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.

← Back to Home

Tuesday, April 7, 2026

Anthropic Grabs Gigawatts Of Google AI Power!

Anthropic Grabs Gigawatts Of Google AI Power!

A new VS Code clone called SideX chases speed and drops Electron weight, as devs probe how far it can go... Europe backs a sovereign office bet with Euro-Office, pushing schools and states away from US cloud giants... macOS takes a hit with a Recovery Mode browser bug and a 49-day networking glitch that quietly breaks apps... Adobe raises alarm by writing into the system hosts file just to police its own tools... High above, Anthropic locks in huge Google Cloud and Broadcom power deals, turning frontier AI into a game of gigawatts... Down in the trenches, Claude Code stumbles, new AI agent tests show bots still get lost on the web, and fresh memory tools like Hippo try to fix what context windows forget... Tonight we track coding agents that even run from a phone, reshaping how and where work happens.

New Tools Rise While Old Platforms Crack

  • New VS Code Clone Promises Speed Without Bloat

    SideX is a Tauri-based rework of VS Code promising the same extensions and layout with a fraction of the size and more native feel. Devs loved the ambition and the performance pitch, while side‑eyeing the early, half‑broken state and wondering if it can really escape Electron’s shadow.

  • Europe Pushes 'Sovereign' Office Suite To Escape Giants

    Euro-Office, built on ONLYOFFICE and released under AGPL, sells itself as a European, self-hostable alternative to Microsoft 365 and Google Docs. Commenters were surprisingly hopeful, seeing it as a rare serious swing at freeing schools and governments from US cloud lock‑in and random pricing changes.

  • macOS Recovery Browser Bug Gives Hackers System Access

    A researcher found macOS Recovery Mode’s Safari let writes to protected system volumes, meaning a malicious page could sneak in persistent root access even on supposedly locked‑down machines. Security‑minded readers were impressed by the bug hunt and unimpressed that Apple left such a gaping hole in its prized defenses.

  • macOS Networking Bug Breaks Apps After 49 Days

    Another macOS gem: a TCP networking bug in the XNU kernel that silently detonates after exactly 49.7 days, breaking tools like OpenClaw. Engineers grimly joked about mandatory reboot calendars and wondered how many mysterious production glitches were actually this ticking time bomb hiding in the stack.

  • Adobe Hijacks Hosts File To Check Its Apps

    Users discovered Adobe Creative Cloud quietly stuffing odd entries into the hosts file just to detect whether its own software is installed. The move felt invasive and amateurish at once, fueling long‑running resentment about Adobe’s subscription model, bloat, and the feeling that your own machine isn’t really yours anymore.

AI Labs Race Ahead While Tools Misfire

  • Anthropic Books Massive Google Cloud For Future AI

    Anthropic signed a huge deal with Google and Broadcom for multiple gigawatts of TPU compute starting in 2027. People read this as a clear sign that frontier AI is now a game only trillion‑dollar partners can play, and that future models will be trained on power budgets that rival small countries.

  • Claude Coding Agent Chokes And Exposes Messy Backend

    A rough patch for Claude Code and leaks about its MCP implementation sparked blunt criticism: if these AI tools are so smart, why are their own backends such a mess? Devs used the outage as proof that dogfooding has limits and that shipping reliable infrastructure still beats vibe‑driven “AI‑all‑the‑things” development.

  • New Test Grades How Well AI Agents Read Websites

    The Agent Reading Test benchmark asks coding agents like Claude Code, Cursor and GitHub Copilot to navigate real‑world web docs and then scores them. Folks liked finally having a way to compare agents beyond marketing slides, and it reinforced the feeling that today’s AI assistants still get hopelessly lost in messy documentation.

  • Hippo Tool Teaches AI What To Remember Or Forget

    Hippo offers “biologically inspired” memory for AI agents, deciding what to keep and what to discard across long projects. Builders of agentic IDEs perked up, sick of context windows overflowing with junk. The mood was cautiously excited: everyone knows memory is broken, but nobody believes there’s a magic fix yet.

  • Run AI Coding Agents On Servers From Your Phone

    Onepilot showed a mobile‑first SSH and AI agent IDE that lets you connect to any server from your phone and let bots do dev work. To some, it’s the future of remote ops; to others, it’s a nightmare vision of debugging half‑baked agent changes from a subway seat at 2 a.m.

Internet Control Fights And Tech From Past And Present

  • Age Checks Turn Into Quiet Global Surveillance Network

    A detailed report on age verification laws in the US, UK and Brazil argued they effectively mandate biometric ID checks, building a private surveillance infrastructure wrapped in child‑safety branding. Readers saw the investor list and shuddered, recognizing the same crowd that already profits from tracking everything else we do.

  • Wikipedia Fights Over AI Bots Rewriting Articles

    Wikipedia’s clash over the Tom‑Assistant and other AI bots highlighted how generative tools can swamp human editors with bland, error‑prone text. Long‑time contributors fear the site becoming a cleanup crew for bot spam, while others argue that without strict rules, quiet algorithmic editing could rewrite history in slow motion.

  • Student Booted After Making Social Site For Campus

    A student built iitsocial.com for his university, only to get his phone seized, the cops called, and expulsion threats from the dean. The story read like a parody of overreaction: instead of supporting a homegrown network, the institution treated basic web dev like a cybercrime and terrified every would‑be builder watching.

  • Inside The Maze That Sends Your Text Message Abroad

    A deep explainer on SMS delivery pulled back the curtain on aggregators, shady routes, pricing games, and why texts vanish into the void. Developers who thought they were “just calling an API” got a crash course in a messy telecom underworld that feels stuck in the 90s but still powers every login code we get.

