Saving 100 terabytes of memory by optimizing 1.1.1.1's DNS cache

https://blog.cloudflare.com/_image?href=https%3A%2F%2Fblog.cloudflare.com%2F_emdash%2Fapi%2Fmedia%2Ffile%2F01M117Y6R8CDGVT4K2XX6VQ2YX.png&w=1200&h=675&f=webp&fit=cover&position=center
Big Pineapple, Cloudflare's DNS platform, reduced its cache memory usage by 56% through five optimizations. The changes freed up over 100 terabytes of memory across the fleet and improved performance with a 43% increase in cache insert throughput and a 19% drop in lookup latency.

Doctors are finally learning to manage antidepressant withdrawal

Small Models Have Arrived

https://calv.info/small-models-have-arrived/cleanshot-2026-08-26-at-2033392x-p7k4wq.png
The author has been impressed with the capabilities of GPT-5.6-Luna, but notes that its high cost is a significant barrier to adoption for consumer AI companies. They believe that fast and affordable models will become increasingly important for businesses, particularly those requiring human-like responsiveness and efficiency.

507 Mechanical Movements

https://507movements.com/img/th/mm_001.png
The website has some working animations, but not all 507 are complete. Users can browse completed animations by color thumbnails and subscribe to updates on progress.

Show HN: We built open OpenRouter that turns usage into a better model

https://raw.githubusercontent.com/experientiallabs/experiential/main/assets/experiential-workflow.png
The user wants to control access to models, set up a custom router for quality, speed, and cost, and start a local OpenAI-compatible gateway. They also discuss the benefits of using a managed gateway versus running one locally.

Gemini-3.5-Transcribe

https://storage.googleapis.com/gweb-uniblog-publish-prod/original_images/Omni_1-1_Flash_hero.png
Gemini 3.5 Transcribe is a precise speech-to-text model designed for intelligent voice interactions, offering advanced capabilities like noise handling and custom vocabulary recognition. The model achieves high accuracy rates across various languages and use cases, with applications in the Gemini app, Android, Google AI Studio, and more.

We found a division by zero bug in FFmpeg with a vibecoded fuzzer

https://code.ffmpeg.org/.within.website/x/cmd/anubis/static/img/pensive.webp?cacheBuster=v1.27.0
Anubis is a website protection system that prevents AI scrapers by requiring proof of work, causing temporary downtime. Users can reload the page or disable plugins like JShelter to access the site.

Show HN: OpenTIE and OpenXWA, Modern Ports of Tie Fighter and X-Wing Alliance

https://raw.githubusercontent.com/elyosh/OpenTIE/main/assets/opentie-001.jpg
OpenTIE is an open-source reimplementation of Star Wars: TIE Fighter, combining 1995 and 1998 game data for a unique experience. It supports various graphics modes, flight simulation speeds, and soundtrack options while remaining under active development with potential bugs.

Microduck

https://pollen-robotics.com/assets/microduck/squad.webp
A 25 cm open-source biped robot is being pre-ordered for $399, shipping before Christmas 2026. The robot's source code and training data are available on GitHub, allowing users to customize and retrain it.

Decompiling a Nintendo 64 game in 84 days

https://blog.chrislewis.au/snowboard-kids-progress.svg#darksafe
The original Snowboard Kids game has been fully decompiled, with all functions having matching C implementations that produce identical machine code to the original game. This was achieved through a collaborative effort involving several individuals and AI models.

AI Engineer Notebooks – free, framework-free RAG/agents/evals on Colab

https://repository-images.githubusercontent.com/1331428199/00fb8916-a7d6-4fde-9f57-93a2c480c40b
This resource teaches the applied-LLM stack through framework-free, on-a-free-API notebooks for AI Engineers and Forward Deployed Engineers. It covers topics such as model APIs, RAG, agents, adaptation, serving and more with real case studies.

The turbulent AI era is here

You don't have permission to access "http://www.gatesnotes.com/work/make-ai-work-for-everyone/reader/a-turbulent-ai-era-and-critical-choices-to-make?" on this server.

M5Stack Launches PaperMono

https://cdn.shopify.com/s/files/1/0056/7689/2250/files/m5stack-paper-mono-news.png?v=1787217130
M5Stack launches PaperMono, a compact ESP32-S3-based E-Ink terminal for low-power projects. It features a 3.97-inch display, touch input, and connectivity options like NFC and LoRa.

