Top Stories
WeatherNext 3
199 points · deepmind.google
Google DeepMind is back with the third generation of its machine-learning weather model, and it’s pushing further into territory that traditional physics-based forecasting has long dominated. WeatherNext 3 promises faster, higher-resolution predictions with better handling of extreme events — the kind of thing that matters enormously for everything from agriculture to disaster response.
The HN crowd is fascinated by how far neural weather models have come in just a few years, but there’s also healthy skepticism about verification, calibration of uncertainty, and whether these systems generalize to the rare, high-stakes events that matter most. It’s a great snapshot of AI colliding with a domain that has decades of hard-won numerical methods behind it.
Keep Our Servers Running
939 points · blog.archive.org
The Internet Archive is running a September donation drive where recurring gifts get matched 3x — and the community rallied hard, sending this to the top of the front page with nearly a thousand points. The Archive remains one of the most beloved institutions on the web, quietly preserving the Wayback Machine, digitized books, software, and audio that would otherwise vanish.
For HN, this is close to home: developers rely on archived pages constantly, and the ongoing legal and financial pressures on the Archive make its survival feel genuinely fragile. The thread is full of both gratitude and worry about how such a critical piece of internet infrastructure stays funded.
216M Spy TVs – The LG Smart TV Problem
444 points · youtube.com
A deep-dive video investigation into how LG’s smart TVs — some 216 million of them — collect and phone home data about what you watch, click, and do. It’s the latest entry in the long-running saga of “smart” consumer hardware that treats its owners as the product.
HN readers have strong opinions here: many advocate never connecting a TV to the internet at all, using external streaming boxes, or blocking telemetry at the network level. The discussion doubles as a practical guide to reclaiming privacy from devices designed to surveil, and a reminder that ACR (automatic content recognition) is now standard across nearly every major TV brand.
Trusting-Trust Attack Against an Entire Linux Distribution
123 points · arxiv.org
Ken Thompson’s classic “Reflections on Trusting Trust” gets a modern, large-scale demonstration: researchers show a compiler-level backdoor that can propagate across an entire Linux distribution without ever appearing in source code. It’s the nightmare scenario for supply-chain security made concrete.
This one hits a nerve on HN because it strikes at the foundation of how we trust software we didn’t write ourselves. The thread digs into reproducible builds, bootstrappable compilers, and whether projects like GNU Mes and diverse double-compilation are enough to defend against an attacker who controls the toolchain.
bzip3
364 points · github.com
A modern spiritual successor to bzip2, bzip3 combines a large-block Burrows-Wheeler transform with context mixing and LZP preprocessing to deliver notably better compression ratios — often beating xz on many workloads while staying fast to decompress.
Compression nerds on HN love a well-engineered new entrant, and this thread is full of benchmark comparisons against zstd, xz, and lzip. The consensus is that bzip3 punches well above its weight, though the usual debates about real-world adoption, streaming support, and where each algorithm fits on the speed-versus-ratio curve are in full swing.
This Month in Ladybird – August 2026
136 points · ladybird.org
The independent, from-scratch Ladybird browser continues its steady march toward being a genuinely usable alternative to the Chromium/WebKit duopoly. The August update covers progress on web-platform compatibility, rendering fidelity, and the engineering grind of implementing decades of accumulated web standards.
HN has been rooting for Ladybird as one of the few serious attempts to build a new browser engine from the ground up, free of Google’s influence. Commenters weigh in on how close it is to daily-driver status and why browser diversity matters for the long-term health of the open web.
Simple Is Not Small
177 points · jyn.dev
A sharp essay untangling two ideas software engineers constantly conflate: simplicity and smallness. A small system isn’t necessarily simple, and a simple system isn’t necessarily small — and confusing the two leads to bad architectural decisions.
This is catnip for the HN engineering crowd, echoing Rich Hickey’s famous “Simple Made Easy” talk while adding fresh nuance. The comments turn into a thoughtful debate about complexity budgets, abstraction, and when adding code actually makes a system easier to reason about.
Leaving VMware Just Got Harder After Broadcom Pulled VDDK Downloads
60 points · virtualizationhowto.com
Broadcom continues to tighten its grip on VMware since the acquisition, and this time it’s pulled public downloads of the Virtual Disk Development Kit (VDDK) — a tool many backup and migration products rely on to get customers off VMware. The move makes an already painful exit even harder.
Enterprise-minded HN readers see this as a textbook example of vendor lock-in weaponized post-acquisition. The thread is full of admins sharing migration war stories, evaluating Proxmox and other alternatives, and venting about how Broadcom’s licensing changes have upended their infrastructure planning.
Caltech Mathathon
216 points · mathathonchallenge.com
Billed as the first hackathon ever devoted to research-level mathematics, Caltech’s Mathathon brings the fast-paced, collaborative energy of coding competitions to open mathematical problems. It’s a fascinating experiment in whether hackathon-style sprints can produce meaningful progress on hard math.
The HN discussion connects this to the growing role of AI in mathematics — from proof assistants like Lean to LLMs tackling olympiad problems — and debates whether research math can really be compressed into a weekend format, or whether the best it can do is spark new collaborations and directions.
Also Trending
- Working on Economics with Fable 5 (38 points) — A hands-on account of using the Fable 5 model to reason through economics problems, part of the wave of people probing what the newest LLMs can and can’t do. wilsoniumite.com
- The Dataflow Model Revisited (78 points) — A VLDB paper revisiting the influential dataflow model that shaped modern stream-processing systems like Beam and Flink. vldb.org
- Replaceable but Employed: Automation and the Meaning of Work (51 points) — An NBER paper examining what happens to workers who are technically automatable but still employed, and what it means for the future of labor. nber.org
- Decoding the NEC V20 Microcode (111 points) — A deep reverse-engineering dive into the microcode of the classic NEC V20 CPU, a favorite genre for the hardware-archaeology crowd. martypc.blogspot.com
- Live Map of Public Transport in Belgium (182 points) — A slick real-time visualization of every train, tram, and bus moving across Belgium. openbaarvervoerbelgie.be