Top Stories

Anonymous GitHub Account Mass-Dropping Undisclosed 0-Days

827 points · github.com

An anonymous GitHub account has been publishing a stream of previously undisclosed zero-day vulnerabilities, with no coordinated disclosure to affected vendors. It’s the kind of story that lights up HN for two reasons: the immediate scramble to assess whether anything you run is on the list, and the perennial debate over full disclosure ethics. Dropping live exploits in the open arms attackers as much as defenders, and the lack of a paper trail makes attribution and cleanup messy.

The thread is already wrestling with whether this is reckless grandstanding or a deliberate prod at vendors who sit on patches. Either way, security teams are spending their weekend cross-referencing.


DSpark: Speculative Decoding Accelerates LLM Inference

764 points · github.com

DeepSeek published a paper on DSpark, a speculative-decoding technique aimed at squeezing more tokens-per-second out of LLM inference. Speculative decoding — drafting candidate tokens with a small fast model and verifying them in a batch with the big one — has become one of the highest-leverage tricks for cutting serving costs, and DeepSeek tends to ship work that’s both practical and reproducible.

HN’s interest here is less academic novelty and more “can I run this.” Inference cost is the dominant line item for anyone deploying models at scale, so any credible speedup gets read closely by the people footing the GPU bill.


OpenRA

731 points · openra.net

OpenRA, the open-source engine that reimagines classic Command & Conquer and Red Alert RTS games for modern systems, surged to the front page. It’s a long-running labor of love — modernized controls, online multiplayer, and ongoing support on platforms the originals long abandoned — and a recurring favorite whenever HN gets nostalgic about the golden age of real-time strategy.

The appeal is partly preservation: keeping a beloved genre playable decades later, with the source open for anyone to study or extend.


Fintech Engineering Handbook

587 points · w.pitula.me

A free, comprehensive handbook on building financial systems — covering ledgers, double-entry accounting, idempotency, reconciliation, and the unglamorous correctness guarantees that separate fintech from typical CRUD apps. With 587 points and almost no comments, it’s the kind of resource people upvote and bookmark rather than argue about.

Money systems punish hand-waving: a dropped event or a rounding bug becomes a real-dollar discrepancy. Guides that codify the hard-won patterns are perennially valuable to engineers crossing into the domain.


The Case for Physical Media Ownership

446 points · dervis.de

A pointed argument for owning physical media — discs, books, files you actually control — in an era where streaming licenses vanish, libraries get silently edited, and “purchases” can be revoked. The piece resonates with HN’s long-standing unease about renting access to culture instead of owning it.

The comment section is the real draw: people swapping strategies for local-first media collections, DRM frustrations, and the slow realization that “the cloud” is someone else’s permission slip.


AI Learns the “Dark Art” of RFIC Design

237 points · spectrum.ieee.org

IEEE Spectrum reports on AI systems tackling radio-frequency integrated circuit design — a notoriously intuition-heavy specialty where seasoned engineers rely on hard-won feel as much as formal method. Automating even part of that “dark art” hints at where ML-for-EDA is heading: not just layout optimization, but the genuinely creative corners of chip design.

HN’s hardware crowd is split between excitement about compressing design cycles and skepticism about whether models can match the edge-case judgment that separates a working RF chip from a paperweight.


Asian AI Startups Launch Mythos-Like Models as Export Ban Drags On

238 points · techcrunch.com

TechCrunch covers a wave of Asian AI startups shipping models in the mold of the “Mythos” generation, positioned to fill demand left by ongoing export restrictions affecting Western labs. It’s a clean illustration of how trade policy reshapes the competitive map — restrictions meant to slow rivals often accelerate homegrown alternatives.

The geopolitics-meets-AI angle reliably draws a big HN thread, with debate over whether export controls actually preserve any lead or just hand market share to fast-moving regional players.


Turn Your Site into a Place People Can Bump into Each Other

241 points · cauenapier.com

A developer released “Town Square,” a drop-in way to add presence and serendipitous interaction to a personal website — letting visitors see and bump into each other rather than browsing alone. It taps the recurring HN nostalgia for a more social, less algorithmic web, the era of webrings and chatrooms before everything funneled through a handful of platforms.

Whether ambient presence is delightful or creepy is exactly the debate the thread is having, alongside the usual technical questions about how it scales without becoming a moderation headache.


Choosing a Public DNS Resolver

178 points · evilbit.de

A practical guide to picking a public DNS resolver — weighing privacy policies, filtering options, performance, and the trust you’re extending to whoever answers your queries. DNS is one of those quiet infrastructure choices that touches every request you make, so HN treats resolver selection as a real privacy and reliability decision rather than a default to ignore.

Expect the comments to relitigate Cloudflare vs. Quad9 vs. self-hosting, with the usual reminders that your resolver sees everything you visit.