Start your morning scandalously informed.
A daily curated digest with the tech news that matter + community vibes, delivered daily, in tabloid style. Like you always wanted.
Sunday, March 1, 2026
AI Cuts War Deals As Phones Lock Down!
Today the tech world stares at power and panic... AI giants shake hands with government while users hunt for exit doors... New deals with a renamed Department of War rattle nerves about where these models will fight next... A loud Cancel ChatGPT drumbeat rolls across the web as trust cracks and people dig up guides on deleting accounts... At the same time, open‑source rebels push local models and fresh forks of core tools, trying to keep control on their own machines... Phone makers quietly strip away recovery menus and sideloading paths, reminding us who really owns the glass in our pockets... Old storage favorites get archived, only to be resurrected overnight by determined coders... In the middle of it all, we see developers torn between turbocharged productivity and a growing fear of long‑term cognitive debt.
AI Marches Into War Rooms And Wallets
Timeline shows AI labs drift into war work
This timeline of Anthropic, OpenAI, and the U.S. state reads like a slow‑motion merger between startup idealism and the security deep state. From classified networks to talk of autonomous weapon systems, the story makes commercial AI look uncomfortably close to the battlefield.
OpenAI confirms pact with Department of War
In a carefully worded post, OpenAI boasts about its new deal with the renamed Department of War, promising a layered “safety stack” while keeping models in its own cloud. Readers see less safety and more mission creep, as a once‑research‑friendly brand leans openly into military AI.
“Cancel ChatGPT” essay turns anger into movement
This blistering piece argues that ChatGPT sits on stolen data, props up surveillance capitalism, and now cozies up to Wall Street and the war machine. The author crowns Anthropic only slightly less guilty, but the real energy is a call to boycott, delete accounts, and walk away from big‑lab AI.
OpenAI staffer fired over prediction market bets
An OpenAI employee allegedly used insider knowledge about Sora for Polymarket bets on Polygon, and got fired for it. The story feels like a tiny Wall Street scandal transplanted into an AI lab, reinforcing the sense that these companies now juggle hype cycles, trading games, and public trust all at once.
Techno‑feudal nightmare warns of AI police state
This furious essay paints billionaires, surveillance tech, and militarized agencies like DHS and ICE as architects of a twenty‑first‑century fascist state. AI is cast as the perfect tool to automate control, from camps to cameras, and many readers nod along even as the rhetoric goes off the charts.
Coders Chase AI Speed, Dread Brain Overdraft
Cognitive debt explains why fast teams feel lost
This piece nails the feeling that AI‑assisted teams ship features like crazy yet can’t remember how anything works six months later. It calls that gap cognitive debt, and the examples of immaculate metrics hiding fragile systems ring painfully true for engineers stuck babysitting “successful” projects.
Essay asks what AI coding really costs us
Here AI coding tools are framed as a spectrum from simple autocomplete to full agents quietly writing entire features. The author loves the speed yet fears skill rot, shallow understanding, and weaker engineering culture. It reads like a confession from someone who can’t put Copilot down but doesn’t trust it either.
HN poll shows devs hooked on AI helpers
A veteran dev admits they feel merely average but super‑charged by AI, and the comments show many others feel the same rush. Some brag about shipping at new speeds; others worry they’re turning into prompt typists. The thread captures a community that loves the power and fears the tradeoff.
Enterprise devs doubt Copilot’s value at work
In this discussion, corporate coders describe GitHub Copilot as noisy, often wrong, and weirdly pushy with keybindings, even as management treats it like magic productivity dust. The mood is wary: people want AI that truly understands their codebase, not just one more subscription humming in the background.
Engineers feel everything changes yet nothing changes
This reflection on LLMs and agents says software is shifting from craft to mass production. The author imagines future teams where specs, tests, and AI do the heavy lifting while humans supervise. It’s both exciting and bleak, capturing that eerie sense that our jobs are transforming in place.
Open Rebels Push Local AI As Giants Lock Down
MinIO archived, fast fork keeps clouds afloat
When MinIO Inc. archived its popular S3‑compatible server, users panicked about a cornerstone of self‑hosted storage going dark. A community fork quickly revived the code, restored the admin console, and rebuilt binaries, showcasing how critical open infrastructure never really dies if enough people depend on it.
Alibaba’s Qwen models rival Sonnet on local rigs
VentureBeat reports that Qwen3.5‑35B and 122B match Claude Sonnet 4.5 on benchmarks while running on decent local GPUs. For power users tired of metered APIs and data‑sharing fears, these open AI models feel like a serious shot across the bow of the closed giants.
Tiny microgpt script teaches DIY model building
This art project packs a full toy GPT training and inference loop into about 200 lines of pure Python. It won’t replace big models, but it demystifies how they tick, letting curious hackers peek under the hood instead of treating large language models as untouchable black boxes.
AMD demo runs trillion‑parameter model at home
An AMD Ryzen AI Max+ cluster driving a trillion‑parameter LLM sounds like sci‑fi, but their showcase claims it’s real. Even if it’s tightly tuned marketing, the message is clear: monstrous models are creeping out of hyperscale data centers and into small labs and prosumer closets.
Samsung update strips Android recovery features away
New Galaxy firmware quietly removes recovery menu tools like sideloading and full factory reset options. Power users see another brick in the walled garden, where vendors control bootloaders, updates, and apps while customers just rent shiny glass slabs that are hostile to real ownership.
Top Stories
AI labs cozy up to the Pentagon
A detailed timeline lays out how Anthropic and OpenAI are moving from lab darlings to defense contractors, including work on autonomous weapons and classified military networks, shocking readers who still saw them as neutral research outfits.
