Scientists capture how cells trigger inflammation (news.stanford.edu)
Researchers report findings on how cells activate inflammatory responses from within, detailing mechanisms that trigger and regulate inflammation at the cellular level. The work, from Stanford News, describes new insights into the intracellular processes involved in immune activation.
Byte-Pair Encoding (en.wikipedia.org) AI
Byte-pair encoding (BPE) is a text encoding method that iteratively merges the most frequent adjacent byte pairs using a learned lookup table, initially described for data compression. A modified form used in large language model tokenizers builds a fixed vocabulary by repeatedly merging frequent token pairs, aiming for practical training rather than maximum compression. Byte-level BPE extends this by encoding text as UTF-8 bytes, allowing it to represent any UTF-8 text.
Notes from from Butterick's Practical Typography (adamadam.blog)
A blogger revisits Robert Bringhurst’s The Elements of Typographic Style through Butterick’s Practical Typography, arguing that good typography should reinforce meaning and that there’s no one-size-fits-all “correct” solution. He draws analogies to theater, highlights practical web typesetting details (soft hyphens, kerning vs. letterspacing, and typographic rendering via SVG when needed), and critiques specific readability choices on his own site. The post also notes design lessons like avoiding distracting table borders and the pitfalls of legacy double spaces.
Show HN: Running local OpenClaw together with remote agents in an open network (github.com) AI
Hybro Hub (hybroai/hybro-hub) is a lightweight daemon that connects locally running A2A agents—like Ollama and OpenClaw—to the hybro.ai portal, letting users use local and cloud agents side by side without switching interfaces. It routes outbound-only connections from the hub to hybro.ai (useful behind NAT), shows whether responses were processed locally or in the cloud, and includes privacy-oriented features like local processing for local-agent requests plus configurable sensitivity detection (currently logging-only). The project provides a CLI to start/stop the hub and launch supported local adapters, with local agents syncing into hybro.ai as they come online.
Deafness reversed: One injection restores hearing in just weeks – ScienceDaily (sciencedaily.com)
Karolinska Institutet researchers report that a single round-window injection of AAV gene therapy delivering a working OTOF gene improved hearing in all 10 patients with congenital genetic deafness. Hearing gains often began within about a month and were clearer after six months, with most participants showing improved sound detection. The treatment was generally well-tolerated over 6–12 months, and the team plans longer follow-up to assess how lasting the effects are.
The White House App Is Riddled with Cybersecurity Vulnerabilities (notus.org)
Cybersecurity researchers and app analysts say the White House’s newly launched “White House App” collects and shares user data—such as IP addresses and time zones—with third-party services without clearly disclosing the extent of that sharing in the app’s store privacy information. They also point to issues involving third-party components like Elfsight widgets and push-notification vendor OneSignal, arguing the disclosures and security approach appear insufficient for a federal app. The White House says the Elfsight vulnerability was reviewed and approved and claims users’ data is safe, while lawmakers and some experts call for stricter standards and faster fixes.
Qwen3.6-Plus: Towards real world agents (qwen.ai) AI
The post introduces Qwen3.6-Plus and discusses how it is being designed to move “real world agents” closer to practical, agent-like performance beyond simple chat.
Show HN: Mtproto.zig – High-performance Telegram proxy with DPI evasion (github.com)
The mtproto.zig GitHub project introduces a high-performance Telegram MTProto proxy written in Zig that disguises traffic as standard HTTPS/TLS 1.3 to evade DPI and censorship. It includes features like TLS/MTProto obfuscation, anti-replay checks, per-user authentication, and multiple TCP/TLS masking techniques. The repo also provides build, test, Docker, and one-line systemd deployment instructions, including optional IPv6 “hop” support via Cloudflare DNS updates.
Maze Algorithms (1997) (astrolog.org)
The article proposes a framework for classifying maze “algorithms” and mazes themselves across seven dimensions: space dimension, topology, cell tessellation, routing rules, passage texture, and how the generator focuses on adding walls versus carving passages. It distinguishes common maze types like 2D/3D and perfect/loop-free mazes, and extends the discussion to hyperdimensional and topologically wrapped layouts. It also breaks down routing and texture properties (e.g., braiding, dead-end patterns, bias/run/elitism, symmetry, and uniformity) and notes that most mazes can be modeled as graphs.
Show HN: Ismcpdead.com – Live dashboard tracking MCP adoption and sentiment (ismcpdead.com) AI
Show HN for ismcpdead.com presents a live dashboard that tracks Model Context Protocol (MCP) adoption and sentiment, aiming to answer whether MCP is “dead.” The site aggregates signals over time to visualize trends in interest and community reactions.
Artemis computer running two instances of MS outlook; they can't figure out why (bsky.app)
A Bluesky post claims that an Artemis spacecraft computer is running two instances of Microsoft Outlook, prompting astronauts to call Houston as NASA prepares to remotely access the system to determine why.