  • Apollo Moon Computer Springs Back To Life On Earth

    A restoration project for the Apollo Guidance Computer showed the 1960s hardware being carefully revived, from core memory to ancient logic modules. Watching this museum piece come back to life charmed readers and underscored how the software that once landed humans on the Moon now fits in a toy microcontroller on your desk.

Top Stories

Anthropic Orders Gigawatts Of AI Power From Google

Artificial Intelligence

Anthropic quietly signed a monster deal with Google and Broadcom for multiple gigawatts of TPU compute starting 2027, signaling an arms race where only a handful of labs can even afford to train frontier AI models.

Read more →Comments

New VS Code Clone Promises Speed Without Bloat

Technology

SideX, a Tauri-based reimplementation of VS Code, hit HN’s front page by promising the same feel with a fraction of the size and more native performance, tapping straight into dev fatigue with Electron’s heaviness.

Read more →Comments

Europe Pushes 'Sovereign' Office Suite To Escape Giants

Technology

Euro-Office, built on ONLYOFFICE under the GNU AGPL, pitched itself as a ‘sovereign’ office suite so governments and companies can escape US cloud lock-in. It hit #1 as devs cheered any serious alternative to Microsoft 365 and Google Workspace.

Read more →Comments

Age Checks Quietly Build Global Face Scan Network

Technology Policy

A deep dive into new age verification laws argued they’re really building biometric ID and surveillance infrastructure, with investors overlapping the usual defense and big-data suspects. HN readers saw it as a privacy nightmare masquerading as child safety.

Read more →Comments

Claude Coding Agent Stumbles And Devs Smell Hype

Artificial Intelligence

Claude Code had a rough day, with outages and leaked implementation details that looked rushed and brittle. The community piled on, using it as exhibit A that AI coding tools are still far from reliable production engineering.

Read more →Comments

macOS Recovery Browser Let Hackers Rewrite The System

Security

A researcher found that macOS Recovery Mode’s Safari let attackers get root-level persistence by writing directly to system partitions, undercutting Apple’s security story and reminding everyone that the ‘safe mode’ isn’t always safe.

Read more →Comments

Wikipedia Battles Rogue AI Editors Flooding The Site

Internet

The dust-up over the Tom-Assistant AI bot on Wikipedia showed how fast generative AI can overwhelm volunteer communities and policies, foreshadowing a messy future where bots quietly reshape the public record if nobody’s watching.

Read more →Comments
All stories (71)
Complete list of news articles from this day

Show HN: I rewrote my 2012 self-signed cert generator in Go – cert-depot.com

by dimastopel

A 2012 weekend project, cert-depot.com, has been rebuilt and relaunched as a modern self-signed certificate generator. The original service was implemented with Node.js, Express, jQuery, and relied on...

Key Points

  • Single-binary Go rewrite using crypto/x509 with no external dependencies.
  • In-memory certificate generation and streaming; no server storage.
  • Supports RSA 2048/4096, ECDSA P-256/P-384, SANs, and ZIP/PEM or PFX/PKCS#12 outputs.

SideX – A Tauri-based port of Visual Studio Code

by 0x1997

SideX is an early-release, open-source effort to port Visual Studio Code’s architecture to a native shell by replacing Electron with Tauri. The project reports over 5,600 TypeScript files adapted from...

Key Points

  • SideX replaces Electron with Tauri (Rust backend + native webview) to port VS Code’s architecture.
  • Working features include Monaco editor, Rust-PTY terminal, Git, themes, native menus, and Open VSX extensions.
  • Many subsystems remain incomplete (extension host, debugging, settings UI), and contributors are invited.

One ant for $220: The new frontier of wildlife trafficking

by gmays

Kenya has become a focal point in a niche but expanding illegal trade in queen ants driven by a global pet hobby. During the rainy season, collectors around Gilgil capture giant African harvester ant ...

Key Points

  • Queens sell for up to $220 and can start long-lived colonies, driving demand.
  • KWS seized 5,000 queens in Naivasha in 2023, destined for Europe and Asia.
  • Trade leverages swarming season near Gilgil and simple mailable packaging.

The Evolution of x86 SIMD: From SSE to AVX-512

by jiehong

This article explores the early evolution of x86 SIMD extensions through the lens of Intel’s MMX technology, emphasizing that the story is driven as much by marketing and corporate politics as by pure...

Key Points

  • Intel’s Israel Development Center in Haifa led the design of the Pentium MMX, Intel’s first flagship processor developed outside the U.S., involving over 300 engineers.
  • MMX’s eight registers were aliased onto the existing x87 floating‑point stack to avoid new processor state and OS changes, forcing use of EMMS when mixing FP and MMX code.
  • Intel trademarked “MMX” and sued AMD over its use of “Matrix Math Extensions,” leading to a settlement where AMD recognized MMX as Intel’s mark but could still use the name.

Does coding with LLMs mean more microservices?

by jer0me

The article describes an observed shift toward microservices driven by LLM-assisted development. Small services with well-defined request/response boundaries make it safer to let LLMs perform large re...

Key Points

  • Microservices with clear interfaces align well with LLM-driven refactoring and iteration.
  • Monoliths face higher risks from implicit coupling compared to isolated services.
  • Operational overhead grows with microservice sprawl, requiring deliberate governance.

France pulls last gold held in US for $15B gain

by teleforce

The Bank of France has completed the withdrawal of its remaining gold held at the Federal Reserve Bank of New York—129 tonnes, around 5% of its reserves—and replaced it with equivalent bars that meet ...