OpenAI signs deal with Department of War
OpenAI’s own post confirms a contract with the newly named Department of War, promising a safety stack while still keeping models cloud‑hosted for the military, fueling fears that flagship AI tools are sliding straight into warfare.
"Cancel ChatGPT" backlash hits the mainstream
A long, angry essay argues that after OpenAI’s Dow deal and growing state ties, ChatGPT has become the villain of the AI race, pushing a full‑on cancellation campaign and channeling the community’s growing distrust of big labs.
MinIO archived, community forks it overnight
S3‑compatible storage workhorse MinIO is abruptly archived by its company, but an independent fork springs up immediately with binaries and a restored console, turning a corporate shutdown into a grassroots fight for critical infra.
Alibaba drops near‑Sonnet open models for home rigs
Alibaba’s Qwen3.5 35B and 122B models reportedly match Claude Sonnet 4.5 on many tasks while running on local hardware, giving power users and companies a serious open alternative to pricey closed APIs.
One‑trillion parameter model runs on AMD cluster
AMD shows off a Ryzen AI Max+ mini‑cluster running a trillion‑parameter model locally, signaling that yesterday’s sci‑fi‑scale LLMs are creeping into the living room and small labs instead of staying locked in hyperscale data centers.
Samsung quietly kills key Android recovery tools
New Samsung Galaxy updates strip recovery menu options like sideloading from stock firmware, setting off alarms among power users who see it as another step toward phones you buy but never truly control.
Semantic Syntax Highlighting for Lisp in Emacs
by oumua_don17The article announces lisp-semantic-hl, a semantic syntax highlighting minor mode for Emacs that supports Common Lisp and Emacs Lisp. Unlike traditional token-based highlighters, it uses information f...
Key Points
- Semantic highlighting for Common Lisp and Emacs Lisp in Emacs via lisp-semantic-hl.
- MELPA/use-package setup with hooks for emacs-lisp-mode and lisp-mode.
- Goals: lexical-environment-based highlighting using CLtL2 APIs and better macro support.
How do I cancel my ChatGPT subscription?
by tobrThis article outlines how to cancel ChatGPT subscriptions across different tiers and purchase channels. Personal subscribers who signed up on chatgpt.com can cancel by navigating to Settings, selectin...
Key Points
- Website cancellations: Settings > Account > Manage > Cancel Subscription
- Business cancellations: Workspace settings > Billing > Manage plan > Cancel subscription
- Cancel at least 24 hours before the next billing date; fees are non-refundable
Package Managers à la Carte: a formal model of dependency resolution
by avsmThe article presents Package Calculus, a formal model intended to unify how diverse package managers perform dependency resolution. It is motivated by the fragmentation that arises when programming la...
Key Points
- Introduces Package Calculus as a unifying formalism for dependency resolution.
- Demonstrates expressiveness via formal reductions modeling real-world package managers.
- Enables cross-ecosystem translation and fuller security visibility using an intermediate representation.
Rust Is Just a Tool
by JuniperMesosThis opinion piece argues for a pragmatic stance toward Rust and programming languages in general. The author commends Rust for its versatility across application and systems domains, its high-quality...
Key Points
- Rust is praised for versatility, tooling, and a strong type system without relying on garbage collection.
- Using Rust does not obligate adherence to community dogma, marketing, or all popular crates.
- Developers should respect alternative languages and acknowledge Rust’s trade-offs and complexity.
US and Israel carrying out strikes against Iran
by simonebrunozziThe article details the onset of coordinated US-Israeli strikes on Iran following months of joint planning, with Israel asserting that targets included top Iranian leadership. Sources cited indicate t...
Key Points
- US and Israel launched coordinated strikes on Iran, with Israel citing senior leadership among targets.
- Trump framed the campaign as large-scale and long-term, warning of possible US casualties and urging regime change.
- Iran retaliated regionally, with explosions reported in the UAE, Bahrain, and Qatar.
U.S. and Israel Conduct Strikes on Iran
by gammaratorThe New York Times live blog reports that the United States has launched strikes on Iran, with President Donald J. Trump declaring the operation massive and ongoing. Israel said it joined the attack. ...
Key Points
- U.S. strikes on Iran announced by President Trump; Israel says it joined the operation.
- Tehran experienced explosions; satellite imagery shows damage at Khamenei’s compound.
- Iran launched missiles toward Israel, prompting emergency measures in Israeli cities.
Inferring Car Movement Patterns from Passive TPMS Measurements
by wisdomseakerThis article reports on a measurement study of Tire Pressure Monitoring System (TPMS) transmissions that demonstrates significant privacy risks for drivers. Modern TPMS messages are broadcast in clear...
Key Points
- TPMS broadcasts with persistent IDs enable long-term vehicle tracking.
- A 10-week roadside study observed 20k+ cars and analyzed 12 verified vehicles.
- Low-cost ($100) receivers make large-scale tracking feasible, prompting calls for secure TPMS redesigns.
The Eternal Promise: A History of Attempts to Eliminate Programmers
by dinvladThis article examines the recurring promise in the software industry to simplify or eliminate programming, linking today’s excitement about code-generating AI and large language models to earlier move...
Key Points
- COBOL’s goal to empower business users did not eliminate programmers; it created specialized roles.
- COBOL remains foundational in critical sectors with ongoing demand for expertise.
- Current AI claims echo 1960s–70s AI optimism, as noted by Simon and Minsky’s predictions.
