Today’s thread is trust eroding at the interface between users and AI systems: a backlash against Gmail’s pushy writing and summary features leads into OpenAI’s move onto AWS as an enterprise distribution shift, a lighter-touch federal AI order, and renewed concern about identity-linked internet access through age verification. Around that sit other forms of platform friction and control: legal threats after a reported security exposure, AI spam hitting job seekers, and Larry Ellison’s comments on recording and surveillance. Elsewhere, readers looked at blockbuster IPO absorption, BYD’s manufacturing through CT scans, and KDE’s path to a Wayland-only future.
-
Gmail thinks I'm stupid, so I left
-
moddedbear.com
- 826 comments
A blog post argues Gmail’s new AI writing and summary features feel intrusive enough to make the service unusable, so the author moved to Fastmail with a custom domain. Commenters mostly agreed that Google’s email UI has become cluttered, slow, and pushy, then turned the thread into a practical comparison of alternatives, migration tactics, and what Gmail still does better.
-
-
Can the stockmarket swallow Anthropic, SpaceX and OpenAI?
-
economist.com
- 1267 comments
An Economist piece asked whether public markets can absorb likely IPOs from OpenAI, Anthropic, and SpaceX. The comments focused less on demand and more on rule changes that would get SpaceX into major indexes faster, with readers split between "this is a wealth transfer into an overpriced IPO" and "the actual portfolio impact is small, especially outside Nasdaq 100."
-
-
OpenAI frontier models and Codex are now available on AWS
-
openai.com
- 131 comments
OpenAI says its newest flagship models and Codex coding agent are now sold through Amazon Web Services Bedrock, letting companies buy and run them via existing AWS contracts instead of going direct to OpenAI. The comments treated this less as a product launch than an enterprise distribution shift that weakens Azure’s grip, pressures Anthropic’s Bedrock advantage, and makes OpenAI far easier to adopt inside large regulated companies.
-
-
Age verification for social media, the beginning of the end for a free internet?
-
mullvad.net
- 525 comments
A Mullvad post argues that age verification for social media is being sold as child protection but risks turning anonymous internet use into identity-linked access. The comments mostly agreed on the surveillance danger, but a big subthread said the California bill being cited is narrower than the article suggests and mostly pushes OS-level parental controls rather than mandatory ID checks.
-
-
Adafruit receives demand letter from Fenwick legal counsel on behalf of Flux.ai
-
blog.adafruit.com
- 283 comments
Adafruit said it received a legal demand letter from Fenwick on behalf of PCB design startup Flux.ai after Adafruit found information exposed by a Flux server misconfiguration and planned to report on it. The post gave very little detail, but the comments quickly turned into a broad warning about Flux’s product quality, billing model, and the risks of using legal threats to suppress security or product criticism.
-
-
Please don't spam people looking for employment. It's just cruel
-
news.ycombinator.com
- 270 comments
A Hacker News user looking for work said AI-generated outreach to people in the "Who Wants to Be Hired" thread is emotionally brutal because each message looks like a lead and turns out to be spam. The comments mostly agreed that public job-search posts are now heavily scraped for sales pitches, recruiter spam, and outright scams, with many people saying the signal has collapsed on both the hiring and job-seeking side.
-
-
Trump signs downsized AI order after weeks of reversals
-
politico.com
- 172 comments
Politico reports that Trump signed a scaled-back AI executive order that drops an earlier 90-day review idea and instead asks some frontier AI companies to voluntarily submit powerful models for government review 30 days before public release. Readers saw it as light on substance in legal terms but potentially important as a procurement lever and a way for incumbents to shape competition.
-
-
CT scans of BYD car parts
-
lumafield.com
- 387 comments
Lumafield posted CT scans of several BYD car parts and used them to argue that BYD’s in-house, highly integrated manufacturing looks unusually mature. Readers largely agreed the scans reinforce that Chinese EVs are no longer obviously cheap-looking clones, but the useful debate was about long-term repairability, residual value, and how much of the article was analysis versus marketing.
-
-
Preparing for KDE Plasma's Last X11-Supported Release
-
blog.davidedmundson.co.uk
- 377 comments
A KDE developer says Plasma 6.7 will be the last release with X11 support before KDE goes Wayland-only in early 2027. The comments split between people saying modern Plasma on Wayland is finally good enough and people pointing to still-missing features in accessibility, remote desktop, window control, and compositor-to-compositor compatibility.
-
-
Larry Ellison: "Citizens will be on their best behavior because we’re recording"
-
techradar.com
- 264 comments
A TechRadar piece resurfaced a 2024 Larry Ellison remark that constant recording would keep both police and citizens "on their best behavior," framing it as a warning about AI-enabled surveillance. The comments mostly agreed the larger issue is not cameras alone but cheap machine interpretation at scale, while also pushing back on the article for stripping the quote of context.
-
-
MAI-Code-1-Flash
-
microsoft.ai
- 254 comments
Microsoft introduced MAI-Code-1-Flash, a new coding model positioned as a fast, lower-cost alternative to Anthropic’s Claude Haiku for tools like GitHub Copilot. The reaction was mostly skeptical: people focused on weak benchmark framing, unclear rollout and pricing, and the fact that several open or cheaper models already seem better.
-
-
Love systemd timers
-
blog.tjll.net
- 297 comments
A blog post argues that Linux users should prefer systemd timers over cron for scheduled tasks, because timers integrate with services, logs, missed-run handling, and randomized scheduling. The comments mostly agreed that timers are better for real operations work, but pushed back hard on claims that their syntax or environment handling is inherently simpler.