Key Points

  • France repatriated 129 tonnes and replaced them with modern-standard bars stored in Paris.
  • A €13B capital gain from higher gold prices contributed to a €8.1B profit for 2025.
  • All 2,437 tonnes are now held in BdF’s La Souterraine vault; remaining 134 tonnes to standardize by 2028.

Euro-Office – Your sovereign office

by XzetaU8

Euro-Office introduces an open-source, web-based office suite aimed at delivering digital sovereignty for collaborative document work. Rather than functioning as a standalone application, it is design...

Key Points

  • Web-based, integrable office suite for collaborative editing across documents, spreadsheets, presentations, and PDFs.
  • Forked from ONLYOFFICE’s AGPL open-source codebase; code is being reviewed and cleaned for easier builds and contributions.
  • Community-led with forthcoming transparent governance; contributors are guided to the DocumentServer repository.

NIMBY Rails

by altilunium

NIMBY Rails is a real-world map railway management sandbox currently in Early Access. The developers explain that the game’s ambition—removing traditional limits on map size, track complexity, and sch...

Key Points

  • Quarterly major updates with no fixed final release date
  • Planned depth features: programmable signaling, real-time timelines, NIMBYism
  • Fully playable now; co-op and mod support via Steam Workshop

Idiomatic Koru Kernels Match Hand-Specialized C

by nemoniac

The article reports on a language performance evaluation where a fused kernel written in Koru was tested against idiomatic implementations in C, Rust, Zig, SBCL (Common Lisp), and GHC (Haskell) using ...

Key Points

  • Koru’s fused kernel matches scalarized hand-specialized C on an n-body benchmark.
  • Fixed-size scalarized C (with and without -ffast-math) and Koru all measured at 1.00 relative performance.
  • Benchmarking used identical conditions (50M iterations, hyperfine, same machine/harness) with code available on GitHub.

Make your own ColecoVision at home, part 5

by classichasclass

This update details the seventh hardware revision of “Leako,” a DIY ColecoVision clone focused on making the design buildable and robust for more people. The project transitions from a DC barrel jack ...

Key Points

  • USB Type‑B replaces DC barrel jack for power, improving usability but requiring good-quality adapters.
  • Mechanical and connector changes reduce board flex; a footprint error was identified and fixed for future runs.
  • Component shortages drove part substitutions (ICL7660, alternative FET) and caused delays (74LS541).

Number in man page titles e.g. sleep(3)

by thunderbong

The article explains the meaning behind the numbers that appear in Linux man page titles, such as read(2) and sleep(3). Prompted by a code review comment correcting a reference to basename, the author...

Key Points

  • Section numbers in man pages denote categories: 1 (commands), 2 (system calls), 3 (library calls).
  • basename is documented as a section 3 libc function, not a section 2 system call.
  • Suffixes like “p” (POSIX) and “x” (X-related) can follow section numbers in man entries.

Age Verification as Mass Surveillance Infrastructure

by rurban

The article outlines how age verification laws in Brazil, the United Kingdom, and the United States are fueling adoption of biometric identity verification systems that also function as surveillance i...

Key Points

  • Age-verification laws are driving biometric identity systems that double as surveillance.
  • Persona’s leaked code shows wide-ranging verification checks and FinCEN/FINTRAC reporting modules.
  • Palantir and Founders Fund-backed firms hold major contracts and data footprints across Brazil, the UK, and the US.

Is Germany's gold safe in New York ?

by KnuthIsGod

Germany continues to store a significant portion of its gold reserves in the United States, with about one‑third—valued around €160 billion—held in the vaults of the Federal Reserve Bank of New York. ...

Key Points

  • About one‑third of Germany’s gold (~€160B) is held at the Federal Reserve Bank of New York.
  • Bundesbank asserts the New York-held reserves have special protections.
  • Experts fear leadership shifts and U.S. political dynamics could challenge established custodial norms.

The Intelligence Failure in Iran

by JumpCrisscross

This article contrasts two eras of U.S. decision making informed by intelligence. It revisits the 2005 conclusion that prewar assessments of Iraq’s WMDs were largely wrong and then examines a later U....

Key Points

  • Intelligence accurately assessed Iran’s capabilities and likely responses before the conflict.
  • The U.S. proceeded with Operation Epic Fury despite assessments arguing for restraint.
  • Iran’s leverage over the Strait of Hormuz and regional fallout aligned with prewar warnings.

Tiny Corp's Exabox

by macleginn

Tiny Corp announced via X that preorders are open for its Exabox hardware, linking directly to a Shopify-hosted product page. In the post, the company notes it had considered raising a funding round t...

Key Points

  • Tiny Corp opened preorders for Exabox via a Shopify link.
  • Exabox is positioned as a way to scale without raising funds for a datacenter.
  • Deployment requirements are minimal: a concrete slab and adequate power.

Ask HN: How do systems (or people) detect when a text is written by an LLM

by elC0mpa

This Ask HN post seeks guidance on identifying AI-generated text, specifically when written by large language models (LLMs). The author requests input on both human-based heuristics and computer-drive...

Key Points

  • Question focuses on detecting LLM-generated text by humans and systems.
  • Author asks if APIs are available for automated detection.
  • The post is an open request for methods and tools, not a set of conclusions.

What Being Ripped Off Taught Me

by doctorhandshake

An AR practitioner recounts being brought in during spring 2024 to help deliver an augmented reality bus tour in a Beijing park for a California-based client, framed by the claim of being “ripped off”...

Key Points

  • Project suffered from absent version control and risky deployment practices.
  • Core AR calibration and sensor configuration issues were unaddressed.
  • Author proposed specific technical and process fixes but reports a $35,000 loss.