Israel and the U.S. launch strikes against Iran
by 0inAccording to the article, the United States and Israel launched a coordinated military operation against Iran, described by President Donald Trump and Israeli Prime Minister Benjamin Netanyahu as an e...
Key Points
- U.S. and Israel launched “Epic Fury” targeting Iran’s leadership and missile program, per official statements.
- Iran reported casualties and said it retaliated with missiles against Israel and U.S. bases in several Gulf states.
- Israel enacted emergency measures, including airspace closure and civil defense activation, amid active missile interceptions.
Unsloth Dynamic 2.0 GGUFs
by toshUnsloth’s Dynamic v2.0 introduces a more granular, model-aware quantization approach for GGUF and safetensors that aims to preserve accuracy while enabling efficient local deployment and fine-tuning. ...
Key Points
- Dynamic v2.0 applies intelligent, per-layer, model-specific quantization across MoE and non-MoE models.
- New quant formats (Q4_NL, Q5.1, Q5.0, Q4.1, Q4.0) target efficiency on Apple Silicon/ARM.
- Evaluation emphasizes KL Divergence and ‘flips,’ with calibrated, fair benchmarking practices.
The man building Team USA's Olympic bobsleds
by wrsh07Marc Van Den Berg, director of technology and equipment for USA Bobsled/Skeleton, has hand-built the bobsleds that Team USA will race at the 2026 Winter Olympics. Working solo from a workshop at the M...
Key Points
- A single engineer built Team USA’s Olympic bobsleds in Lake Placid over two years.
- Ten sleds were air-shipped from New York to Europe for multiple men’s and women’s events.
- Van Den Berg kept designs secret to avoid copying and is in Italy with the team as events begin.
A Fuzzer for the Toy Optimizer
by chunklesThis post presents a pragmatic approach to validating a toy load/store optimizer using fuzz testing and an interpreter-based oracle. The author generates random programs composed of loads, stores, and...
Key Points
- Fuzzer + oracle checks heap-state equivalence pre/post optimization.
- Programs randomly combine loads, stores, and escape ops on three args.
- Verification covers no-alias and full-alias scenarios to catch bugs.
OpenAI – How to delete your account
by carlosrgOpenAI’s help article provides step-by-step guidance for permanently deleting a ChatGPT account and outlines what happens to data and subscriptions afterward. Users can delete accounts either by submi...
Key Points
- Account deletion is permanent and triggers data deletion within 30 days.
- App Store/Google Play subscriptions must be canceled separately.
- Two deletion methods: Privacy Portal or self-serve in ChatGPT settings.
SHELL: Global Tool for Calling and Chaining Procedures in the System (1965) [pdf]
by NaOHThis 1965 MIT Project MAC memorandum by Louis Pouzin outlines a system-wide “SHELL” intended to standardize the calling and chaining of procedures. It defines the fundamental behavior of commands invo...
Key Points
- Defines console-invoked command conventions and argument parsing.
- Stresses user-oriented design with clear, edited console outputs.
- Requires command-specific decoding and conversion of printable BCD inputs.
Tell HN: 3 months ago we feared AI was useless. Now we fear it will take our job
by giuliomagnificoA Tell HN post details a swift change in market sentiment surrounding artificial intelligence. Two to three months prior, investors were anxious that massive AI-related capital expenditures by technol...
Key Points
- Market fears have shifted from AI CAPEX being wasteful to AI displacing jobs.
- The software and services sector is cited as down roughly 20% year-to-date via an S&P index.
- The author references a WSJ podcast and questions what could trigger the next drop.
Show HN: SplatHash – A lightweight alternative to BlurHash and ThumbHash
by unsorted2270SplatHash is introduced as a compact image placeholder algorithm that compresses any image into exactly 16 bytes and reconstructs a 32×32 blurred preview. It provides Go (reference), TypeScript/JavaSc...
Key Points
- Encodes images into fixed 16-byte hashes with 0.067 ms decode to 32×32 previews.
- Bit-exact implementations across Go, TypeScript/JS, and Python.
- Uses Oklab, Gaussian basis, and ridge regression; supports alpha and fixed-size output.
The Future of AI
by BerislavLopacThis essay, adapted from a February 25, 2026 talk in London, examines two linked challenges in contemporary AI: moral alignment and information reliability. It proposes the “Parents’ Paradox,” contras...
Key Points
- Labeled deepfakes still influenced viewers in a Nature study (Jan 2026).
- Labeling alone may be insufficient to counter AI-driven misinformation.
- Feedback loops from training on inaccurate user data risk degrading ground truth.
More Cows, More Wives
by oxwThis article explores cross-cultural diversity in marriage through a detailed case study and comparative ethnographic evidence. It begins with E.E. Evans-Pritchard’s 1930s fieldwork among the Nuer in ...
Key Points
- Nuer “ghost marriage” preserves lineage by attributing children to a deceased man.
- Anthropological accounts from Samoa and North-Western Melanesia highlight diverse courtship and marriage norms.
- Cross-cultural data show polygyny predominates across many societies, with monogamy less common globally.
Don't trust AI agents
by gronky_Gavriel Cohen argues that AI agents should be treated as untrusted and potentially malicious, and that security should be achieved through architecture, not just permissions or prompts. He contrasts O...
Key Points
- Per-agent, ephemeral container isolation is central to NanoClaw’s design.
- Defense-in-depth includes an external mount allowlist and read-only host mounts.
- OpenClaw’s defaults and complexity are presented as risk factors.
