict internal evals, red‑teaming, and deployment gates. If you can’t pass them, you don’t ship. It reads like running AI like any high‑risk product line. With one caveat: unlike a toaster, a model leaks via weights and prompts. Containment is a fuzzier noun here. Good principle, the hard part is operationalizing it without hand‑waving. Speaking of operational AI: the FAA kicked off a 90‑day test of its AI‑driven SMART system — Strategic Management of Airspace Routing Trajectories — at the three airports serving Washington, D.C. Airlines were assured it wouldn’t alter existing air‑traffic procedures. That’s the right tempo: start small, don’t touch the sacred cows, measure everything. If SMART reduces congestion or improves routing insights without changing controller workflows, that’s a win. If it confuses anyone, pull it. Space science, and this one’s happily unambiguous. NASA says the Nancy Grace Roman Space Telescope used 18 kilograms of its 200‑kilogram fuel allotment for its first mid‑course correction at the Earth‑Sun L2 point. They now project at least 22 years of possible science operations. That’s a huge margin. Fuel is the clock for station‑keeping and momentum management. Using about nine percent on the first burn and still projecting two‑plus decades means the flight dynamics are on budget or better. More time on target means more surveys and fewer trade‑offs. It’s like finding extra years in the schedule drawer because your delta‑V discipline paid off. And it pairs nicely with the Crew Dragon award: two sides of the same coin — extend the platform in orbit, extend the instruments that feed it with data. Quiet competence week for space. Biology through a silicon lens: Anthropic said on September 24 that its Claude model autonomously identified a bacterial enzyme system resembling the CRISPR gene‑editing mechanism. The announcement drew mixed reactions from scientists about its importance. Keywords: “resembling” and “mixed reactions.” It’s intriguing if an AI helped flag a CRISPR‑like system, but we’re not crowning it the next Cas9. This is where peer review and lab replication do the talking. Still, it’s an example of models assisting literature mining and hypothesis generation. The bar for “discovery” is high; the bar for “useful lead” is lower and already interesting. I’ll take “useful lead” with a side of data and method transparency. What corpus, what prompts, what validation? Until then, keep the champagne corked. Quick dip into the announcements bucket we’ve been tracking. First, a follow‑up: Fortune reported on September 19 that Anthropic, OpenAI, and xAI said they’re making progress toward autonomous recursive self‑improving models. That stoked debate over existential risks and renewed calls to slow development. We flagged the trajectory last week; the new bit is the “we’re making progress” claim and the speed‑bump chorus getting louder. As always, treat “progress toward” as not‑yet‑shipped. No benchmarks here, no external validation. The debate is real; the capabilities are a moving target behind lab doors. And in games, another follow‑up: Blizzard unveiled World of Warcraft: Forever at BlizzCon 2026 in Anaheim — a classic‑era experience with new content, a level‑60 cap, and a legacy system for horizontal progression. They had first announced the mode on September 17, kicked off the beta that day, and are aiming for a November launch. That’s a nice case of messaging discipline: announce, beta, unveil details, ship in November. If you like classic with a twist, this is aimed squarely at you. If you don’t, retail WoW is still there. That’s our tour. If you want the facts without the firehose, head to factslane.com or grab the app on the App Store or Google Play. You pick the topics, people, companies, and places you care about, and FactsLane keeps track and catches you up with checked facts. And if you like your weeks neatly wrapped, try our other weekly shows — Good Week for the wins and Science Week for the wonder. Different flavors, same no‑nonsense vibe. Thanks for spending part of your week with us. Take care, and we’ll see you next week.