PostHog (YC W20) Is Hiring

by james_impliu

PostHog, a startup from Y Combinator’s W20 batch, is recruiting for a range of positions spanning engineering, product, support, and go-to-market roles. The company describes its platform as a combina...

Key Points

  • PostHog is hiring across multiple technical and go-to-market roles.
  • The company offers a transparent public handbook and compensation calculator.
  • Plans include building 20 more products in the next few years.

Sam Altman May Control Our Future – Can He Be Trusted?

by adrianhon

An investigative account details how OpenAI’s chief scientist, Ilya Sutskever, compiled and circulated confidential memos alleging that CEO Sam Altman misled executives and directors and misrepresente...

Key Points

  • Sutskever’s memos alleged Altman misled leadership and misrepresented safety protocols.
  • OpenAI’s nonprofit governance prioritizes safety and empowered the board to fire the CEO.
  • Altman’s abrupt removal surprised Microsoft and other stakeholders.

NY Times publishes headline claiming the "A" in "NATO" stands for "American"

by latexr

The New York Times announced a correction to a headline from a recent article about President Trump’s threats to leave NATO. The original headline misidentified NATO’s full name, referring to it as th...

Key Points

  • The New York Times is correcting a headline that misstated NATO’s full name.
  • NATO stands for North Atlantic Treaty Organization, not North American Treaty Organization.
  • The correction pertains to a headline on a story about President Trump’s threats to leave NATO.

Germany Doxes "UNKN," Head of RU Ransomware Gangs REvil, GandCrab

by Bender

German authorities have identified the hacker known as “UNKN/UNKNOWN” as 31-year-old Russian Daniil Maksimovich Shchukin, alleging he led the GandCrab and REvil ransomware operations. The BKA attribut...

Key Points

  • BKA names “UNKN” as Daniil Maksimovich Shchukin, alleged leader of GandCrab and REvil.
  • Alleged extortion totals: ~€2M from two dozen attacks causing €35M+ in damage in Germany.
  • US DOJ links Shchukin to $317K+ in crypto tied to REvil proceeds.

When Virality Is the Message: The New Age of AI Propaganda

by virgildotcodes

The article explores how generative AI is reshaping wartime propaganda by making polished, viral content cheap and fast to produce. In March 2026, AI-made videos portraying Donald Trump and Benjamin N...

Key Points

  • Iranian-linked and U.S. government accounts used AI-styled, entertainment-themed videos to message during wartime.
  • Attribution of Iranian videos is disputed between Revayat-e Fath/IRGC links and the self-claimed Explosive News Team.
  • Platform dynamics and generative AI enable rapid, viral spread of polished propaganda content.

I Won't Download Your App. The Web Version Is A-OK

by ssiddharth

This opinion piece challenges the industry trend of prioritizing native mobile apps over functional websites. The author describes constant prompts steering users from the browser to app stores and no...

Key Points

  • Browsers provide more user control than native apps, which can enforce telemetry and retention.
  • Many apps are thin clients; only hardware-heavy use cases clearly require native implementations.
  • UX micro-interactions and performance issues can severely affect perceived app quality.

The Team Behind a Pro-Iran, Lego-Themed Viral-Video Campaign

by tantalor

An Iran-based YouTube channel, Akhbar Enfejari (Explosive News), shifted in February from low-traffic political commentary to AI-generated animations styled like Lego movies that criticize the U.S. Th...

Key Points

  • Explosive News’ AI, Lego-style videos went viral after a February pivot.
  • Media cited state-linked reposts and watermarks; the channel denies regime ties.
  • Content combines sensational imagery, rap, and conspiracies to maximize reach.

Book Review: There Is No Antimemetics Division

by ibobev

Stephen Diehl reviews Sam Hughes’s “There Is No Antimemetics Division,” a science fiction novel built around the notion of antimemes—phenomena that erase themselves from perception and memory, making ...

Key Points

  • Antimemes drive the novel’s central conflict by resisting perception and memory.
  • The book collects and refines Sam Hughes’s SCP Foundation wiki entries into a standalone narrative.
  • The cosmology prioritizes information (noosphere) over matter, shaping plot and stakes.

Claude Code Down

by theahura

A developer recounts an assessment of Model Context Protocol (MCP) implementations in AI coding tools, focusing on Claude/Claude Code and Codex/Codex CLI. They claim MCP, described as a straightforwar...

Key Points

  • Claude Code and Codex CLI are reported to have incomplete or incorrect MCP implementations.
  • The Claude Code repository reportedly has 5,000+ open issues, including bugs and regressions.
  • Codex is said to lack support for MCP prompts and resources.

sc-im Spreadsheets in Your Terminal

by m-hodges

sc‑im (Spreadsheet Calculator Improvised) is a terminal-based spreadsheet that emulates Vim-style interaction on top of ncurses and builds upon the classic ‘sc’. The project announces v0.8.5 (05/21/20...

Key Points

  • v0.8.5 released; maintainer seeks donations for continued development.
  • Supports CSV/TAB/XLSX (import/export), ODS import, Markdown export, with Vim-like TUI and extensive features.
  • Build via make with ncurses and other deps; macOS install via Homebrew; configure via scimrc.

AI Singer Now Occupies Eleven Spots on iTunes Singles Chart

by flinner

A content creator, Dallas Little, has launched an AI-generated artist named “Eddie Dalton” whose tracks have rapidly climbed the iTunes charts. Following the release of several AI-made songs—including...

Key Points

  • Eddie Dalton holds 11 iTunes Top 100 singles spots and a No. 3 iTunes album.
  • Dallas Little uses AI to create and release Eddie Dalton’s songs and visuals quickly.
  • Luminate cites roughly 6,900 track sales and no radio/streaming despite chart success.