Better Activation Functions for NNUE
by luuThe article reports on replacing squared clipped ReLU (SCReLU) with Swish in the intermediate layers (L₁ and L₂) of Viridithas 19’s NNUE chess engine network. Initially implemented via Hard-Swish (β=1...
Key Points
- Replacing SCReLU with Swish improved strength but initially reduced sparsity.
- L1 activation penalty recovered block-sparsity and inference performance.
- Swish produced smoother evaluation distributions and Elo gains over baseline.
Show HN: Now I Get It – Translate scientific papers into interactive webpages
by jbdamask“Now I Get It” is a web tool designed to make scientific literature easier to understand and share. Users upload a scientific PDF and receive a plain-language, interactive webpage summarizing the mate...
Key Points
- Uploads of scientific PDFs are transformed into interactive, plain-language webpages.
- A gallery highlights examples across AI, security, cryptanalysis, clinical guidance, and historical philosophy.
- Entries display titles, authors, and dates, illustrating the tool’s breadth and source coverage.
Customer Update on Simplenote
by 0inSimplenote has released a customer update indicating a shift in its product lifecycle. The company states that the app is no longer in active development, signaling the end of ongoing feature work and...
Key Points
- Simplenote ends active development but remains available.
- Only basic functionality will be maintained; no new features planned.
- Support is provided via the public support forum.
Show HN: Decided to play god this morning, so I built an agent civilisation
by uravWerld is an open-source, local-first simulation that models open-ended evolution of autonomous agents on a Watts–Strogatz small-world network. Agents are driven by NEAT-style neural networks with evol...
Key Points
- Local, pure-Python simulation with NEAT-style evolving neural networks on a small-world graph
- Next.js dashboard (Werld Observatory) polling SQLite for real-time visualization
- Robust runtime features: checkpoints, resume, bounded ticks, watchdog, and SIGTERM-safe handling
Addressing Antigravity Bans and Reinstating Access
by RyanShookGoogle announced it is restoring access for users impacted by recent Antigravity Terms of Service enforcement that caused bans to spill over to Gemini CLI and Gemini Code Assist due to backend abuse-p...
Key Points
- Automated unban will restore access within 1–2 days for recently flagged accounts.
- New self-service reinstatement requires recertification via a Google Form.
- Second ToS violation results in a permanent ban; OAuth piggybacking is prohibited.
Please do not use auto-scrolling content on the web and in applications
by speckxThe article highlights accessibility risks of auto-scrolling content in mobile apps and on the web. During a native Android app audit, enabling the TalkBack screen reader produced incessant ticking ea...
Key Points
- Auto-scrolling triggers excessive audio cues in Android TalkBack, hindering usability.
- The issue appears in both native Android apps and web pages using JavaScript scrollLeft.
- VoiceOver on iOS did not show the same behavior in the tests; avoid or control auto-scrolling.
The Life Cycle of Money
by nanacnoteThis explainer builds an ontology of modern money by treating it as a legal and institutional relationship recorded on balance sheets rather than a physical commodity. It classifies money into base mo...
Key Points
- Money is a balance-sheet relationship, not a commodity.
- Modern currencies are fiat and rely on legal acceptance and network effects.
- The Federal Reserve issues the U.S. monetary base under the 1913 Act.
Everything Changes, and Nothing Changes
by todsacerdotiThe article asserts that large language models are rapidly reshaping software engineering by making code generation abundant and inexpensive, shifting the discipline from craft-driven coding toward ma...
Key Points
- AI could write most code, with some engineers already not coding manually in leading labs.
- Core practices—stacked diffs, CI/CD, automated tests, rollbacks—grow in importance with AI-generated code.
- Engineering value shifts toward architecture and constraints; guidance like AGENTS.md helps models produce cleaner code.
OpenAI Fires an Employee for Prediction Market Insider Trading
by bookofjoeOpenAI dismissed an employee after an internal investigation found they used confidential company information to trade on prediction markets, including Polymarket. The decision, shared in an internal ...
Key Points
- OpenAI fired an employee for using confidential data on prediction markets.
- Unusual Whales flagged clustered, suspicious trades around OpenAI events.
- Polymarket’s Polygon-based ledger is traceable despite pseudonymity.
Timeline: Anthropic, OpenAI, and U.S. Government
by vldsznThis timeline captures a rapid sequence of developments involving Anthropic, OpenAI, and the U.S. government in late February 2026. After months of discussions with the Department of War, Anthropic re...
Key Points
- U.S. blacklists Anthropic as a defense “supply chain risk,” barring it from contracts and triggering a six-month transition.
- Anthropic refuses mass domestic surveillance and fully autonomous weapons, plans to challenge the designation; commercial access unaffected.
- OpenAI raises $110B and agrees to deploy models on Pentagon classified networks with stated safety red lines.
Show HN: Gitcredits – movie-style end credits for any Git repo in your terminal
by swq115Gitcredits is a lightweight command-line tool that generates movie-style end credits for any Git repository directly in the terminal. After installation via Go (go install github.com/Higangssh/gitcred...
Key Points
- Terminal tool displays movie-style credits for Git repositories.
- Optional GitHub CLI integration adds stars, description, language, and license.
- Installs via Go or source; requires Git and Go 1.21+, MIT-licensed.
Latency numbers every programmer should know
by ksecThis console-based reference compiles approximate latency figures for common operations across computing stacks, presented as a port of “Jeff Dean’s latency numbers.” Hosted on GitHub (chubin/late.nz)...
Key Points
- Console port of Jeff Dean’s latency numbers (GitHub: chubin/late.nz), referencing colin-scott’s interactive resource.
