Today centers on control over technology: the EU’s Chat Control vote puts privacy, platform scanning, and procedural legitimacy at the top, while the FTC’s John Deere settlement pushes in the opposite direction by giving owners more leverage to repair their own equipment. AI follows as both product race and working reality, with OpenAI’s GPT-5.6 framed around efficiency and real-world behavior, a piece on LLM burnout, and a look at the electric grid as an AI bottleneck. Elsewhere, infrastructure and hardware show up in Meta’s reuse of old RAM, and modern warfare in essays on fragile logistics and cheaper strike drones.
-
EU Parliament greenlights Chat Control 1.0
-
patrick-breyer.de
- 460 comments
A controversial EU vote revived the temporary legal basis for platforms to scan some private messages for child sexual abuse material until 2028. The key wrinkle is procedural: more MEPs voted against than for, but rejection required an absolute majority of all members, which turned the result into a flashpoint over EU process, privacy, and trust.
-
-
John Deere owners will get the right to repair equipment under FTC settlement
-
apnews.com
- 271 comments
The FTC settled with John Deere, requiring it to give tractor owners and independent shops access to repair tools, software, and manuals instead of reserving them for dealers. Readers saw it as a real right-to-repair win, but a narrow one, because the fine is tiny and Deere still has plenty of room to make repairs expensive and awkward.
-
-
GPT-5.6
-
openai.com
- 736 comments
OpenAI announced GPT-5.6, a new three-tier model family called Sol, Terra, and Luna, with a model card, API docs, and benchmark-heavy launch materials claiming better coding, design, and cost-per-task than Claude Fable and earlier GPT versions. The comments treated the release less as a pure capability jump than as a practical contest over token efficiency, prompt behavior, coding harness quality, guardrails, and whether OpenAI’s benchmark lead will hold up in real use.
-
-
My thoughts on the Bun Rust rewrite
-
andrewkelley.me
- 546 comments
Andrew Kelley, Zig’s creator, published a response to Bun’s announced rewrite from Zig to Rust. Readers focused less on his technical rebuttals than on the post’s openly personal tone, and several called out one factual claim about Bun not fuzzing its code as likely overstated or wrong.
-
-
Meta reuses old RAM in new servers with custom bridge chip
-
theregister.com
- 210 comments
A paper and follow-on reporting describe how Meta is attaching older DDR4 memory to newer servers through a custom CXL bridge chip, effectively turning retired RAM into a slower second tier of memory instead of throwing it away. Readers focused on infrastructure saw this less as a quirky hardware hack than as a response to AI-driven memory shortages, with real implications for server design, costs, and even carbon footprint.
-
-
The glass backbone: Why the Army's logistics will break in the next war
-
mwi.westpoint.edu
- 363 comments
A West Point essay argues the US Army’s logistics model is built for uncontested rear areas that no longer exist, because cheap drones, long-range fires, and persistent surveillance now make fuel dumps, depots, and convoys easy targets. Readers zeroed in on the practical implication that modern armies may fail from supply breakdown long before they run out of weapons, with Ukraine as the live example and US defense manufacturing as the deeper vulnerability.
-
-
I think I have LLM burnout
-
alecscollon.com
- 338 comments
A blog post argues that constant work with large language models can feel mentally draining even when it boosts output, because the job shifts from building to supervising, reviewing, and correcting machine-generated work. The comments strongly agreed, then added a sharper point: the real burnout driver is not the model itself but the flood of cheap output, constant context switching, and pressure to keep humans reviewing at machine speed.
-
-
US seeks cheaper hunter-killer drones after Iran destroys $1B worth of Reapers
-
arstechnica.com
- 305 comments
Ars Technica reports that after Iran destroyed about $1 billion worth of MQ-9 Reaper drones, the US is looking for cheaper long-range strike drones instead of relying so heavily on large, exquisite aircraft. Commenters mostly agreed the loss exposes a procurement system built for expensive, slow-to-iterate platforms while cheap mass-produced drones are reshaping modern warfare.
-
-
What's slowing down the AI buildout
-
worksinprogress.news
- 144 comments
A Works in Progress essay argued that AI expansion is being slowed less by chip supply than by the electric grid, especially transmission, siting, and peak-demand constraints. Commenters mostly accepted the grid bottleneck, then split over what follows from it: build more generation and wires, colocate data centers with power, or take the bottleneck as evidence that AI demand itself is overhyped.
