Today’s thread is resistance to AI buildout, led by Nashville using eminent domain to stop a data center near its zoo and reinforced by scrutiny of xAI’s Memphis power setup, both pointing to fights over zoning, power, noise, and local oversight. The AI layer above that looks unsettled too: a human approval game suggesting people miss risky agent actions, and a fast-moving model ranking that mattered less than how close Chinese frontier models now seem to U.S. leaders. Elsewhere, developers wrestled with GitHub reliability, AMD’s bet on etched-in-silicon inference, Android app control, Rust performance, broadcast consolidation, and faster PCB manufacturing.
-
Nashville uses eminent domain to block data center near zoo
-
costar.com
- 466 comments
Nashville’s council voted to use eminent domain to stop a planned data center beside the city zoo, turning a local zoning fight into a national signal about how hard the backlash against AI-era infrastructure has become. Commenters mostly saw it as part of a wider revolt over power, noise, trust, and Big Tech’s reputation, even though many also called eminent domain a dangerous tool.
-
-
GitHub Actions and Pages are experiencing degraded availability
-
githubstatus.com
- 258 comments
GitHub reported degraded availability for Actions and Pages, and developers in the comments described a long outage that blocked CI, releases, and even some self-hosted runner workflows. The discussion focused less on this specific incident than on a pattern of worsening reliability, with people tying it to GitHub’s Azure migration, AI-driven traffic growth, and lock-in around GitHub Actions.
-
-
Humans missed 1 in 3 threats approving AI agent commands across 40k game runs
-
scalex.dev
- 187 comments
A developer published stats from an online “AI agent permission” game that simulates approving coding-agent commands under time pressure. Across about 40,000 runs, players missed roughly one-third of intentionally risky actions, and the comments mostly treated that as another sign that constant human approval is a weak security control.
-
-
Qwen3.8 Max now ranked as the best overall model by agentic index
-
artificialanalysis.ai
- 263 comments
A benchmark page from Artificial Analysis briefly showed Alibaba’s Qwen3.8 Max as the top model on its “Agentic Index,” then flipped after a same-day methodology update. The comments mostly treated the ranking itself as shaky and focused instead on what feels more durable: Chinese frontier models now look close enough to US leaders that price, tooling, and model behavior matter more than tiny benchmark deltas.
-
-
AMD acquires Taalas to boost inference performance by etching models in silicon
-
theregister.com
- 240 comments
AMD bought Taalas, a startup building chips that hardwire AI model weights into silicon to make inference much faster and cheaper. The conversation focused less on the acquisition itself than on whether frozen-model hardware is a breakthrough for cheap, low-latency “good enough” AI or a dead end in a market where models change every few months.
-
-
Federal Communications Commission scraps limit on broadcast TV ownership
-
nbcnews.com
- 95 comments
The FCC voted to remove the long-standing national cap that limited how much of the U.S. broadcast TV audience one company could reach through local station ownership. Commenters mostly treated it as both a likely illegal move, because Congress set the 39% cap in statute, and a practical boost for further consolidation of local news by groups like Sinclair.
-
-
xAI, SpaceX, and the Race for AI Buildout
-
illegal.solutions
- 107 comments
A blog post argues that xAI’s Memphis data center ran gas turbines without proper permits to power a rapid AI buildout, and that the legal and political response shows how large AI infrastructure projects can outrun local oversight. The comments mostly treated this less as a Musk story than as a case study in how companies test environmental and zoning limits when the upside is huge.
-
-
On non-rooted Android 17, ADB uninstall of system apps fails
-
github.com
- 170 comments
A change in Android 17 appears to remove a long-used ADB trick for non-rooted phones that let users "uninstall" preloaded system apps for a single user. The post says those apps will now be disabled or reverted instead, and the comments split between "disable is enough" and "this is another step toward phones users do not really control."
-
-
Branchless Rust: Making a Filter 4x Faster by Removing an If
-
greyblake.com
- 104 comments
A Rust performance post shows how rewriting a filter loop to avoid an `if` made worst-case runtime about 4x faster by removing branch misprediction costs. The comments mostly agreed with the microarchitectural lesson, then pushed further into SIMD, compiler limits, and the tradeoffs the blog skimmed past.
-
-
Launch HN: ProvenMetal (YC S26) delivers circuit boards in days instead of weeks
-
provenmetal.com
- 129 comments
ProvenMetal says it can get assembled circuit boards built domestically in days by automating the messy front half of PCB manufacturing: quoting, design-for-manufacturing checks, and especially component sourcing. The discussion focused less on patriotic manufacturing claims and more on whether this is a real speed advantage over existing US and Chinese options, with commenters pointing to defense procurement, credit terms, and live BOM tooling as the actual wedge.
-
-
Zapscape (CVE-2026-64561): Guest-to-Host Escape in KVM/x86
-
- 9 comments
A GitHub post disclosed “Zapscape,” a claimed guest-to-host escape in KVM on x86 that could let a virtual machine break isolation and affect its host. The comments quickly narrowed the blast radius: it appears most relevant when providers expose nested virtualization, though operators would still treat it as urgent patch-and-migrate infrastructure work.
-
-
France is banning unsolicited telemarketing calls starting next week
-
apnews.com
- 51 comments
France is rolling out a near-total ban on unsolicited telemarketing calls, replacing a weaker opt-out system with an opt-in rule. Commenters mostly treated it as overdue and useful, but said the real test is whether regulators can punish domestic callers and cut off the VoIP providers that let offshore spam operations keep reaching people.