- Nanosecond-scale CPU/memory operations vs. microsecond/millisecond storage and network latencies.
- Concrete examples across SSD/HDD storage and intra-datacenter vs. intercontinental network round trips.
No Bookmarks
by herbertlIn this personal essay, writer Niklas Göke explains his choice to read without bookmarks, a habit that began when he once went to bed without one and later had no trouble finding his place. He notes t...
Key Points
- Reading without bookmarks can reinforce attention and memory.
- Occasional difficulty re-finding a place becomes a deliberate memory exercise.
- Self-discovered rituals can be more rewarding than prescribed methods.
What AI coding costs you
by tomwojcikThis piece charts the progression of AI-assisted coding from early integrated tools to agent-driven workflows and more capable models. It explains how first-wave assistants like Cursor and Copilot, po...
Key Points
- Early AI tools reduced search overhead by indexing codebases with RAG.
- Agent workflows increased autonomy but introduced reliability issues.
- Improved models (e.g., Opus 4.5) made AI-driven workflows more dependable.
Woxi: Wolfram Mathematica Reimplementation in Rust
by adamnemecekWoxi is a Rust-powered interpreter for the Wolfram Language designed to make scripting and notebook-based workflows accessible without the overhead of traditional tooling. The project focuses on imple...
Key Points
- Rust-based interpreter for Wolfram Language focused on CLI and Jupyter.
- Full Jupyter support with an in-browser JupyterLite instance.
- Claims faster execution than WolframScript due to no kernel/license overhead.
From Noise to Image – interactive guide to diffusion
by simedwThis interactive guide explains how diffusion-based text-to-image systems transform random noise into coherent images aligned with a user’s prompt. It outlines the enormity of the image search space a...
Key Points
- Diffusion models denoise from random noise to produce images aligned with text prompts.
- Generation occurs in latent space with an encoder/decoder mapping to pixel images.
- Random seeds, step count, and prompt detail significantly affect output and runtime.
Is GitHub Copilot still relevant in the enterprise?
by AznHisokaThis article offers a user’s account of GitHub Copilot in an enterprise where it is the sole approved AI tool. The author reports minimal reliance on Copilot after disabling its auto-suggestions, citi...
Key Points
- GitHub Copilot is the only approved AI tool in the author’s workplace.
- Auto-suggestions were disabled due to inaccuracies and troublesome Tab keybinding behavior.
- Copilot’s Agent mode caused unintended code changes, prompting rollback via commit history.
Cognitive Debt: When Velocity Exceeds Comprehension
by pagadeThis article introduces the concept of “cognitive debt,” explaining how AI-assisted development can accelerate code production beyond engineers’ capacity to understand what’s being built. It contrasts...
Key Points
- AI-accelerated production can outpace comprehension, creating “cognitive debt.”
- Traditional output metrics hide comprehension deficits; reliability metrics reveal them later.
- AI inverts code review throughput, pressuring review depth and quality.
747s and Coding Agents
by cckolonA developer reflects on how AI coding agents have reshaped their workflow and learning. Framed by an earlier conversation with a veteran 747 pilot who felt his job offered little room for further impr...
Key Points
- AI coding agents now implement end-to-end features with minimal human input.
- This shift increases productivity but slows developers’ skill growth.
- Intervening on agent-generated code can be harder than building solutions incrementally.
Stop Burning Your Context Window – How We Cut MCP Output by 98% in Claude Code
by mksgluContext Mode targets the often-overlooked cause of context-window bloat in Claude Code: raw outputs from external tools. As an MCP server, it runs each tool call in an isolated subprocess and only adm...
Key Points
- Up to 98% reduction in tool-output size via subprocess stdout-only routing.
- SQLite FTS5 + BM25 + Porter stemming return exact, relevant code blocks.
- Extended session duration (~3 hours) and minimal context usage after 45 minutes.
A Man Who Stole Infinity
by rbanffyQuanta Magazine reports on a discovery that may revise the legacy of Georg Cantor, whose 1874 paper established that infinities can vary in size and reshaped modern mathematics. Mathematician-journali...
Key Points
- A newly surfaced 1873 letter tied to Cantor is presented as evidence of plagiarism behind his 1874 infinity paper.
- Demian Goos located the letter while reviewing Cantor’s estate at the University of Halle with Karin Richter’s help.
- Cantor’s 1874 work profoundly reshaped mathematics by rigorously confronting the concept of infinity.
Obsidian Sync now has a headless client
by adilmoujahidObsidian introduced a headless client for Obsidian Sync, available in open beta, enabling command-line synchronization of Obsidian vaults without the desktop application. Designed for CI pipelines, ag...
Key Points
- Open-beta headless client enables CLI-based Obsidian Sync without the desktop app.
- Requires an active Obsidian Sync subscription and careful backup/conflict-avoidance practices.
- CLI supports listing vaults, creating remotes with encryption options, and setting up local-to-remote sync.
Verified Spec-Driven Development (VSDD)
by todsacerdotiThe article introduces Verified Spec-Driven Development (VSDD), an AI-orchestrated methodology that unifies Spec-Driven Development, Test-Driven Development, and Verification-Driven Development into a...
Key Points
- VSDD merges SDD, TDD, and VDD into a single, AI-led pipeline.
- Defined roles include Architect, Builder (e.g., Claude), Tracker (Chainlink), and Adversary (e.g., Gemini-based Sarcasmotron).
- Phase 1 specifies rigorous contracts and a verification architecture, including formal properties and a purity boundary.