Previewing the Model Hardware Standard

https://www.anthropic.com/_next/image?url=https%3A%2F%2Fwww-cdn.anthropic.com%2Fimages%2F4zrzovbb%2Fwebsite%2F0aa7623ba0e545ce39e0de725583c0445ce6b76d-1999x1250.png&w=3840&q=75
Researchers have developed the Model Hardware Standard (MHS), a shared specification for AI agents to safely operate physical devices, enabling them to work in parallel and perform intricate tasks. MHS reduces integration time from weeks or months to hours or minutes, allowing researchers to focus on high-level experimental design and reasoning.

Suica, Japan's First IC Transit Card

https://assets.tokyodev.com/tokyodev-production-r2/variants/63rvivq9bhtbm6imkjsgucw7j1dn/f71b87b38636b7591ceb09953914ef235c0903dc949c343b7f802d202ec96b92
Suica, a Japanese IC card, was developed by Sony and JR East in the 1990s to improve train station efficiency. It uses FeliCa technology to process transactions locally on the card itself.

Afterglow: Run classic After Dark screen savers on modern macOS

https://morphing.cloud/afterglow/afterglow-app-icon.webp
Afterglow is an emulator for running original 68k After Dark screen savers on modern macOS. It allows users to import and run classic modules without dependencies on ROMs or classic Mac OS, preserving the software for years to come.

Emacs 31: An unofficial guide to Markdown-ts-mode

https://avatars.githubusercontent.com/u/16169950?s=400&u=e79b2220a9b9edfe6bd2a92e709d63d7bf97be72
Emacs 31 has released a new markdown-ts-mode, an experimental mode that covers the CommonMark and GitHub Flavored Markdown specs. The mode is feature-rich but still evolving, requiring users to opt-in by loading the library manually.

Meta Paid $17B – Gets to Write Safety Rules for Other SocMedia Platform

https://www.techdirt.com/wp-content/uploads/2024/03/ctrl-alt-speech-logo.png
Meta has settled with 52 state and local Attorneys General over child safety on its platforms, paying between $12.7 billion to $18 billion and implementing various changes to improve child safety. The settlement includes features such as time limits, age verification, and content restrictions that may harm free speech under the First Amendment.

Bild AI (YC W25) is hiring product and AI engineers

Bild AI Jobs

Gemini Omni 1.1 Flash

https://storage.googleapis.com/gweb-uniblog-publish-prod/images/Omni_1-1_Flash_hero.width-200.format-webp.webp
Gemini Omni 1.1 Flash introduces creative controls and generative video capabilities for developers, enabling real-world reasoning in generative creation. The new suite supports professional use via the Gemini API in Google AI Studio with improved visual consistency and narrative adherence.

Show HN: Voronoi Go

https://voronoigo.com/og-image.png
Play the ancient game of Go without the grid. Capture territory and best your opponent in a flowing, geometry-driven battle of wits.

Engineered yeast for converting plastic and biomass compounds into food

With billions of people experiencing hunger in 2025 and unable to afford a healthy diet, food insecurity remains a critical global challenge…

Show HN: The load-bearing vocabulary of Claude

Nvidia agrees to acquire Hugging Face for $13B

https://www.businessinsider.com/public/assets/games/games-discovery/discover-sm.png
Nvidia is in talks to acquire Hugging Face, a popular AI platform, for over $13 billion. The deal could give Nvidia a bigger foothold with developers and drive more workloads onto its chips.

The mechanics of the Nepali flash flood

MIT's Ad Hoc Committee on AI Use in Teaching, Learning, and Research Training

MIT's Ad Hoc Committee on AI Use in Teaching, Learning, and Research recommends eight guiding principles to navigate the challenges and opportunities of generative AI. The committee emphasizes the importance of preserving human presence, collaboration, mentorship, and shared intellectual life in residential education.

Aphantasia Beginner's Guide

https://aphantasia.com/_next/image?url=https%3A%2F%2Fmedia.aphantasia.com%2Fimages%2Fwhat-is-aphantasia.jpg&w=2048&q=75
Aphantasia is a condition where people cannot visualize images in their mind's eye, instead thinking conceptually. Those with aphantasia may have different experiences and perceptions of the world compared to those who can vividly imagine.

Launching Route 53 Files

https://www.daemonology.net/blog/generate-IAM-bundle.png
Route 53 Files is a new file system that connects AWS compute resources with Amazon's highest-availability database, allowing seamless editing of DNS records without additional steps. It enables shared access to hosted zones across teams and integrates with AWS Identity and Access Management (IAM) for access control.

Show HN: Watches user sessions, finds bugs that matter, and fixes them

https://raw.githubusercontent.com/opslane/opslane/main/docs/assets/readme/demo-thumbnail.jpg
Opslane is a tool that finds and fixes user-facing bugs in applications, using session recordings to identify issues. It investigates problems based on user impact and verifies fixes before creating a pull request.

Autism mutations drive neurodevelopmental pathology