Slap: Functional Concatenative Language with a Borrow Checker? (taylor.town)
The article introduces “Slap,” a small functional, stack-based language combining APL-like terseness with Rust-inspired linear/borrow-checker-style type safety. It explains how Slap prevents type mismatches and unsafe memory behavior using parametric and linear types, plus “stack effects” for type-checking control and data flow. It also shows examples of managed effects for interacting with the host (e.g., for games/demos) and notes the interpreter/compiler is a compact C program with working native and WebAssembly builds.
Lemonade by AMD: a fast and open source local LLM server using GPU and NPU (lemonade-server.ai) AI
Lemonade is an open-source local LLM server that runs on PCs using available GPUs and NPUs, aiming for quick setup and private, local-first AI for text, images, and speech. It supports an OpenAI-compatible API and integrates with a range of apps, with a lightweight native backend and cross-platform availability (Windows, Linux, and macOS beta).
The Last Quiet Thing (terrygodier.com)
The piece contrasts the reliability of a basic Casio F-91W with the constant upkeep demanded by modern smart devices, apps, and subscriptions. It argues that everyday technology has shifted from “done” products to ongoing relationships that offload maintenance and “support work” onto users, reinforced through blame-oriented tools like Screen Time. The author finds the most “luxurious” relief in wearing a watch that collects no data, updates never, and asks nothing.
OpenAI Acquires TBPN (openai.com) AI
OpenAI says it has acquired TBPN, announcing the deal on its website without providing additional article details beyond the acquisition announcement. The post is meant to inform readers about the transaction and its implications.
SQLite in Production: Lessons from Running a Store on a Single File (ultrathink.art)
The article describes running a production e-commerce store with Rails 8 using SQLite across four database files stored on a shared Docker volume, highlighting WAL mode and practical configuration details. It recounts a failure where rapid blue-green deploys overlapped and concurrent writers corrupted the expected order records—payments succeeded but two orders were missing. The author traces the issue to deployment pipeline concurrency on the shared volume, shares debugging tips (including sqlite_sequence), and lists SQLite-specific gotchas before concluding SQLite is viable with careful deploy pacing and moderate write loads.
George Goble has died (legacy.com)
The obituary listing for George Goble announces his death and provides memorial details via Legacy.com, including the date and place of remembrance.
The Axios supply chain attack used individually targeted social engineering (simonwillison.net)
Axios’s postmortem says the recent supply-chain incident was enabled by a tailored social-engineering campaign aimed at a specific maintainer. The attackers impersonated a cloned founder identity, drew the target into a realistic Slack and Teams setup, claimed the target’s system or Teams was outdated, and used a remote access trojan to steal credentials for publishing the malicious dependency.
Jack Dorsey says Block employees now bring prototypes, not slides, to meetings (businessinsider.com)
Block CEO Jack Dorsey said the company has shifted from slide decks to bringing prototypes to meetings, arguing prototypes offer more realism and can be updated in real time. He said the cost of making the wrong decision is becoming smaller, and linked the change to a broader tech trend away from presentations. Dorsey also cited AI-driven efficiency amid Block’s earlier layoffs.
Linux Running in a PDF (2025) (linux.doompdf.dev)
The article claims to show an experimental “Linux running in a PDF” proof-of-concept, embedding code/data inside a PDF that can execute in a PDF viewer environment. The provided text appears to be obfuscated/script-heavy and focused on implementing the logic needed to simulate or bootstrap parts of a Linux-like environment from within the document.
There Is a RAM Shortage (npr.org)
NPR reports on a shortage of computer RAM memory and how it is affecting technology manufacturers, supply chains, and product pricing across the industry. The story outlines the causes behind the tight supply and the practical impacts on computing systems and related businesses.
FAA prohibits SFO's parallel approaches (reuters.com)
The FAA is prohibiting some parallel approaches at San Francisco International Airport (SFO), restricting certain landing operations, according to Reuters. The rule applies to specific landing scenarios at SFO and is intended to address safety and operational concerns.
Astronomers Find a Third Galaxy Missing Its Dark Matter (universetoday.com)
Astronomers report a third ultra-diffuse galaxy, NGC 1052-DF9, that appears to lack dark matter, continuing a pattern seen in earlier objects such as DF2. The team argues the galaxies were likely produced in a “Bullet Dwarf” style collision, where dark-matter halos pass through while normal matter collides and separates into a trailing structure. Further observations are planned to test additional faint galaxies along the same trail.
Iran's Network of Cameras Bolsters Air Defenses, Expert Says (wsj.com)
The report says Iran is strengthening its air defenses by using a network of cameras to improve detection and monitoring, according to an expert cited in coverage tied to the Iran conflict.
Why the Most Valuable Things You Know Are Things You Cannot Say (deadneurons.substack.com)
The article argues that expert judgment can be learned through repeated real-world calibration, but it cannot be fully taught through language because the underlying model depends on high-dimensional inputs and subtle interactions that people may not even perceive until they have experience. It contrasts “book smarts” (knowledge transmissible via explicit rules and frameworks) with “street smarts” (legible outputs but illegible reasoning) and critiques institutions that reward articulacy over predictive accuracy. Overall, it frames the limit as an information-transfer problem and a bootstrapping challenge in developing the perceptual features experts rely on.