Show HN: SQLite for Rivet Actors – one database per agent, tenant, or document
by NathanFlurryRivet introduces a serverless, stateful actor model that packages compute, state, and operational tooling into isolated “actors.” Each actor includes in-memory state, persistent storage using SQLite o...
Key Points
- Stateful, serverless actors with SQLite/JSON persistence and real-time/WebSocket support
- Self-host, managed cloud, and open-source deployment paths with broad ecosystem integrations
- SDKs, engine components, and community resources for building and operating at scale
Ghosts'n Goblins – "Worse danger is ahead"
by elvis70This article examines the origins of Famitsu, launched by LOGiN in 1986 with an initial focus on Nintendo’s Famicom, and its long-running role in publishing national sales charts. Leveraging these cha...
Key Points
- Famitsu’s early charts and UK rankings show Ghosts’n Goblins topping both markets simultaneously in 1986.
- The UK’s 1986 game market was computer-led, while Japan/US were console-led by the Famicom/NES.
- Arcades linked markets internationally; non-Nintendo publishers like Capcom enabled broader home conversions.
Seeing Like a Sedan
by surprisetalkThe article contrasts two competing strategies for autonomous ride-hailing. Waymo’s robotaxis use lidar, radar, and cameras to construct a detailed 3D understanding of the environment, exemplifying a ...
Key Points
- Waymo’s multi-sensor stack contrasts with Tesla’s vision-only Cybercab approach.
- Waymo is live in several cities; Tesla is piloting in Austin with Model Y and safety monitors.
- Market size and per-vehicle cost will heavily influence robotaxi scaling.
The whole thing was a scam
by guilamuAI researcher Gary Marcus contends that OpenAI’s Sam Altman was privately negotiating a government AI agreement that displaced Anthropic even as he publicly supported Anthropic CEO Dario Amodei. Citin...
Key Points
- Author cites NYT reporting that Altman worked on the deal before his public support for Amodei.
- Claims the U.S. government rejected Anthropic and labeled it a supply chain risk while accepting similar terms elsewhere.
- Notes Greg Brockman’s $25M donation to Trump’s PAC preceding the deal timeline.
CSP for Pentesters: Understanding the Fundamentals
by zdwThis article provides a practical overview of Content Security Policy (CSP) aimed at pentesters and web security practitioners. Inspired by a CTF incident where an XSS payload was blocked by CSP, the ...
Key Points
- CSP enforces allow-listed sources for resources via browser-applied policies.
- Directive fallback behavior (to default-src) is a common source of misconfigurations.
- script-src, object-src, and base-uri require careful configuration to prevent bypasses.
Werner Herzog Between Fact and Fiction
by HookeThis piece analyzes Werner Herzog’s latest articulation of “ecstatic truth,” a concept he has promoted for decades to describe a poetic, non-literal truth in cinema. Opening with a 1970/71 episode fro...
Key Points
- Herzog’s “ecstatic truth” embraces fabrication to reach deeper cinematic truths.
- A 1970/71 scene from The Land of Silence and Darkness exemplifies his approach.
- The article questions how to distinguish artistic license from manipulation in the era of deepfakes and AI.
How Long Is the Coast of Britain? (1967)
by HookeA 1967 article in Science examines the challenges of defining the length of highly irregular geographical curves, such as coastlines. It explains that as measurement resolution increases, the measured...
Key Points
- Irregular coastlines can yield undefinable lengths as measurement resolution increases.
- Statistical self-similarity enables characterization of such curves via scaled parts of the whole.
- A fractional dimension D (>1) quantifies the complexity of these curves.
Show HN: Tomoshibi – A writing app where your words fade by firelight
by hakumeiTomoshibi is a distraction-free writing app that encourages forward progress by letting earlier lines slowly fade from view while preserving all text behind the scenes. Presented in English and Japane...
Key Points
- Earlier lines fade visually but are saved locally, promoting forward writing without losing content.
- Minimal interface with a breathing flame and four themed modes supports a focused ambiance.
- No account is needed; a Mac app is planned in addition to the browser version.
The happiest I've ever been
by bewal416This personal account details how the author, feeling unfulfilled in a first post-college job, found meaning by becoming a middle school youth basketball head coach in Indiana in early 2020. After a c...
Key Points
- Head coaching role emerged unexpectedly and led to a structured, successful season.
- Individual player development and positive team culture were central goals.
- Planned Pacers court experience was canceled due to the COVID-19 shutdown.
Techno‑Feudal Elite Are Attempting to Build a Twenty‑First‑Century Fascist State
by measurablefuncThis essay presents a critical analysis of the United States’ political trajectory, arguing that the country is moving toward a 21st‑century form of authoritarianism led by an oligarchic, “techno‑feud...
Key Points
- The essay argues neoliberalism enabled a shift of power to private oligarchic networks.
- Big Tech platforms are characterized as rent‑extracting and central to surveillance/disinformation.
- Cites accounts linking DHS/ICE to plans for expanded detention infrastructure in the U.S.
"Cancel ChatGPT" movement goes mainstream after OpenAI closes deal with U.S. Dow
by AndrewKemendoThis article portrays a split in the AI industry’s approach to government use. It claims Anthropic, maker of the Claude AI, was designated a supply chain risk and banned from use by U.S. governmental ...
Key Points
- Article claims Anthropic was banned from U.S. government use after opposing certain deployments.
- OpenAI, under Sam Altman, reportedly pledged ChatGPT and other tech to U.S. government needs.
- Backlash is described, alongside a claimed $730B valuation round for ChatGPT with major investors.
