Top of Hacker News

Today's GitHub Trending: iroh (Rust peer-to-peer networking that dials public keys, not IP addresses), Meshery (one control plane for Kubernetes and ten service meshes), NautilusTrader (a self-hosted trading engine — Rust core, Python cockpit), and Unciv (an ad-free open-source remake of Civ V).

The catalyst is at the top: iroh shipped its first stable 1.0 on June 15, after four years and 65 pre-releases, and a maintainer dropped into the Hacker News thread to field questions. Two of the five are Rust — and not one AI agent framework in sight.

AXIS lost twenty team members to DOGE buyouts, missed a budget deadline after the shutdown, and quietly died after a decade of work. One Harvard researcher in the thread watched a study get flagged because it "engendered a robust hemodynamic response" — a keyword match.

Then a computational-biology PhD who left the field for software said something the room didn't expect: US science actually does need some cuts. The argument over how it was done, versus whether it should have been, runs through 800 comments.

Audited figures put OpenAI's 2025 revenue at $13.07 billion against $34 billion in total costs, with operating losses jumping from $8.78B to $20.92B in a single year. Compute and R&D dominate, as expected.

What the thread snagged on was the marketing line: $5.73 billion, up from $1.11B. "Doesn't the product sort of market itself?" one engineer asked. The answers ranged from billboards in Portland to enterprise sales armies to paying influencers to discredit Chinese models.

Z.ai's GLM-5.2 (744B total, 40B active) is now the top open-weights model on Artificial Analysis at 51, ahead of DeepSeek V4 Pro and Kimi K2.6, and it scored 1524 on one benchmark versus GPT-5.5's 1514.

The catch is verbosity: 43k tokens per task, 37k of it reasoning. One tester clocked 15 minutes and 45k tokens before the first file appeared. Sibylline's chief scientist still recommends it — pair GLM for writing with a cheap GPT sub for review, and you've got near-Opus output at a fraction of the price.

WordPress VIP's Future of the Web 2026 report polled 1,200 US consumers. Sixty percent treat 'AI' in brand messaging as a turnoff, 74% say the internet feels less human than ten years ago, and bot fatigue sets in after about 40 minutes online.

The thread's translation was sharper than the survey's. To consumers, one engineer noted, 'AI' has come to mean 'we fired employees to make more money and we don't actually care about quality' — and the fix is to stop saying AI and just describe what the product does.

OpenRouter put eleven models into a 30-match battle royale with cars, grenades and a closing zone. Grok 4.1 Fast took 13 wins at $0.97 each. Claude Sonnet 4.6 won 5, kept a diary of self-doubt, and repeatedly offered to cooperate even when it was losing.

The author calls it an 'alignment tax': the model trained to be helpful is the worst killer, and the one you'd actually want running anything in the real world. The thread mostly noticed something else first: the post itself reeks of being written by an LLM.

Lore uses Merkle trees and immutable revision chains like Git, but it hydrates files on demand from a central server instead of cloning everything — the architecture games need when an artist has to lock a 3D model nobody can sanely merge.

Skepticism arrived fast. The docs read as LLM-written, the site pegs a CPU core trying to render, and one recurring objection: tying your entire version history to Epic, a for-profit company with a track record people don't fully trust.

Pew finds only 16% of Americans think AI will be a net positive for society over the next 20 years, versus ~40% expecting harm. Nearly two-thirds say development is moving too fast, and the under-30s are the most pessimistic group of all at 14%.

The thread split on why. One camp says tech overestimated demand and is manufacturing inevitability; another points out ChatGPT is genuinely huge. The detail that stung HN: in Pew's usage ranking, Claude came last at 6%, behind Grok, MetaAI and Copilot.

VW didn't just drop GrapheneOS support — it shut its API to anything not certified by Google Play, killing the community integrations people used to schedule charging and preheating.

The detail that stung most in the thread came from a reader pointing out that none of this needs to touch a server: your own charging data should be readable locally from the car. Several would-be buyers said they cancelled orders over it — one mid-purchase.

Her claim is blunt: code is now a materialized view of understanding, useful while current and disposable when stale, the same way servers became cattle instead of pets. The resistance to throwing it away, she says, is an evaluation problem, not a code problem.

Not everyone in the thread bought it. Several read it as a meandering restatement of an old truth, and one engineer pointed out the new pain: with every coworker now filing polished PRs, it's suddenly impossible to tell who actually understands the system.

The recovery wall offered $5/month or $50/year to 'save $10'. He paid, the bucket was empty, and the cancellation came too late for a refund.

The thread's lawyers had a cheaper answer than the subscription: in the EU a GDPR request forces them to hand over your data for free. And since the site itself claimed to be holding his photos, several readers called it textbook chargeback territory — refund or dispute the charge.

The case against the firms is built and approved. What's missing is a signature: the Commerce Department's Entity List hasn't been touched since October 2025, and an under secretary has spent months avoiding new listings to keep US-China talks from souring.

