Today’s thread is infrastructure power being used as policy: a judge says the government illegally used a national security blacklist against Anthropic, a volunteer hosting collective warns how sanctions can turn shared internet services into liability, and the EPA opens room for data center power projects to dodge some pollution rules. From there the day shifts to control at the platform layer, with app-only workflows, an automated copyright takedown hitting Luanti on Google Play, and a case for keyboard-complete GUIs. AI also runs through the slate, from open-weight GLM-5.3 and Claude-like coding prose to faster exploit generation and htmx’s new AI hooks.
-
Judge rules Trump administration’s blacklisting of Anthropic was illegal
-
nytimes.com
- 368 comments
A federal judge ruled that the Trump administration illegally blacklisted Anthropic from defense supply chains, finding the move was retaliation for the company’s speech and contract terms rather than a real national security judgment. The case turns on the government using a national security label to punish a vendor, not on forcing the Pentagon to buy Anthropic’s tools.
-
-
U.S. sanctions against the A/I Collective
-
inventati.org
- 421 comments
This post links to the A/I Collective’s own statement after the U.S. sanctioned the Italian volunteer-run hosting group as a “global terrorist” for providing email, blogs, chat, and other digital services to activist networks. The useful signal is not just outrage at the politics, but a concrete warning about how far infrastructure liability and U.S. financial leverage can reach once a provider is accused of knowingly serving designated groups.
-
-
EPA says power for data centers can sidestep pollution laws
-
epa.gov
- 242 comments
The EPA issued guidance saying some on-site power plants built for data centers can avoid parts of the Clean Air Act’s Acid Rain Program if they are not connected to the public grid. The reaction was mostly furious, but the most useful comments pointed out this is likely a long-standing loophole for backup and remote generators that AI-scale projects are now exploiting.
-
-
“It works better in the app”
-
shkspr.mobi
- 429 comments
A blog post about Google Calendar points out a basic task that works only in the desktop web UI, not in the mobile app or mobile web, and uses it as a springboard to criticize “use the app” culture. Commenters turned it into a wider argument about app-forced workflows, tracking, and how often mobile apps are worse than the websites they replace.
-
-
Luanti removed from Google Play due to baseless AI copyright notice
-
blog.luanti.org
- 132 comments
Luanti, the open source voxel game platform formerly known as Minetest, said Google Play removed its Android app after Tracer.AI filed a copyright notice on Microsoft’s behalf claiming it infringed Minecraft. Comments focused less on whether the claim will stick and more on how easy it is for large rights holders and their vendors to knock smaller projects offline with low-risk automated takedowns.
-
-
GLM-5.3 is now open-weight
-
huggingface.co
- 201 comments
Z.ai released the weights for GLM-5.3, a large open model, and commenters quickly zeroed in on its cheaper, smaller sibling GLM-5.3-Flash as the more practical story. The consensus was that Flash is unusually strong for coding and agent work, often compared favorably with DeepSeek Flash and even Anthropic Opus for some tasks, but speed, token burn, and the economics of running it locally are still real constraints.
-
-
GUIs should be fully keyboard-driven
-
ckardaris.com
- 285 comments
A blog post argues that GUI apps should support every action from the keyboard, not because GUIs should mimic terminal apps, but because many modern interfaces have regressed on speed, consistency, and accessibility. Commenters largely agreed and turned it into a broader indictment of web-style app frameworks, while pushing for better discoverability, stronger native framework support, and treating keyboard navigation as basic accessibility rather than a power-user luxury.
-
-
Show HN: The load-bearing vocabulary of Claude
-
louisabraham.github.io
- 314 comments
A Show HN post visualizes words that have surged in GitHub pull requests and isolates a distinct “Claude-like” vocabulary cluster, with terms like “load-bearing” and “seam” spiking after recent model releases. Readers mostly treated it as proof that AI-written engineering prose is now easy to spot and increasingly irritating to work with.
-
-
Just the rumour of a bug is enough to find an exploit these days
-
anil.recoil.org
- 88 comments
A security-focused blog post argues that modern AI tools can turn even vague hints about a software bug into working exploits fast enough to break traditional disclosure and patching practices. Readers saw it as a real shift in software security: not a brand-new technique, but one that is now cheap, automated, and brutal for maintainers and slow-moving companies.