'Play like a dog biting God's feet': Steven Isserlis on György Kurtág at 100
by mitchbobCellist Steven Isserlis reflects on his four-decade relationship with Hungarian composer György Kurtág, coinciding with Kurtág’s centenary. He recounts their first meeting at the International Musicia...
Key Points
- Isserlis details decades of coaching with György Kurtág, beginning at Prussia Cove.
- Kurtág’s methods employ imagery, tonal focus, and the structural role of silence.
- Specific works (“Gérard de Nerval,” “Schatten”) and tools (hotel mute) exemplify these practices.
Our Agreement with the Department of War
by surprisetalkOpenAI announced it has reached an agreement with the U.S. Department of War (Pentagon) to deploy advanced AI systems in classified environments. The company asserts the agreement contains stronger sa...
Key Points
- Agreement with the Pentagon to deploy AI in classified environments with strict guardrails.
- Cloud-only architecture with OpenAI-controlled safety stack and verification via classifiers.
- Contract limits independent use for autonomous weapons and high-stakes automated decisions requiring human oversight.
Block the "Upgrade to Tahoe" Alerts
by todsacerdotiThe article details how to prevent recurring “Upgrade to Tahoe” notifications when running macOS Sequoia, using macOS configuration profiles to defer major update activities for 90-day intervals. It n...
Key Points
- Use a configuration profile to defer major macOS updates for 90 days and suppress Tahoe upgrade prompts.
- A macOS 15.7.3 bug makes the 90-day deferral act as a rolling period.
- The Stop Tahoe Update project provides a mobileconfig profile and steps to install it via System Settings.
Qwen3.5 122B and 35B models offer Sonnet 4.5 performance on local computers
by lostmsuAlibaba’s Qwen AI team unveiled the Qwen3.5 Medium Model series—a set of four large language models with agentic tool calling—aimed at bringing high performance to local machines. Three models (Qwen3....
Key Points
- Three Qwen3.5 models are open source (Apache 2.0) and downloadable; a fourth is proprietary via Alibaba Cloud API.
- Benchmark claims indicate Qwen3.5 beats GPT-5-mini and Claude Sonnet 4.5 on third-party tests.
- Quantization enables >1M-token context on 32GB consumer GPUs; architecture uses Gated Delta Networks + sparse MoE.
MinIO Is Dead, Long Live MinIO
by zufallsheldThe article reports that MinIO’s open-source GitHub repository, a widely used object storage project with 60k stars and over a billion Docker pulls, was archived on February 12, 2026, following a Dece...
Key Points
- MinIO’s repo was archived on Feb 12, 2026, after maintenance mode in Dec 2025.
- A fork (pgsty/minio) restores the admin console and distribution pipelines, with CVE fixes.
- AGPL v3 ensures the community’s right to fork and maintain the project.
The archivist preserving decaying floppy disks
by BrajeshwarThe article highlights the urgent need to preserve data stored on aging floppy disks, once the backbone of personal and institutional computing. With tens of billions produced across varied sizes and ...
Key Points
- Archivist Leontien Talboom is imaging diverse floppy formats to rescue historical data.
- “Copy That Floppy!” provides a practical preservation guide under the Future Nostalgia project.
- Legacy reliance on floppies persisted in critical sectors well into recent years.
Iran's Ayatollah Ali Khamenei is killed in Israeli strike, ending 36-year rule
by andsoitisThe article states that Iran’s Supreme Leader Ayatollah Ali Khamenei has been killed in Israeli strikes carried out with U.S. support. It says President Donald Trump announced the death on social medi...
Key Points
- Article reports Khamenei’s death in Israeli strikes with U.S. support, announced by Donald Trump and cited via NPR.
- Khamenei’s rise and consolidation of power through the IRGC and strategic deterrence mechanisms are detailed.
- Domestic interventions and crackdowns, including the 2009 election and reported protest casualties, are recounted.
Laravel Inertia Toast
by veekthovenLaravel Inertia Toast is a package that adds customizable toast notifications to Laravel + Inertia.js applications, offering a unified experience across backend and frontend. It features a fluent PHP ...
Key Points
- Provides unified toast notifications for Laravel + Inertia.js across backend and frontend.
- Supports Vue 3 and React adapters with Tailwind CSS (v3/v4) and TypeScript.
- Includes clear installation and setup steps for Composer and npm, plus Tailwind configuration.
The war against PDFs is heating up
by pseudolusThe article recounts the origin and rise of the Portable Document Format (PDF), introduced by Adobe in 1993. Initially dismissed by a Gartner consultant as “the dumbest idea,” the format faced practic...
Key Points
- Adobe launched PDF in 1993 despite skepticism and internal doubts.
- IRS adoption of PDFs helped cement the format’s widespread use.
- The article questions PDF’s durability amid the AI revolution.
Poll: Code with AI or Not?
by bitbasherA seasoned software developer describes how AI has transformed their workflow, functioning more like a mentor than a junior assistant. They report using AI in assistant mode to review code and help wi...
Key Points
- AI serves as a mentor-like assistant for code review and front-end tasks.
- Pattern-based architectures (CRUD, MVC, layered layers) suit AI-driven automation.
- Proposed agent orchestration would produce PRs for human review and merge.
Deterministic Programming with LLMs
by todsacerdotiThis essay explores how large language models (LLMs) can be used in more deterministic ways, drawing practical lessons from mathematics. It notes that while LLMs can produce text resembling mathematic...
Key Points
- LLMs are powerful but require verification; formal systems like Lean provide deterministic checks.