-
-
Postgres rewritten in Rust, now passing 100% of the Postgres regression tests
-
github.com
- 342 comments
A solo developer’s Rust port of PostgreSQL now passes the full upstream regression suite, with claims that an unreleased branch is much faster on both transaction and analytics benchmarks. The conversation focused less on the milestone itself and more on whether test parity, heavy AI assistance, and a lot of unsafe code are enough to trust a database rewrite.
-
-
Why developers are ditching GitHub for Codeberg and self-hosting alternatives
-
howtogeek.com
- 243 comments
A How-To Geek article argues that some developers are moving projects off GitHub to Codeberg, Forgejo, Gitea, and self-hosted setups. The comments mostly said the headline wildly overstates the scale, but they also surfaced real reasons people are hedging away from GitHub anyway: outages, moderation risk, AI scraping, and better control over CI on self-hosted forges.
-
-
Muse Spark 1.1
-
ai.meta.com
- 169 comments
Meta launched Muse Spark 1.1, a paid closed model API aimed at coding, tool use, and agent workflows, with pricing that undercuts some premium rivals on cached context. The post drew interest because Meta looks competitive again, but a lot of the conversation focused on shaky self-reported benchmark claims, limited availability, and frustration that this is not open weights.
-
-
Remote Attestation
-
liamcvw.com
- 106 comments
A blog post explains remote attestation, the hardware-backed way a machine can prove to another system what software it booted and what keys it holds. Readers saw the enterprise security value, but most of the signal was about where attestation breaks down in practice and how the same mechanism can be turned into DRM, platform lock-in, and user control.
-
-
No leap second will be introduced at the end of December 2026
-
datacenter.iers.org
- 175 comments
A formal bulletin from the International Earth Rotation and Reference Systems Service says there will be no leap second at the end of December 2026. The comments turned it into a useful explainer on why leap seconds exist, why Earth’s rotation is hard to predict, and why many engineers want leap seconds phased out entirely.
-
-
DKIM2 and DMARCbis Have Landed
-
stalw.art
- 92 comments
A Stalwart Mail post explains two new email authentication standards, DKIM2 and DMARCbis, aimed at fixing long-standing breakage around forwarding, mailing lists, and message rewriting. Readers saw them less as a new lock-in scheme than as overdue plumbing repairs, though many still complained that operating your own mail server remains fragile because deliverability is driven as much by provider reputation and IP policy as by standards.
-
-
Benchmarking coding agents on Databricks' multi-million line codebase
-
databricks.com
- 65 comments
Databricks published an internal benchmark of coding agents on real tasks from its own multi-million-line codebase, comparing models and agent harnesses on pass rate and cost per task. The standout result was that the Pi harness often beat vendors’ native tools on both token efficiency and total task cost, while open models like GLM 5.2 landed surprisingly close to top proprietary models.
-
-
TLS certificates for internal services done right
-
tuxnet.dev
- 90 comments
A post argued for securing internal-only services with public TLS certificates, split-horizon DNS, ACME, and a reverse proxy instead of running a private certificate authority. Most comments rejected split-horizon DNS as the hard part, and pointed to DNS-01 validation, wildcard certs, or an internal CA as simpler and safer depending on how much control you have over clients.
-
-
Buried Apple feature turns an iPhone into the perfect kids' dumb phone
-
wired.com
- 157 comments
A Wired piece points out that Apple’s little-known Assistive Access mode can turn an iPhone into a highly restricted, large-button phone for kids, seniors, or anyone trying to strip a smartphone down to calls, texts, maps, and a few approved apps. The useful signal is that people found it genuinely practical, but also surfaced sharper options like MDM for tighter lockdowns and a few annoying limitations in Apple’s implementation.
-
-
Spider venom kills varroa mites without harming honeybees
-
connectsci.au
- 138 comments
An Australian research team says peptides derived from spider venom killed varroa mites in lab tests while leaving honeybees alive, offering a possible new way to fight the parasite that drives major colony losses. Commenters liked the direction but stressed that beekeepers need proof it works cheaply and safely in real hives, where current mite control is exhausting and often damages bees too.