-
-
How to Make a Nintendo 64 Game in 2026
-
phoboslab.org
- 231 comments
A developer posted a detailed write-up on shipping a brand-new Nintendo 64 game in 2026, covering the modern open source tools, cartridge publishing path, and hardware quirks that made it possible. Readers used it as proof that retro console development is no longer a museum exercise and is now approachable enough for hobbyists, small teams, and even experimental AI-assisted workflows.
-
-
Something is changing in the unit economics of software
-
nicolo.xyz
- 54 comments
A blog post argues that AI is breaking classic SaaS economics because every useful model call has a real per-use cost, unlike traditional software where serving one more user was close to free. Commenters mostly accepted that inference changes cost structure, but pushed harder on what actually breaks first: margins, pricing, or the need for many smaller SaaS products at all.
-
-
LLMs won't break symmetric crypto
-
bfswa.blog
- 93 comments
A blog post argues that current AI systems are unlikely to crack modern symmetric cryptography like AES and SHA-2, even if AI speeds up math research elsewhere. Comments mostly agreed on the core claim but sharpened the distinction between breaking the underlying cipher and finding bugs in crypto code, key generation, or deployment.
-
-
Taste Is All That's Left
-
notashelf.dev
- 129 comments
A blog post argues that once AI makes producing code and content cheap, human advantage shifts from execution to “taste” and judgment about what should exist and what is good. Comments mostly agreed that AI raises the premium on discernment, but many pushed back that maintainability, architecture, security, and long-term quality still matter more than the essay’s fuzzy use of “taste.”
-
-
Improving GPT‑5.6 Sol in ChatGPT, expanding GPT‑5.6 Luna access for free users
-
openai.com
- 90 comments
OpenAI says it improved GPT-5.6 Sol in ChatGPT and is giving free ChatGPT users broader access to GPT-5.6 Luna plus the reasoning toggle. The comments mostly treated this as a product and market move, with attention on whether the free tier is finally good enough to change mainstream AI usage and whether the growing model menu has become a usability mess.
-
-
Show HN: I spent 2 years designing a mechanical Magic Keyboard
-
electronicmaterialsoffice.com
- 212 comments
A keyboard maker showed off Altar II, a 4.75mm low-profile mechanical keyboard meant to replace Apple’s Magic Keyboard, with custom switches, a detachable dial, haptics, and a macOS app. Readers loved the industrial design and the ambition, but kept circling the same practical questions: no Touch ID, no ISO layouts yet, a flat chassis, and a polarizing number-row design.
-
-
What I love about Django
-
buttondown.com
- 108 comments
A blog post praised Django for its long-term stability, batteries-included design, and practical developer experience, and the comments mostly agreed while zeroing in on where Django still bites. The main split was not whether Django is useful, but whether its ORM and conventions stay productive as codebases get large and performance-sensitive.
-
-
Unearthing my 1996 windowed OS in machine code for Am29000 homebrew computer
-
nanochess.org
- 32 comments
A retrocomputing post reconstructs a custom 1996 windowed operating system written directly in machine code for AMD’s little-known Am29000 RISC chip, then brings it back to life in a new emulator. The comments add useful context on what made the 29K unusual, where it actually shipped, and how some missing hardware details were filled in by a former AMD engineer.
-
-
Mario Meets Pareto
-
mayerowitz.io
- 147 comments
A blog post uses Mario Kart character stats and animated charts to explain the Pareto frontier, the idea that some choices are strictly worse than others while a small set represents the real tradeoff space. Readers largely liked the teaching approach, then pushed on where the model breaks down in real systems and in Mario Kart itself.
-
-
Unions Can Save Tech Workers and Tech Work
-
jacobin.com
- 109 comments
A Jacobin piece argues that tech workers should unionize, using unionized University of California IT staff as its main example and framing AI, layoffs, and workplace ethics as reasons to act now. Commenters largely agreed that tech’s old boom conditions are fading, but split hard on whether unions would preserve leverage or just import rigid rules into a field that has rewarded mobility and individual bargaining.
-
-
Almost no skill required to cook a steak
-
blog.sydorets.com
- 303 comments
A blog post used steak-cooking as a metaphor for AI coding, arguing that getting good software from LLMs still requires taste, judgment, and enough technical knowledge to tell when the output is wrong. Readers mostly ignored the AI point and argued about the steak analogy itself, with a smaller but sharper set of comments landing on a familiar software conclusion: AI is great for common, well-bounded work, but weak where context, tradeoffs, and maintainability matter.
-
-
Pushing the limits of RISC-V emulation
-
shuklaayu.sh
- 33 comments
A blog post walks through speeding up RISC-V code execution on another CPU by moving from straightforward interpretation toward recompilation techniques, and reports how far each step improved performance. The comments mostly put the results in historical context, saying 10x slowdown is normal for plain emulation and pointing to older JIT and portable-binary ideas the post did not cover.
-
-
Prime Agent: A self-improving RLM agent
-
primeintellect.ai
- 59 comments
Prime Intellect posted Prime Agent, a coding-agent harness built around Recursive Language Models and a self-improvement loop where agents can share code and evolve their own tooling. Readers were interested in the idea but mostly pressed on two points: whether the benchmark claims mean anything, and whether the project is just another bloated agent wrapper that frontier models may soon outgrow.
-