Top Stories

Mistral Raises €3B

507 points · mistral.ai

Mistral has closed a €3 billion round, framing itself squarely as Europe’s answer to the American AI giants — “sovereign, open-weight AI at the frontier.” For a company that made its name shipping capable open-weight models while OpenAI and Anthropic went closed, this is a war chest that keeps the open camp funded and in the game.

The HN crowd is dissecting what “sovereign AI” actually buys Europe versus what it costs, and whether Mistral can sustain open weights at frontier scale with backers who will eventually want returns. It’s the fundraising story of the day and a bellwether for whether the open-model bet still commands serious capital.

I’ve Factored the RSA Keys of a Certificate Authority from the 90s

358 points · mcpherrin.ca

A researcher walks through factoring the RSA keys of a 1990s-era Certificate Authority — the kind of small-modulus key that was normal decades ago and is trivially breakable now. It’s a hands-on reminder of just how far compute and factoring techniques have moved, and a cautionary tale about long-lived roots of trust.

HN loves this genre: concrete cryptography, reproducible steps, and a clear lesson about why key sizes and rotation matter. It also lands with extra weight amid the broader “harvest now, decrypt later” and post-quantum anxiety running through the community.

Jellyfin 12.0

393 points · jellyfin.org

The open-source media server hits a major 12.0 milestone. Jellyfin is the self-hosting community’s darling — the ad-free, no-subscription alternative to Plex — so a headline release always draws a crowd eager to talk about what’s new and what finally got fixed.

Comments center on the usual self-hosting themes: transcoding and hardware acceleration, client app parity, and the ongoing Plex-vs-Jellyfin tug-of-war as people reassess who controls their media library.

WeatherNext 3

338 points · deepmind.google

Google DeepMind’s latest weather-forecasting model pushes machine-learning meteorology further, promising faster and more accurate predictions than traditional physics-based numerical models. ML weather forecasting has been one of the most quietly impressive applied-AI stories of the last few years.

HN is interested in the accuracy-versus-interpretability tradeoff and how forecasting agencies will fold these models into operational pipelines that have relied on supercomputer simulations for decades.

Watch Los Angeles Get Built, One Building at a Time (1880–2026)

300 points · parcelscope.net

An interactive visualization animates the growth of Los Angeles building by building across nearly 150 years, drawing on parcel and construction-date data. It’s the kind of beautiful data-driven map that HN can’t resist clicking through.

Discussion mixes admiration for the rendering work with curiosity about the data sources and the eras where the city’s growth visibly accelerated.

TALA Is Open-Source

236 points · d2lang.com

The team behind D2, the modern diagramming language, has open-sourced TALA — its layout engine purpose-built for diagrams. Auto-layout is the hard, thankless part of any diagramming tool, so opening up a dedicated engine is a meaningful gift to the dev-tools ecosystem.

Developers are weighing TALA against Graphviz and other layout engines, and welcoming more competition in the “diagrams as code” space.

We Have a Year to Fix Security Everywhere

243 points · jyn.dev

A pointed essay arguing that the window to shore up software security across the ecosystem is closing fast — and laying out why the author thinks the next year is decisive. It taps directly into the supply-chain and trust anxieties that have dominated infra discussions lately.

HN is debating the urgency framing: some find it a necessary wake-up call, others push back on the deadline as rhetorical, but nearly everyone agrees the underlying problems are real.

Leaving VMware Just Got Harder After Broadcom Pulled VDDK Downloads

205 points · virtualizationhowto.com

Broadcom has pulled public downloads of the Virtual Disk Development Kit — the very tooling many backup and migration products rely on to get workloads off VMware. The timing, amid an exodus driven by Broadcom’s licensing changes, reads to many as making the exit door heavier on purpose.

This is catnip for the enterprise-IT contingent on HN, who are trading war stories about post-acquisition VMware, migration paths to Proxmox and others, and the risks of building on vendor-controlled tooling.

There’s a New “Google Jail” for Independent Wikis

195 points · weirdgloop.org

Operators of independent wikis describe watching their search traffic collapse as Google appears to bury community-run sites in favor of larger platforms. It’s the latest entry in the long-running lament over Google Search quality and the survival of the independent web.

The comments are full of small-site owners sharing similar traffic cliffs and debating whether the answer is SEO, diversification, or simply life after Google referrals.

How Well Do Agents Use Test/Verification Techniques?

94 points · danluu.com

Dan Luu digs into whether AI coding agents actually make good use of tests and verification — or just paper over failures. Given how much of software work is now delegated to agents, the question of whether they verify their own output is central rather than academic.

HN engages seriously with this one: it’s a rigorous, skeptical look from a respected writer, and it dovetails with everyone’s lived experience of agents that confidently ship broken code.