Top Stories

How Google Helped Destroy Adoption of RSS Feeds

537 points · openrss.org

This 2023 essay is back on the front page and clearly still touches a nerve. It argues that Google’s 2013 shutdown of Google Reader wasn’t just the loss of a beloved product — it kneecapped RSS as a mainstream way to follow the web, pushing readers toward algorithmic feeds the platforms control. The piece walks through how Chrome quietly dropped native RSS support and how Google’s dominance shaped what “reading the internet” came to mean.

The HN crowd is deeply sympathetic to open, decentralized standards, so this is catnip: RSS is the classic example of a user-empowering protocol that lost not on technical merit but on platform incentives. Expect nostalgia for Reader mixed with renewed calls to keep self-hosted feeds alive.

Diátaxis

371 points · diataxis.fr

Diátaxis is a framework for structuring technical documentation around four distinct modes — tutorials, how-to guides, reference, and explanation — each serving a different user need. Instead of one sprawling doc that tries to do everything, it insists you separate learning-oriented material from problem-solving material from pure reference.

It resonates on HN because bad docs are a universal engineering pain, and Diátaxis gives teams a concrete vocabulary for why their documentation feels like a mess. It’s been adopted by major open-source projects, and the discussion tends to split between people who found it transformative and those who think rigid taxonomies can become their own kind of bureaucracy.

Seedance 2.5

357 points · seed.bytedance.com

ByteDance’s Seed team dropped Seedance 2.5, its latest AI video generation model, pitching “one-take creation” and flexible referencing — meaning you can feed it reference images or clips to steer character, style, and scene consistency across a generated shot. It’s another step in the fast-moving race to make text-to-video coherent enough for real production work.

For HN, this lands squarely in the ongoing debate about where generative video is headed: the technical leaps are undeniable, but so are the questions about training data, deepfakes, and what happens to creative labor. ByteDance shipping competitive models also reinforces that the frontier isn’t just an OpenAI-and-Google story.

Go 1.27 Interactive Tour

236 points · victoriametrics.com

VictoriaMetrics put together a hands-on tour of what’s new in Go 1.27, walking through the release’s language tweaks, standard-library additions, runtime and toolchain improvements with runnable examples rather than dry changelog bullets. It’s the kind of practical, example-first writeup Go developers love.

Go remains one of the most-used languages among HN’s backend and infrastructure crowd, so every release tour draws a big, engaged audience comparing notes on performance wins and ergonomics. The interactive format is a nice touch that makes the upgrade feel tangible.

Running Kimi K3 on MI355X at Better Performance per Dollar Than B300

135 points · wafer.ai

Wafer benchmarked running Moonshot’s Kimi K3 model on AMD’s MI355X accelerators and reports better performance-per-dollar than Nvidia’s B300. If the numbers hold up under scrutiny, it’s a meaningful data point in the argument that AMD is becoming a credible alternative for large-model inference — historically an area where Nvidia’s software moat has been hard to crack.

This is exactly the kind of story that lights up HN: real benchmarks, a challenge to Nvidia’s dominance, and cost math that matters enormously to anyone deploying LLMs at scale. Commenters will pick apart the methodology, but the direction of travel — more competition in AI hardware — is what people are cheering.

Postmortem for Kernel Soundness Bug #14576

152 points · leodemoura.github.io

Leonardo de Moura, creator of the Lean theorem prover, wrote up a postmortem for a soundness bug in Lean’s kernel — the small trusted core that’s supposed to guarantee proofs are actually correct. A soundness bug there is the nightmare scenario for a proof assistant, since it could in principle let a false statement be “proven.” The writeup details how it slipped in, how it was caught, and what’s being hardened as a result.

For HN’s formal-methods and programming-language contingent, this is fascinating precisely because the stakes are so philosophical: the whole point of a proof kernel is trust, and a candid postmortem from the person who built it is a masterclass in engineering humility and rigor.

Show HN: A 15-Year-Old’s Cycloidal Gearbox

181 points · github.com

A 15-year-old shared the cycloidal gearbox he designed and built — a compact, high-reduction gear mechanism prized in robotics for its precision and low backlash. The repo includes his CAD and build details, and the HN reaction has been overwhelmingly encouraging.

Stories like this are the warm heart of Hacker News: raw curiosity and self-directed learning, no credentials required. Beyond the feel-good factor, the comments turn into a genuinely useful thread of mechanical-engineering advice and pointers for a clearly talented young builder.

I Don’t Recommend Tailwind CSS

100 points · en.andros.dev

A developer lays out the case against Tailwind CSS, arguing that utility-class soup clutters markup, hurts readability, and reinvents styling problems that semantic CSS and modern features already solve. It’s a deliberate contrarian take on one of the most polarizing tools in front-end development.

Tailwind debates are a reliable HN flamewar, and this one delivers — passionate defenders point to velocity and consistency, while critics nod along at the maintainability concerns. Whatever side you land on, the thread is a good snapshot of where the CSS-methodology wars stand right now.