81yo Dodgers fan can no longer get tickets because he doesn't have a smartphone

by josephcsible

A widely circulated X post claims that an 81-year-old lifelong Los Angeles Dodgers fan, who has maintained a season pass for over five decades, was informed that printed tickets will no longer be prov...

Key Points

  • An X post alleges a Dodgers season pass holder can no longer receive printed tickets and must use digital tickets.
  • The post claims the individual struggles with digital devices and that the team hasn’t responded.
  • The article offers no official Dodgers statement or independent confirmation beyond the post.

A Cryptography Engineer's Perspective on Quantum Computing Timelines

by thadt

A cryptography engineer announces a shift toward urgent deployment of post‑quantum cryptography, prompted by two recent research papers that substantially reduce the resources needed to break 256‑bit ...

Key Points

  • Resource estimates to break 256‑bit ECC have dropped sharply in two recent papers.
  • Experts signal a 2029 PQC migration deadline and heightened urgency.
  • The author calls for immediate deployment of available PQC (e.g., ML‑DSA).

Reducto releases Deep Extract

by raunakchowdhuri

Reducto introduced Deep Extract, an agent-based system for structured data extraction that emphasizes iterative verification and correction. Instead of single-pass parsing—prone to omissions and recon...

Key Points

  • Deep Extract uses an iterative agentic loop to verify and correct extraction results.
  • Production beta handled 28M+ fields and documents up to 2,500 pages with 99–100% field accuracy on critical docs.
  • Citations mode provides bounding boxes for auditability; users can define correctness rules in prompts.

Why do Macs ask you to press random keys when connecting a new keyboard?

by zdw

The article explains macOS’s keyboard-identification dialog that appears when connecting a new keyboard. Because keyboards typically cannot report exact key positions, macOS asks users to press the ke...

Key Points

  • macOS infers ANSI, ISO, or JIS layouts via keypresses next to left/right Shift.
  • Visual/legend-based selection is unreliable due to variations and locales.
  • Apple keyboards skip the dialog via known vendor/model IDs; third-party IDs are inconsistent.

Sky – an Elm-inspired language that compiles to Go

by whalesalad

Sky is an experimental programming language that seeks to unify full‑stack development under a single, portable binary by compiling to Go. It borrows Elm’s functional programming strengths—Hindley–Mil...

Key Points

  • Compiles to Go to ship a single, portable full‑stack binary.
  • Elm-inspired FP features and TEA, plus LiveView-style server-driven UI using SSE.
  • Self-hosted toolchain (~4MB Go binary) with type-safe Go FFI and no Node/TS/npm.

The Last Quiet Thing

by coinfused

This brief essay underscores the continuous demands modern connected devices place on users. It lists routine responsibilities—updating, charging, configuring, troubleshooting, managing, and maintaini...

Key Points

  • Devices require persistent upkeep and management.
  • Connectivity, apps, and accounts are prerequisites for many features.
  • Updates, subscriptions, and location tracking extend user involvement.

Adobe modifies hosts file to detect whether Creative Cloud is installed

by rglullis

The article describes a technique reportedly used by Adobe’s Creative Cloud on Windows and macOS to let Adobe’s website determine whether Creative Cloud is installed. Upon visiting adobe.com/home, the...

Key Points

  • Creative Cloud adds hosts file entries to enable installation detection via adobe.com.
  • The site loads cc.png from a Creative Cloud subdomain; success indicates installation.
  • Chrome’s Local Network Access restrictions prompted Adobe to abandon localhost checks.

Show HN: GovAuctions lets you browse government auctions at once

by player_piano

The article introduces GovAuctions, a site designed to simplify access to U.S. government surplus auctions by aggregating listings from major official platforms into a single, clean, searchable interf...

Key Points

  • Aggregates listings from major official government auction platforms into one interface.
  • Supports browsing and email alerts; bidding occurs on original platforms via click-through.
  • Free to search with no account required, addressing fragmentation and outdated interfaces.

Eighteen Years of Greytrapping – Is the Weirdness Finally Paying Off?

by jruohonen

This retrospective reviews an 18-year greytrapping experiment on a self-hosted mail system operated under nxdomain.no/bsdly.net. After starting with a Debian Linux setup, the operator moved edge defen...

Key Points

  • Spamd with greylisting and blocklists (SPEWS, Spamhaus) reduced mail server load.
  • Greytrapping began in 2007 and has run as a long-term experiment.
  • Email has shifted from self-hosted setups to centralized cloud providers.

The Cult of Vibe Coding Is Insane

by drob518

This article critiques an approach to AI-assisted software development it calls “vibe coding,” framed around a reported leak of Claude’s source code and subsequent derision of its quality. The author ...

Key Points

  • Pure vibe coding is unrealistic; AI needs human structure and oversight.
  • AI can rapidly reduce technical debt when guided by ‘Ask mode.’
  • Brief code inspection plus structured prompts can eliminate redundancy.

Zooming UIs in 2026: Prezi, impress.js, and why I built something different

by tinchox6

The article examines two dominant approaches to zooming in web interfaces—Prezi’s closed, presentation-focused platform and impress.js’s open-source, step-based presentation framework—and introduces a...

Key Points

  • Zumly treats zoom as app navigation, dynamically mounting views akin to SPA routing.
  • Prezi and impress.js excel at presentations but are limited for navigable web apps.
  • Zumly is framework-agnostic and available via a live demo and GitHub repo.