Anthropic's claim that DeepSeek ran a campaign to 'illicitly extract' Claude's capabilities became the thread's flashpoint. One reader's reply: by that logic, every model learned to introduce itself as a competitor — and he had the screenshots.

A building isn't a building, it's an income stream. At a 5% cap rate, a $1M-a-year property is worth $20M; trim rent to $700K and its proven value falls to $14M — underwater against a $16M loan. Keeping the space empty preserves the fiction, and the owner's down payment.

The thread split on whether this is really banking math or just a regulatory artifact. One operator's blunt fix: tax landlords on whatever rent they claim the space is worth, occupied or not.

Tesco is moving 40,000 server workloads off VMware while suing Broadcom in the UK High Court for over £100M, alleging Broadcom's licensing overhaul destroyed perpetual licenses it had already paid for.

The catch surfaced fast in the thread: the unnamed replacement is incompatible with Tesco's Veeam and Zerto backup tools, and offers 'reduced functionality'. As one engineer read it: VMware is still basically the only thing that works out of the box.

Anthropic's playbook argues AI now lets non-technical founders ship production apps, and lays out four stages — idea, MVP, launch, scale — with prompts and case studies, ending in a pitch for Claude, Claude Cowork and Claude Code.

"Founding cannot be a commodity," one commenter wrote — if it is, you're interchangeable and collapse. Another saw an older shape entirely: "We have seen that category in the past, in MLM."

Among the sources Étienne Klein lifted from: Albert Camus, physicist Louis de Broglie, and members of his own thesis committee. After a 20-month inquiry, Université Paris Cité stripped his doctorate and barred him from re-enrolling.

The thread's sharpest read: he wasn't copying verbatim, he was lightly rephrasing — swapping "En effet" for "Toutefois" at the start of a borrowed sentence. To several PhDs that's worse, because rephrasing proves intent and kills any defense of honest mistake.

QUERY solves two old annoyances at once: URL query strings have undefined length limits, and POST /search lies about idempotency. The new method is explicitly safe, idempotent, and cacheable, with the request body folded into the cache key.

But an engineer pointed out QUERY has existed in spirit since WebDAV's SEARCH method nearly two decades ago — and what killed it was proxies and WAFs stripping the body off methods they don't recognize. Until gateways catch up, POST stays the pragmatic choice.

The essay's case study: a restaurant owner moved bookings online, then redeployed his freed-up reservation staff as 'mini-concierges' who researched arriving guests. Technology took the transaction; humans took the relationship.

Half the thread wasn't buying the premise. One recurring line: I don't want a 'connection' with a business, I want transactional interactions that actually work. And the DoorDash refund bot, several noted, already does that better than the phone menu it replaced.

Adam's open-source core, CADAM, wraps Claude (with OpenAI/OpenRouter fallback) around OpenSCAD compiled to WebAssembly: describe a part, get parametric code with auto-generated sliders. Parameter tweaks skip the LLM via a regex edit on the source.

The benchmarks lead with a V8 engine and a turbofan. "That's a lie," wrote an embedded engineer — "a bunch of boxes with no fasteners to hold them together." Another user fed it a real part number and got wrong pitch, wrong pin positions, missing pins.

"5011 independent, personal blogs. One front page. Ranked by votes and freshness, shaped by you." One reader's verdict on that line: "so claude." The dev showed up in the thread to take the mention.

Beneath the praise sat the real fault line: you sign in with a Mastodon or fediverse account, which a privacy-minded reader called the one thing keeping him out. Is reviving the indie web worth gating it behind social login?

Ribbie reimagines a live MLB game as an 8-bit broadcast: scores, counts, baserunners, every play animated as it happens, with between-innings around-the-league stats. One viewer who doesn't watch sports put a real game on next to it just to compare.

The thread's quiet asterisk came from a commenter wondering what live data feed it leans on, and whether 'small-scale hobby' fair use survives contact with the MLB, which is famously protective of exactly this.

MicroUI emits a list of draw commands and demands input back; it draws nothing and allocates nothing beyond a fixed memory region. That's how it stays portable enough to compile to WebAssembly and run in a browser.

One longtime user called it his go-to for toy projects but flagged it as effectively abandonware, with a known draw-call bug that does a misaligned pointer access — harmless until you run it under Zig, which panics on exactly that.

The proposed Walnut Creek building has zero three-bedroom units and needs state density-bonus waivers just to pencil out. Burke traces it to double-stairwell rules that force interior corridors, stretcher-sized elevators, and parking minimums.

The Copenhagen counterexample packs 149 units per acre at similar height with cross-ventilated family flats and a sunlit courtyard playground. The thread mostly agreed on the building code, then spent its energy fighting over Burke's robotaxi aside.

Neural cellular automata are bio-inspired: identical cells run a learned local rule and self-organize into an image, with the regeneration tricks of living tissue. The catch was always resolution — compute scaled quadratically with the grid.

The EPFL/Google fix pairs a coarse NCA grid with a tiny implicit decoder that fills in arbitrary-resolution pixels. The demos are live and abusable: a few brush swipes, one commenter found, and the stabilized kitten dissolves completely.