-
-
Htmx 4.0
-
four.htmx.org
- 120 comments
The htmx 4.0 release post announced a major update to the HTML-first web library, including breaking changes like explicit attribute inheritance and new features such as hx-live and AI “skills” files. Commenters were largely enthusiastic, but the useful signal was about where htmx fits, how teams are using it with server-rendered stacks and LLMs, and where it still loses to full SPA frameworks.
-
-
Please stop flooding our projects with AI slop to furnish your CV
-
neilalexander.dev
- 141 comments
A blog post from an open source maintainer argues that AI-generated pull requests are flooding projects with low-value changes made mainly to pad résumés, not to help the software. Commenters broadly agreed that even correct tiny fixes can be net negative because review and long-term maintenance cost more than the change is worth, though some pushed back that maintainers are also rejecting harmless improvements out of frustration with status games.
-
-
Sovereign Tech Agency invests €500k in Flatpak
-
modal.cx
- 135 comments
Germany’s Sovereign Tech Agency is putting €500,000 into Flatpak, the Linux desktop app packaging and sandboxing system, to fund permission, security, and desktop-integration work. The reaction was split between people who see Flatpak as necessary infrastructure for secure Linux desktops and people who think its security model, UX, and architecture are too compromised to deserve scarce public money.
-
-
Migrating to HTTPX2
-
github.com
- 78 comments
OpenAI posted a migration guide for its Python SDK moving from the stalled `httpx` library to `httpx2`, a compatibility-focused fork backed by Pydantic. The practical changes are small for most users, but the story turned into a referendum on dependency governance, because major AI SDKs are now standardizing on a fork after losing confidence in the original maintainer and release process.
-
-
I used AWS cognito for a startup. I wouldn't do it again
-
joshkaramuth.com
- 118 comments
A startup founder posted a broadside against AWS Cognito, saying the managed auth service saved money up front but cost far more in debugging, rigid configuration, weak customization, and migration pain. Commenters largely agreed that Cognito works and is cheap, but said its real trap is how many irreversible choices and lock-in costs you only discover after launch.
-
-
The Twelve-Factor App (2025)
-
12factor.net
- 112 comments
A 2026 repost of Heroku’s long-running Twelve-Factor App guide sparked a mostly positive check-in on which ideas still hold up. Readers still treat it as foundational for cloud app design, but many now reject its advice to store config and especially secrets in environment variables.
-
-
Stripe said to abandon $50B pursuit of PayPal
-
bloomberg.com
- 253 comments
Bloomberg reports that Stripe and its private-equity partners have dropped their roughly $50 billion effort to buy PayPal after PayPal’s share price rebounded above the implied offer value. Commenters mostly used the news as a referendum on PayPal itself, arguing that it feels hated in tech circles but remains deeply entrenched in parts of Europe and long-tail e-commerce.
-
-
Stopping the smart TV from being used against you
-
s-config.com
- 106 comments
A blog post warns that some smart TVs and monitors can trigger unwanted software installs on a connected Windows PC through Windows Update, even if the TV itself never goes online. The useful part is not the anti-smart-TV rant but the reminder that display identification can be abused as a delivery path for junkware, and that simple hardware or OS-level workarounds exist.
-
-
Smaller reactors bring nuclear power closer to fulfilling its promise
-
nature.com
- 163 comments
A Nature article argues that small modular reactors could make nuclear power easier to finance and build by shrinking plants into factory-made units, with first real tests due later this decade. Commenters were far less convinced, focusing on whether smaller reactors fix nuclear’s core cost problem or just repackage it for data centers and public subsidies.
-
-
Select * from Internet.blogposts
-
pfrazee.leaflet.pub
- 59 comments
A blog post argues that Bluesky’s underlying protocol, atproto, lets anyone treat public posts like a queryable database instead of a tightly controlled API. Commenters mostly focused on whether that openness is genuinely better than RSS and the open web, and on the tradeoff between public indexing and users’ expectations of privacy.
-
-
Show HN: OpenTIE and OpenXWA, Modern Ports of Tie Fighter and X-Wing Alliance
-
- 54 comments
A developer posted OpenTIE and OpenXWA, open-source modern ports of the classic Star Wars flight sims TIE Fighter and X-Wing Alliance that keep the original game data but add native builds and modern rendering features. The reaction was mostly nostalgia, with the most useful discussion focused on when a rewrite beats emulation and what these ports could unlock beyond just replaying old games.