- A reported 2026 pipeline used ChatGPT and Aristotle to produce a Lean-verified solution to an Erdős problem.
- Deterministic approaches in programming are encouraged, with coding agents cited as context.
Pentagon chief blocks officers from Ivy League schools and top universities
by geoxDefense Secretary Pete Hegseth has moved to restructure professional military education by eliminating certain Senior Service College fellowship programs at a set of universities starting with the 202...
Key Points
- Memo eliminates certain SSC fellowships at named universities from 2026–2027 onward.
- New partner institutions proposed include Liberty University and several public/private universities.
- Existing Army and Space Force partnerships at affected schools face uncertain impact.
Show HN: Xmloxide – an agent made rust replacement for libxml2
by jawigginsXmloxide is introduced as a pure Rust reimplementation of libxml2, designed to deliver memory safety and high performance while maintaining broad compatibility with existing XML/HTML workflows. It rep...
Key Points
- Pure Rust, memory-safe replacement for libxml2 with full W3C XML test conformance.
- Comprehensive feature set: multiple parsing APIs, error recovery, HTML 4.01, XPath 1.0, validation, C14N, XInclude.
- Tooling and integration: xmllint CLI, C/C++ FFI, minimal dependencies (encoding_rs; clap for CLI).
H-Bomb: A Frank Lloyd Wright Typographic Mystery
by mrngmThis article investigates a typographic oddity at Frank Lloyd Wright’s Unity Temple in Oak Park, Illinois: an inverted “H” in the bronze entrance lettering above the west doors. The building, opened i...
Key Points
- The west entrance shows an inverted “H” in the bronze inscription.
- Both entrance signs were removed and reinstalled in 1973 after gunite work.
- A 2010 theft removed 58 of the original 72 bronze letters.
Microgpt
by tambourine_manMicrogpt is a compact, educational implementation of a GPT-like language model delivered as a single Python file of roughly 200 lines and no external dependencies. The guide explains how the script en...
Key Points
- Single-file (~200 lines) pure Python GPT implementation with no dependencies.
- Trains on ~32,000 names using a simple character-level tokenizer with BOS.
- Source code available via GitHub gist and a dedicated web page.
Just two days of oatmeal cut bad cholesterol by 10%
by gradus_adA clinical trial conducted by the University of Bonn and published in *Nature Communications* investigated whether a very brief, intensive oat-based diet could improve key cardiovascular risk markers ...
Key Points
- A two-day, oat-based, calorie-restricted diet lowered LDL cholesterol by about 10% in adults with metabolic syndrome, with effects lasting at least six weeks.
- The oat group showed greater improvements in cholesterol, weight (about 2 kg loss), and blood pressure than a calorie-restricted control group without oats.
- Oatmeal altered the gut microbiome, increasing bacteria that produce phenolic compounds like ferulic acid, which may help improve cholesterol metabolism.
Running a One Trillion-Parameter LLM Locally on AMD Ryzen AI Max+ Cluster
by mindcrimeAMD’s guide demonstrates local, distributed inference of a one trillion-parameter class LLM using a compact four-node cluster based on the Ryzen AI Max+ AI PC platform. The setup employs Framework Des...
Key Points
- Four-node Framework Desktop cluster with Ryzen AI Max+ 395 and 128 GB per node runs Kimi K2.5 locally.
- Uses AMD ROCm and llama.cpp RPC on Ubuntu 24.04.3 LTS, with a 375 GB GGUF model via Hugging Face.
- BIOS and Linux memory configuration (iGPU setting and TTM) support extended VRAM; nodes linked by 5 Gbps Ethernet.
Samsung Galaxy update removes Android recovery menu tools, including sideloading
by pabs3Samsung is removing several longstanding tools from the Android recovery menu on Galaxy phones as part of One UI 8.5 updates. The recovery environment, commonly used for device maintenance and manuall...
Key Points
- One UI 8.5 removes multiple Android recovery options on Galaxy devices.
- Update ties to February 2026 security releases and blocks downgrades.
- Galaxy S26 Ultra on January 2026 build still shows full recovery tools.
SpacetimeDB ThreeJS Support
by ryker2000The post introduces SpacetimeDB 2.0 as a database-first approach to real-time and multiplayer applications built with Three.js. In this architecture, game logic executes within the database, which the...
Key Points
- SpacetimeDB 2.0 runs game logic inside the database and streams state to clients.
- Three.js clients subscribe over WebSockets to receive fine-grained diffs instead of polling.
- Pricing includes a free tier and a paid plan starting at $25/month.
Microsoft announces new "mini PCs" for Windows 365
by mikeceMicrosoft is expanding its Windows 365 hardware ecosystem with two partner-produced thin client mini PCs designed for commercial users to access cloud-hosted Windows desktops. The ASUS NUC 16 for Wind...
Key Points
- Two new thin client devices for Windows 365 from ASUS and Dell support up to three displays.
- Availability begins in Q3: ASUS in the U.S. and Europe; Dell in 58 countries.
- Windows Cloud PCs will add Bluetooth onboarding and branded login screens in 2026.
Sub-second volumetric 3D printing by synthesis of holographic light fields
by zdwThe article presents Digital Incoherent Synthesis of Holographic light fields (DISH), a volumetric 3D printing method that overcomes the common trade-off between resolution and build speed. Unlike com...
Key Points
- DISH prints volumetric objects via continuous multi-angle projections without sample rotation.
- Achieves 19-μm resolution across a 1-cm depth and 0.6-second builds for millimetre-scale objects.
- Integration with fluid channels enables mass production in low-viscosity acrylate materials.