Smart people recognize each other – science proves it

by 01-_-

A peer-reviewed study reported in the journal Intelligence examines why some people more accurately gauge others’ intelligence. Conducted by a German team led by Christoph Heine, the research enlisted...

Key Points

  • More intelligent evaluators judged others’ intelligence more accurately.
  • Reliable cues were speech clarity, content, and vocabulary—not appearance or posture.
  • Gender and traits like empathy and openness did not predict judgment accuracy.

Got kicked out of uni and had the cops called for a social media website I made

by 1whizkid1

A student at IIT Delhi describes launching iitsocial.com around 1 a.m., a site that scraped data for all IIT Delhi students to auto-create profiles. The platform enabled anonymous accounts, comments, ...

Key Points

  • A student built and launched iitsocial.com by scraping IIT Delhi student data.
  • The site went viral overnight with heavy anonymous activity and tagging.
  • Campus security and Delhi Cyber Police contacted the creator after complaints.

Agent Reading Test

by kaycebasques

Agent Reading Test is an open benchmark created to evaluate how AI coding agents handle real-world documentation websites. It focuses on common failure modes in web content ingestion—such as truncated...

Key Points

  • Benchmarks agent reading of documentation by simulating realistic tasks and using canary tokens.
  • Targets known web-content failure modes (CSS, client-side rendering, truncation, tabs).
  • Provides a 20-point scoring system; typical current-agent scores are expected to be 14–18.

Show HN: I successfully failed at one-shot-ing a video codec like h.264

by bushido

The article showcases Sinter, an experimental, patent-free video codec created to test one-shot AI agent team workflows using Claude Code. Rather than targeting production readiness, the project explo...

Key Points

  • Experimental patent-free codec built with Claude Code agent teams
  • Around 18.6x larger than H.264 at similar luma quality in tests
  • Gap driven by missing tools; patent-safe ceiling estimated at 4–6x H.264

Londoners are sick of viral videos telling lies about their city

by pseudolus

London Centric’s latest weekend update highlights a growing reputational challenge for London driven by viral videos that depict the city as collapsing into crime and disorder. Building on a prior inv...

Key Points

  • Met Police are continuing to investigate a TikTok account that posted false videos about London.
  • London Centric says viral content is harming London’s global reputation and may affect tourism and investment.
  • Reader anecdotes from Suriname and Australia show misinformation about London has spread internationally.

Show HN: Ghost Pepper – 100% local hold-to-talk speech-to-text for macOS

by MattHart88

Ghost Pepper is a free macOS menu bar app that delivers 100% local, hold-to-talk speech-to-text on Apple Silicon devices running macOS 14 or later. Users hold the Control key to record and release to ...

Key Points

  • 100% local, hold-to-talk STT for macOS 14+ on Apple Silicon
  • Uses WhisperKit for transcription and Qwen 2.5 for cleanup; models cached locally via Hugging Face
  • No cloud use or disk logging; supports MDM pre-approval for Accessibility permissions

Wikipedia's AI agent row likely just the beginning of the bot-ocalypse

by hackernj

Wikipedia blocked an autonomous AI agent, Tom-Assistant, after it created and edited articles without formal bot approval. The agent, built by Covexent CTO Bryan Jacobs, contributed via the account To...

Key Points

  • Wikipedia blocked Tom-Assistant for operating without bot approval.
  • Wikipedia bans LLMs for new content due to policy violations.
  • The incident reflects a broader trend toward more autonomous “agentic AI.”

Show HN: I built a 2-min quiz that shows you how bad you are at estimating

by convexly

The article presents the “Calibration Challenge,” a short, free, no‑signup quiz that evaluates how well a person’s confidence aligns with actual correctness. In under two minutes, users answer 10 ques...

Key Points

  • A free, 2‑minute, no‑signup calibration quiz uses a confidence slider across 10 questions.
  • Convexly tracks decisions, provides immediate behavioral insights, and monitors outcomes over time.
  • Designed as a “fitness tracker for decision‑making,” with start and Pro upgrade options.

AI agents over SSH from your phone. Starting to think this is where dev is going

by elmlabs

The article describes Onepilot, a mobile-first SSH and AI agent IDE that turns an iPhone into a control center for remote development and automation. It offers a unified dashboard where users can see ...

Key Points

  • Onepilot provides a mobile-first SSH IDE with terminal, file, git, and cron management for remote servers.
  • It offers a guided workflow to deploy and configure AI agents, including model, channel, and personality selection, without manual DevOps steps.
  • The platform integrates with 23+ LLM providers and messaging channels like Telegram, Discord, and Slack, differentiating it from traditional SSH clients.

Show HN: Docking – extensible Linux dock in Python

by edumucelli

Docking is presented as a Linux-focused dock application built entirely in Python. The project emphasizes practical, everyday usability by aiming to adapt to individual workflows, highlighting speed, ...

Key Points

  • Extensible Linux dock written entirely in Python
  • Includes built-in applets for out-of-the-box functionality
  • Designed for full integration with the Linux desktop environment

A macOS kernel bug can cause OpenClaw to stop working after 49.7 days

by RyanZhuuuu

Photon documents a reproducible macOS TCP failure caused by a 32‑bit unsigned integer overflow in Apple’s XNU kernel after roughly 49 days, 17 hours of continuous uptime. The overflow freezes the kern...

Key Points

  • A 32‑bit overflow in XNU’s tcp_now freezes TCP timestamps at ~49.71 days uptime on macOS.
  • TIME_WAIT never expires post‑overflow, exhausting ephemeral ports and blocking new TCP connections; ICMP still works.
  • Photon reproduced the bug, traced it to XNU code, and offers a reproduction guide and mitigation (reboot).