-
-
Apple rejected my dictation app for using the accessibility API
-
mitmllc.com
- 166 comments
A Mac developer wrote about Apple rejecting the App Store version of a dictation app because it uses macOS accessibility permissions to paste transcribed text into other apps. The comments mostly agreed this is a real App Store policy wall, not a technical block, and focused on the tradeoff between user safety, accessibility power, and Apple’s inconsistent review process.
-
-
Coreutils for Windows
-
github.com
- 246 comments
Microsoft released a Windows-native package of Unix-style core command-line tools, based on the Rust uutils project, to make familiar Linux and macOS commands available without Cygwin or WSL. The reaction was split between people happy to see better native tooling on Windows and people pointing out confusing command conflicts, missing pieces, and overlap with tools that already exist.
-
-
Squillions: How money laundering won
-
lrb.co.uk
- 227 comments
A London Review of Books essay argues anti-money-laundering rules create huge surveillance and inconvenience for ordinary people while doing little to stop large-scale criminal finance, which still flows through cash, trade, luxury goods, shell businesses, and crypto. Commenters largely agreed that AML has become dragnet compliance theater, then split over whether cash is a civic safeguard, a tax-evasion tool, or both.
-
-
Use your Nvidia GPU's VRAM as swap space on Linux
-
- 125 comments
A GitHub project called nbd-vram turns unused Nvidia GPU memory into Linux swap space, mainly for laptops with fixed RAM and an idle discrete GPU. Readers were intrigued by the hack, but the useful signal was that the current implementation is far slower than the hardware suggests and comes with real power, reliability, and desktop-stability tradeoffs.
-
-
macOS needs its grid back
-
blog.hopefullyuseful.com
- 264 comments
A macOS developer posted GridLion, a paid utility that restores the old two-dimensional Spaces grid Apple removed years ago, arguing that spatial desktop layouts were faster and easier to remember than today’s one-dimensional Mission Control. The comments turned into a broader indictment of modern macOS window management and security UX, with plenty of concrete workarounds and a few defenses of Apple’s added friction.
-
-
A walking tour of surveillance infrastructure in Seattle (2020)
-
coveillance.org
- 318 comments
A 2020 guide to Seattle’s visible surveillance systems, from cameras to license-plate readers and Wi-Fi trackers, sparked a broad debate about what these tools actually do and whether they make cities safer. Readers mostly agreed the guide’s politics and technical details were uneven, but used it as a prompt to argue that surveillance keeps expanding because it is easier than fixing policing, prosecution, or the social roots of crime.
-
-
Morningstar values SpaceX at $780B, half its IPO target
-
reuters.com
- 195 comments
Reuters reported that Morningstar put SpaceX’s value at about $780 billion, far below the roughly $1.5 trillion figure tied to its planned IPO. Commenters mostly agreed the lower number is still rich, but the sharper concern was not SpaceX’s rockets so much as the attempt to use rule changes and index inclusion to funnel passive fund money into the stock quickly.
-
-
My thoughts after using Clojure for about a month
-
acdw.net
- 192 comments
A blogger wrote up their first month with Clojure, praising its data-centric style, sequence abstractions, and small core while still struggling with the dense closing-bracket syntax and dynamic typing. The comments mostly agreed that Clojure’s real draw is immutable data plus a strong REPL and editing workflow, then veered into a long argument about whether the JVM now closes Clojure’s old runtime gap with Go and Erlang.
-
-
Why Janet? (2023)
-
ianthehenry.com
- 268 comments
A blog post argues for Janet, a small embeddable Lisp-like language, as a sweet spot for scripting, exploratory programming, and shipping tiny standalone binaries. The comments mostly agree Janet is unusually pleasant for shell-style scripting and embedding, but they also surface real limits around package management, parser tradeoffs, namespacing, and some embedding constraints.
-
-
Fooling around with encrypted reasoning blobs
-
blog.cryptographyengineering.com
- 33 comments
A cryptography blog post dug into the encrypted “reasoning” blobs some AI models send back and found they can act as reusable hidden state, with side effects like replaying internal chains of thought and leaking information through timing. The comments mostly treated it as a clever protocol and systems finding, not a practical break, while surfacing why providers encrypt these traces and where the real state still lives.
-
-
How we index images for RAG
-
kapa.ai
- 29 comments
Kapa.ai posted a writeup on handling images in retrieval-augmented generation by converting each image into a text description at ingest time, then indexing that text instead of sending images through a multimodal pipeline on every query. Readers mostly treated this as a practical, already-proven production pattern, with the useful debate centered on where it beats multimodal retrieval and where it loses detail.
-
-
How is Groq raising more money?
-
zach.be
- 69 comments
A blog post asks why Groq can still raise a reported $650 million after Nvidia took a license to its chip technology and hired much of its leadership and engineering team. The comments mostly land on a simpler answer: Groq was not fully acquired, but people are deeply skeptical that the remaining company still has enough product quality, current models, and staff to justify a big new round.
-
-
Multicore suppport for DOS is real – partly
-
vogons.org
- 16 comments
A Vogons post digs up proof that DOS programs can in fact use more than one CPU core, though DOS itself provides none of the operating system support you would expect for multicore. The HN comments say the trick is less “multicore DOS” than “a DOS app bootstrapping its own tiny SMP runtime,” with demos and old systems showing it is practical if you stay away from DOS and BIOS services on worker cores.
-