-
-
Bootstrappable Builds: How and Why
-
lwn.net
- 56 comments
An LWN article explained bootstrappable builds, the effort to rebuild a modern software stack from a tiny hand-auditable seed instead of trusting opaque binaries. Readers mostly treated it as a niche but serious supply-chain and resilience project, then argued over whether this solves a real security problem or mostly scratches an engineering itch.
-
-
Hilariously fast volume computation with the divergence theorem (2018)
-
alyssarosenzweig.ca
- 65 comments
A 2018 blog post walks through a very fast way to compute the volume of a closed triangle mesh by turning a 3D volume integral into a sum over its surface triangles using the divergence theorem. Readers mostly treated it as a clean rediscovery of a long-known geometry trick, then added useful context on equivalent formulas, historical references, and where the method breaks or still works in practice.
-
-
Don't use musl if you care about performance
-
blog.brokk.ai
- 57 comments
A blog post argues that binaries linked against musl, a small alternative Linux C standard library often used for static builds and tiny containers, can take a real performance hit versus glibc. Comments mostly agreed that musl’s allocator is the biggest problem, but said slower memcpy and string routines also matter enough that swapping in a better allocator does not fully close the gap.
-
-
Debugging my new network, when 10 Gigabit Ethernet Runs at 300 Megabits
-
hanselman.com
- 31 comments
A blog post walks through why a brand-new 10 Gigabit Ethernet setup on Windows only delivered about 300 Mbps, and how the culprit turned out to be NIC driver defaults and offload settings rather than bad cabling or the switch. The comments broadened it into a practical troubleshooting guide for home and small-office networks, with useful edge cases around 100 Mbps TV ports, pause frames, ASPM, and simple physical-layer mistakes.
-
-
Terminal-Bench-Science: Evaluating AI agents on scientific research workflows
-
terminal-bench-science.ai
- 35 comments
Terminal-Bench-Science is a new benchmark that tests AI agents on end-to-end scientific workflow tasks inside a sandboxed terminal, using task-specific checks instead of toy coding puzzles. The reaction was mostly that this is the right kind of eval to build, but readers immediately zeroed in on whether the tasks, scoring, and public release will hold up once labs start training to the benchmark.
-
-
My Business Is Dying
-
bankstatementconverter.com
- 85 comments
A solo founder wrote that his once-profitable bank-statement conversion SaaS is sliding as Google AI answers cut search traffic and users increasingly turn to chatbots or AI-generated scripts instead of a dedicated tool. Commenters mostly agreed this is exactly the kind of narrow utility app AI is compressing, though many also said pricing, distribution, and product fit problems were already visible.
-
-
Get your Windows license refund
-
en.refund4freedom.org
- 263 comments
A campaign site explains how laptop buyers in parts of Europe can try to reclaim the cost of a preinstalled Windows license if they reject the terms and install another OS. The comments mostly treated it as a long-running but still messy symptom of PC bundling, with practical advice split between chasing refunds and buying from Linux-friendly vendors instead.
-
-
Inception-style curved map for turn-by-turn directions
-
orbify.eu
- 137 comments
Orbify posted a live demo of a navigation map that bends the street grid upward so you can see both the road-level view and the overhead route at once. Readers thought the effect was visually striking and potentially useful for tricky turns or landmarks, but many said the current version is too nauseating, too heavy, and weak at showing what happens after sharp consecutive turns.
-
-
Run Qwen3.8 27B locally: real numbers from my Mac Studio
-
terminalbytes.com
- 65 comments
A blog post benchmarked running the Qwen 3.8 27B language model locally on a Mac Studio and reported much lower token speeds than expected. Most of the useful signal in the comments is that the slowdown likely reflects a bad software stack or model build, not a fundamental knock on the model or on local inference itself.
-
-
EasyEffects can massively improve laptop speaker sound quality
-
osnews.com
- 46 comments
An OSNews post argues Linux users can dramatically improve weak laptop speakers with EasyEffects, a PipeWire audio-processing tool that adds EQ, loudness, and other DSP filters. Comments largely confirm the improvement is real, then push the conversation toward calibration, latency tradeoffs, and whether this should be built into the desktop by default.
-