HackerRank (YC S11) Is Hiring

by rvivek

HackerRank is recruiting for an FDE role as it rolls out a revamped, AI agent–oriented approach to developer hiring. The company states it supports over 2,500 companies and has a platform with more th...

Key Points

  • HackerRank is hiring an FDE to deploy its AI agent–driven hiring process with customers.
  • Assessments shift to real-world tasks, AI orchestration evaluation, and an Agentic IDE experience.
  • Hybrid roles available in Santa Clara, New York, London, and Bangalore; applicants should evidence Smart, Curiosity, Agency, and AI fluency.

Root Persistence via macOS Recovery Mode Safari

by yaseeng

A security researcher details two vulnerabilities in Safari when run within macOS Recovery Mode. The first, rated CVSS 8.5 and affecting macOS Sequoia and older, allows arbitrary file writes to system...

Key Points

  • Arbitrary write/root persistence via Recovery Mode Safari on macOS Sequoia and older (CVSS 8.5).
  • Unrestricted file read issue in macOS Tahoe via Recovery Mode Safari (CVSS 4.6).
  • Changing Safari’s download setting enabled persistent writes to “Macintosh HD” from Recovery Mode.

SOM: A minimal Smalltalk for teaching of and research on Virtual Machines

by tosh

SOM (Simple Object Machine) is introduced as a minimal Smalltalk tailored for teaching and virtual machine research. The article demonstrates a simple language example (a Fibonacci method) and mention...

Key Points

  • Multiple SOM implementations provide the same language with differing techniques and optimizations.
  • SOM underpins research projects and additional language implementations (e.g., SOMNS, Moth, TruffleMATE).
  • Community resources include GitHub code, Twitter updates, a mailing list, and Slack for collaboration.

Show HN: Tusk for macOS and Gnome

by factorialboy

Tusk is a free, open-source PostgreSQL client offered for both macOS and GNOME/Linux, aimed at developers and database administrators who need schema browsing, table inspection, data manipulation, and...

Key Points

  • Free, open-source PostgreSQL client for macOS and GNOME/Linux
  • Shared core: schema browsing, table inspection, data editor, and SQL editor
  • Platform-specific extras: macOS (SSL/TLS, Activity Monitor, JSON tree), GNOME (.pgpass import, role browser, pinned columns)

Show HN: TTF-DOOM – A raycaster running inside TrueType font hinting

by 4RH1T3CT0R

TTF-DOOM is a compact 3D raycasting engine implemented entirely within a TrueType font’s hinting virtual machine. Unlike prior font-based computation that relied on HarfBuzz’s WebAssembly shaping, TTF...

Key Points

  • First 3D renderer to run inside TrueType font hinting bytecode, not via WebAssembly shaping.
  • Font acts purely as a renderer; JavaScript handles game state and passes parameters via variation settings.
  • Overcomes hinting VM constraints with recursion, hit-flags, axis jitter, and MPPEM-based coordinate conversion.

NYC Families Need over $125,000 in Income to Live in Any Borough

by boh

Two new reports indicate that families in New York City need household incomes exceeding $125,000 to live without government or private assistance in any of the city’s five boroughs. The article highl...

Key Points

  • NYC households require over $125,000 to meet basic needs without assistance across all boroughs.
  • A key report from the Fund for the City of New York measures the 'self-sufficiency standard.'
  • This analysis has been released regularly since 2000, offering a long-term affordability benchmark.

Show HN: Hippo, biologically inspired memory for AI agents

by kitfunso

Hippo is a biologically inspired, CLI-based shared memory layer designed to unify and manage memories across AI agents like Claude Code, Codex, Cursor, and OpenClaw. Instead of saving everything, Hipp...

Key Points

  • Auto integration: hippo init wires hooks, creates dirs, and schedules daily hippo learn --git and hippo sleep.
  • Cross-tool import and conversation capture with dry-run, tagging, and duplicate detection.
  • Biologically inspired memory model with default decay, episodic-to-semantic consolidation, and reinforced retrieval.

Show HN: Anos – a hand-written ~100KiB microkernel for x86-64 and RISC-V

by noone_youknow

Anos is a hand-written, microkernel-based hobby operating system targeting x86_64 and RISC-V that has reached “toy kernel” maturity. It currently comprises the STAGE3 microkernel, a SYSTEM user-mode s...

Key Points

  • Microkernel STAGE3 plus SYSTEM supervisor enable userspace services and drivers.
  • Fast, capability-governed syscalls (SYSCALL/SYSRET, ecall) with zero-copy IPC.
  • Targets x86_64 and RISC-V; runs on real hardware with userspace drivers and multitasking.

Ask HN: How do you handle marketing as a solo technical founder?

by lazarkap

This short guidance piece for solo technical founders emphasizes customer discovery and iterative refinement as the foundation for successful product building. It recommends beginning by talking to cu...

Key Points

  • Talk to customers before building to validate needs.
  • Iterate with user feedback until the product solves their problems.
  • Use word-of-mouth first; formal marketing comes later.

The anatomy of SMS delivery: from request to carrier

by Bridgexapi

This article breaks down what actually happens inside an SMS delivery system after a developer sends a request to an SMS API. It argues that most developers mistakenly equate the API call with message...

Key Points

  • An SMS API call is just an entry point into a hidden routing and execution system, not the full delivery process.
  • Most SMS issues arise from opaque routing and execution decisions, not the basic act of sending a message.
  • Routing is based on detailed routing profiles and strict access controls, which define behavior, pricing, and allowed traffic.

Anthropic expands partnership w Google and Broadcom for multiple GW of compute

by l1n

Anthropic is expanding its compute footprint through a new agreement with Google and Broadcom to secure multiple gigawatts of next‑generation TPU capacity, expected to begin coming online in 2027. The...

