Top Stories

California moves to exempt Linux from its age-verification law after backlash

898 points · tomshardware.com

After a wave of pushback from the open-source community, the same California legislator who authored the state’s controversial operating-system-level age-verification law has filed an amendment to exempt Linux. The original bill would have required OS vendors to collect and verify users’ ages — an obvious non-starter for community-maintained distros with no single legal entity to hold accountable. HN is reading this as a rare win for grassroots advocacy and a reminder of how easily well-intentioned regulation can sweep up the long tail of non-commercial software.

Using AI to write better code more slowly

701 points · nolanlawson.com

Nolan Lawson pushes back on the velocity narrative around AI coding tools and argues the real win — at least for experienced engineers — is quality, not speed. His core claim: the moments where AI saves him real time are narrow (boilerplate, test scaffolding, throwaway scripts), but the moments where it nudges him toward a cleaner abstraction or catches a logic gap he’d have shipped are surprisingly common. It’s a thoughtful counterpoint to both the “10x productivity” hype and the “it’s all slop” backlash, and the comment section is full of senior engineers nodding along.

Mullvad’s exit IP VPN servers mitigation rollout

364 points · mullvad.net

Mullvad is rolling out new exit-IP infrastructure to mitigate the increasingly common pattern of websites blocking or degrading VPN traffic en masse. The writeup is a candid look at the cat-and-mouse game between privacy providers and IP-reputation services, and includes specifics on how they’re rotating ranges, which datacenters they’re moving away from, and what the user-visible impact will be. It’s also a quiet reminder that “use a VPN” is no longer the simple privacy answer it was a few years ago.

Norway’s 2 petabytes of Huawei flash storage and LLM training

278 points · blocksandfiles.com

Norway’s sovereign AI effort has quietly become one of the larger Huawei flash deployments in Europe — roughly two petabytes feeding a national LLM training pipeline. The piece raises two threads HN won’t let go of: the geopolitical optics of a NATO country leaning on Chinese storage for strategic AI infrastructure, and the more practical question of whether bespoke national models can keep pace with frontier labs that have an order of magnitude more compute. Expect this story to keep coming up as more European governments stand up their own sovereign AI stacks.

Motorola phones have started hijacking the Amazon app to insert affiliate codes

191 points · 9to5google.com

Researchers found that recent Motorola firmware is intercepting traffic from the Amazon app and silently injecting Motorola-owned affiliate tags into purchase URLs, redirecting commission payments away from the legitimate creators and affiliates who originally referred the user. It’s a striking example of OEM monetization creeping into territory that most users — and most regulators — would call straightforwardly deceptive. The legal questions (wire fraud? consumer protection? FTC?) are nontrivial, and the thread has plenty of folks asking whether this is uniquely bad or just the most blatant instance of a wider trend.

Nobody cracks open a programming book anymore

225 points · unix.foo

A short essay on how AI assistants have reshaped the learning loop for developers — fewer trips to K&R, SICP, or “Programming Pearls,” and more conversational back-and-forth with a model that already has them in its training data. The author isn’t purely nostalgic; the post is more interested in what’s lost when junior devs skip the slow, deliberate reading that built deep mental models in earlier generations. The HN comments split predictably between “books were never the point” and “you’re going to feel this in five years.”

Hacker News front page as a site

280 points · thefrontpage.dev

A clean little side project that reimagines the HN front page as a proper, browsable site rather than a wall of grey links — think richer previews, story-level context, and better navigation between threads. It’s the kind of thing HN both loves (someone built a thing!) and reflexively critiques (don’t fix what isn’t broken, the minimalism is the brand). Worth a click if only to see how much extra usability is unlocked just by getting out of the 1996 layout.

How Shamir’s Secret Sharing works

218 points · ente.com

Ente (the open-source photos service) put out a clear, visual explainer of Shamir’s Secret Sharing — the scheme that lets you split a secret into N shares such that any K of them can reconstruct it but K-1 reveal nothing. The post walks through the polynomial intuition without drowning in math, and ends with a practical note on where Ente uses it in their own key-recovery flow. A good pointer to send junior engineers who keep asking how threshold cryptography actually works.