This week’s ingest is a study in contrasts: culture-war predictions collapsing at the box office, regulators finally circling AI’s stickier corners, and a hardware release that reads more like a punchline than a product launch. Below is our roundup, with commentary on why each item earned a slot. A few duplicates arrived from different outlets, so we’ve consolidated those. Note: this is a draft pending editorial review.
Sometimes the story is the non-story. The manufactured outrage over Christopher Nolan’s casting choices didn’t dent the film’s trajectory, with the epic reportedly “on track to make $200 million” on opening weekend. It’s a useful reminder that online backlash volume rarely correlates with ticket sales. Via Wired.
Science publishing keeps generating governance scandals, and the American Diabetes Association ouster is the latest. When authors respond to a blocked op-ed by posting it as a preprint, you get a clean illustration of why institutional gatekeeping is losing its leverage. Worth watching as more details surface, via Ars Technica.
OpenAI’s first hardware isn’t the Jony Ive device everyone expected—it’s a $230, 13-switch keyboard billed as a “command center for agentic work.” The gap between that lofty framing and what’s essentially a macro pad speaks volumes about how the industry is scrambling to invent physical interfaces for software agents nobody quite knows how to supervise yet. Covered by both Computerworld and InfoWorld.
Microsoft is trimming OneDrive support for older Windows 10 builds next month, though 22H2 users get a reprieve until 2028. It’s a small deadline, but a familiar pattern: cloud services become the lever that nudges holdouts toward Windows 11. Details at The Register.
Moonshot AI’s 2.8-trillion-parameter Kimi K3 lands as the largest open-source model yet, and the more interesting story is the one about infrastructure and cost. Sheer parameter count is becoming a national-ambition signal as much as a technical one, and “open” at this scale raises real questions about who can actually run it. Via TechRepublic.
Russia’s reusable answer to the Falcon 9 may see grasshopper-style hop tests as early as 2028—which is another way of saying it’s years from anything operational. The gap between announcement and flight remains the defining feature of most SpaceX rivals. Via Ars Technica.
A billing bug briefly told AWS customers they owed billion-dollar sums, and Amazon’s plea to “not panic” is the quotable moment. Beyond the comedy, it’s a reminder of how much blind trust we place in automated cloud metering, and how little visibility users have when it breaks. Via The Register.
Fubo restored some NBCU channels but is charging $15 more for the privilege—and still lacks Versant channels. This is the streaming bundle slowly reconstituting the cable economics it once promised to escape. Consumers are footing the carriage-dispute bill either way. Via Ars Technica.
Decades-old “text salting” tricks are fooling LLM-powered spam filters, which is quietly one of the most instructive security stories of the week. It suggests that swapping heuristic filters for language models doesn’t automatically inherit robustness—it can reintroduce old weaknesses in new clothes. Via The Register.
San Francisco is ordering Apple and Google to pull AI “nudify” apps, with officials estimating the platforms “likely made millions” in fees. This puts app-store gatekeepers on the hook not just for hosting harmful tools but for profiting from them—a framing that could reshape platform liability arguments. Via
Leave a Reply