Key Points

  • Anthropic secures multi‑GW next‑gen TPU capacity with Google and Broadcom, starting 2027.
  • Run‑rate revenue surpassed $30B in 2026; $1M+ customers doubled to 1,000+ since February.
  • Most new compute will be in the U.S., expanding a $50B infrastructure commitment.

After 20 years I turned off Google Adsense for my websites

by datadrivenangel

A veteran blogger ended a two-decade relationship with Google AdSense, which he joined in 2005 to better understand online advertising and supplement income. While early results included high-value cl...

Key Points

  • AdSense revenue declined from ~$1,000/year to about $100/year.
  • Expanded ad formats caused unintended, intrusive ad placements.
  • Removing ads aims to clarify non-commercial status and reduce legal risk.

The back story behind the first "$1.8B" dollar "AI Company"

by chermanowicz

Gary Marcus critiques a New York Times article that presented Medvi as a $1.8 billion “AI company,” noting the piece went viral as a triumph for AI-enabled entrepreneurship. Marcus argues the NYT cove...

Key Points

  • NYT’s viral profile framed Medvi as a $1.8B “AI” success.
  • Referenced lawsuit alleges California anti-spam violations by Medvi’s affiliates.
  • Prior YouTube and Futurism critiques offer additional context the NYT underplayed.

Porting Go's strings package to C

by ingve

A developer details porting components of Go’s standard library to C to make a Go-to-C subset practical. After starting with the io package for core I/O abstractions, the work shifted to bytes and str...

Key Points

  • Operator precedence differences (shifts vs. +/−) require careful parenthesization in C to match Go.
  • Dependencies math/bits and unicode/utf8 were ported first as pure-function building blocks.
  • bytes.Equal in C uses zero-copy string reinterpretation and memcmp to mirror Go’s no-allocation equality.

Netflix Void Model: Video Object and Interaction Deletion

by bobsoap

Netflix introduced VOID (Video Object and Interaction Deletion), a video inpainting method that removes target objects and their induced physical effects from footage. Built atop CogVideoX and fine-tu...

Key Points

  • VOID removes objects and their induced interactions using a two-pass inpainting pipeline.
  • Pass 2 warped-noise refinement improves temporal consistency over the base pass.
  • Setup requires Gemini (Google AI API), SAM2, CogVideoX base model, and a 40GB+ VRAM GPU.

Solod – A Subset of Go That Translates to C

by TheWiggles

Solod (“So”) introduces a systems-oriented subset of Go that transpiles directly into readable C11 while eliminating a runtime. The project targets developers who want Go’s syntax, type safety, and to...

Key Points

  • Transpiles a strict Go subset to readable C11 with zero runtime and manual memory management.
  • Supports key Go features but excludes channels, goroutines, closures, and generics.
  • Provides native C interop without cgo and integrates with Go tooling and module workflows.

What happens when a destructor throws

by jandeboevrie

The article explains the C++ runtime behavior when a destructor throws, grounding the discussion in RAII and exception handling mechanics. It notes that destructors are integral to resource management...

Key Points

  • Destructors are implicitly noexcept(true) since C++11 unless declared otherwise.
  • Exceptions escaping a noexcept destructor or thrown during unwinding call std::terminate.
  • Using noexcept(false) on a destructor allows exceptions to propagate only when no other exception is active.

Graph-go – zero config, full visibility

by devGrimm

graph-info, part of the graph-go project, provides zero-config infrastructure visibility by connecting to the Docker daemon to auto-discover services and build an interactive dependency map. It suppor...

Key Points

  • Auto-discovers Docker-hosted services and builds an interactive dependency map with health.
  • Supports major databases, Elasticsearch, and S3-compatible storage with detailed metadata.
  • Offers Docker-first install, optional YAML config for external services, and a ready-to-run demo.

Linux extreme performance H1 load generator

by MDA2AV

Glass Cannon (gcannon) is introduced as a high-performance HTTP/1.1 and WebSocket load generator that leverages Linux’s io_uring for batched asynchronous I/O. It is positioned as the official load gen...

Key Points

  • Leverages Linux io_uring for high-throughput HTTP/1.1 and WebSocket load generation.
  • Provides exact per-request latency histograms and a rich TUI with history and adaptive charts.
  • Offers JSON output for automation and supports only http:// with comprehensive CLI controls.

Dear Heroku: Uhh What's Going On?

by crcastle

A Judoscale-authored piece questions the clarity of Heroku’s recent strategic shift. On February 6, Heroku announced it was transitioning to a “sustaining engineering” model focused on stability, secu...

Key Points

  • Heroku announced a shift to a sustaining engineering model and halted new Enterprise Account contracts.
  • Heroku released March updates: bigger slugs, more frequent SSL/TLS cert provisioning, and a rebuilt CLI (v11).
  • The author calls for clearer messaging, a public roadmap, and business clarification to set user expectations.

Apollo Guidance Computer Restoration Videos and Press Coverage

by mariuz

This article chronicles the restoration of an Apollo Guidance Computer (AGC) discovered by Jimmie Loocke among scrapped NASA equipment purchased in Texas in 1976. In preparation for the Apollo 11 50th...

Key Points

  • A privately owned Apollo Guidance Computer was restored to operation and publicly demonstrated.
  • The team published detailed technical resources on AGC memory diagnosis and repair.
  • Recovered Apollo-era software from nine core rope modules, including a complete Sundial‑E.
Made with <3 by @siedrix and @shesho from CDMX. Powered by Forge&Hive.