-
-
Suspecting AI cheating, Ivy League prof ordered in-person final; scores fell 50%
-
- 149 comments
An Ars Technica piece reports that a Brown professor, suspecting heavy AI use on take-home economics exams, switched to an in-person final and average scores dropped by about half. The comments mostly treat this less as a Brown scandal than as proof that unproctored assessment and grade-based credentials are breaking under generative AI.
-
-
Hy3
-
hy.tencent.com
- 77 comments
Tencent posted Hy3, a new Apache 2.0 licensed large language model that aims to deliver near top-tier results at lower cost and smaller size than flagship open models. The comments mostly treated it as another strong open-model release, but the useful signal was that real-world usability still looks mixed because of serving issues, benchmark skepticism, and hardware tradeoffs versus DeepSeek, Qwen, Gemma, and GLM.
-
-
Show HN: Getting GLM 5.2 running on my slow computer
-
- 80 comments
A solo developer posted Colibrì, a tiny C inference engine that runs the huge open model GLM 5.2 on a 32 GB laptop by keeping the fixed parts in RAM and streaming expert weights from SSD on demand. Readers were impressed by the hack, but the useful signal was about the tradeoff: it proves very large mixture-of-experts models can limp along on ordinary hardware, though current speeds and SSD concerns keep it in experiment territory.
-
-
GLM 5.2 is nearly as accurate as a human book keeper
-
toot-books.pages.dev
- 105 comments
A startup benchmark claimed GLM 5.2 can prepare a UK VAT return almost as accurately as a careful human bookkeeper, with the final amount off by just 7 pence in one test. The comments largely agreed AI can help with narrow bookkeeping tasks, but pushed hard on what the benchmark left out: finding documents, handling exceptions, preventing fraud, and taking legal responsibility when the output is wrong.
-
-
AI changes the economics of software rewrites
-
thetruthasiseeitnow.com
- 103 comments
A blog post argued that AI lowers the cost of rewriting old software systems, especially when moving between common frameworks and languages. Most comments agreed AI can speed up code translation, but said the hard part of rewrites is still deciding what should change and proving you did not break hidden behavior.
-
-
Cargo-nextest: 3x faster than cargo test, per-test isolation, first-class CI
-
nexte.st
- 48 comments
Cargo-nextest is a replacement Rust test runner that executes each test in its own process, aiming to make `cargo test` runs faster and more reliable, especially in CI. The comments are mostly from satisfied users running large suites, with the sharpest caveats around doctest support, database-heavy tests, and endpoint security tools that choke on lots of short-lived processes.
-
-
Girls just wanna have fast MPMC queues with bounded waiting
-
nahla.dev
- 26 comments
A blog post walks through a bounded multi-producer, multi-consumer queue in Rust and originally called it wait-free before adding a correction that it is not. The comments quickly turned into a useful reality check on queue guarantees, pointing to older production-grade designs and explaining why the hard part is not speed but which progress guarantees you are willing to give up.
-
-
A road to Lisp: Why Lisp
-
scotto.me
- 95 comments
A blog post argues for learning Lisp by focusing on what longtime users find distinctive: macros, interactive development, and the way Lisp lets programmers reshape the language itself. The comments were less starry-eyed than usual, pushing on where Lisp is actually unique, where it is dated, and which parts still matter in 2026.
-
-
MIRA: Multiplayer Interactive World Models Trained on Rocket League
-
mira-wm.com
- 16 comments
MIRA is a playable AI-generated version of Rocket League that replaces the game engine with a 5 billion parameter neural network trained on 10,000 hours of gameplay. People who tried it said it feels surprisingly close to the real game, but inputs can be laggy or ignored, which points to both how far world models have come and where they still break as interactive systems.
-
-
Bonnie Tyler has died
-
bbc.com
- 121 comments
A BBC obituary for Welsh singer Bonnie Tyler prompted a big nostalgia-heavy thread centered on “Total Eclipse of the Heart” and “Holding Out for a Hero,” plus a side fight over whether celebrity death notices belong on Hacker News at all. The useful signal was less the obituary than the cultural footprint of those songs and how much of that came from songwriter Jim Steinman.
-