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.
Wednesday, March 18, 2026
Meta’s $2B Plot To Scan Kids!
Big tech power plays dominate today as Meta stands accused of routing over $2B through nonprofits to push phone-level age checks that could turn devices into built-in ID scanners... In Illinois, lawmakers advance an OS based Children’s Social Media Safety Act, raising fresh fears of mandatory surveillance baked into every app... New whistleblowers say outrage boosting feeds on Meta and TikTok favor anger over mental health, while Honda slows its EV push and Meta quietly shrinks its metaverse dream... On the AI front, tiny GPT‑5.4 models bring smart tools to cheap chips, Mistral Forge courts developers into a new cloud lock in, and Zeroboot’s ultra fast VM sandboxes hint at agents that act, not just chat... In the middle of it all, we see coders turning Claude Code and fresh workflows into a new kind of superteam.
Big Tech And Governments Turn Up The Control
Meta’s Secret $2B Push To Scan Your Kids
A Reddit user dug through filings and claims Meta quietly routed over $2B through nonprofit fronts to push age-verification laws. The catch: they’d force Apple and Google to bake new surveillance tech into phones. Readers see it less as child protection and more as a power grab over the entire app ecosystem.
Illinois Wants Your Phone To Check Your Age
An Illinois bill would push age checks down to the operating system, not just social apps. With talk of an OS “Children’s Social Media Safety Act,” people worry this is how Meta-style age controls become mandatory for every app, turning phones into built-in ID scanners by law rather than by choice.
Whistleblowers: Outrage Was A Feature, Not A Bug
New whistleblower claims say Meta and TikTok let more harmful, enraging posts rise after internal research showed outrage boosted engagement. Instead of dialing it down, they apparently leaned in. Many see this as proof the big feeds value watch time over mental health, and no one is shocked, just annoyed it’s now on record.
Honda Backs Away From Electric Cars As Rivals Charge
Honda is quietly killing off its EV plans in key markets just as cheaper Chinese models start knocking on the door. Commenters are baffled that a major brand is retreating while others double down. It feels like watching a classic car company choose short-term comfort over the long-term electric future.
Meta Shrinks Its Metaverse Dream Yet Again
Meta is discontinuing Horizon Worlds on Meta Quest, peeling apart its platforms and shrinking the once-grand metaverse pitch into just another mobile app. The community reaction mixes schadenfreude with sympathy: billions later, the big VR bet looks more like a slow-motion climbdown than the future of the internet we were promised.
AI Arms Race Hits Turbo And Gets Weird
Tiny GPT‑5.4 Models Bring Big Brains To Cheap Chips
New GPT‑5.4 mini and nano models promise much of the big model’s power in small, fast packages. People see this as a tipping point: soon, phones, toasters, and every dull SaaS dashboard will quietly run on-device AI. The excitement is real, but so is the dread of smart features stapled onto everything for no reason.
Mistral Launches Forge To Lock In AI Developers
Mistral AI rolled out Forge, a heavy-duty platform for running its models in production, clearly targeting the same wallets feeding OpenAI and Google. Devs like seeing a strong European contender, but also joke that every lab is now building its own mini cloud empire, and everyone wants you deeply, hopelessly locked in.
Garry Tan Turns Claude Code Into A Dev Superteam
YC’s Garry Tan shared his Claude Code setup, using a tool called gstack to spin up specialized AI helpers for planning, refactors, and code review. Engineers love the pragmatism but admit it highlights a new reality: the real power now sits with people who can orchestrate many AI tools at once, not just write clever code.
Scathing Essay Mocks Bosses Chasing AI Coding Speed
A widely shared post argues that obsessing over AI-assisted code-writing speed is missing the point. The real slowdown is bad specs, unclear priorities, and broken review processes. Developers are clearly relieved someone said it out loud: the problem isn’t slow keyboards, it’s leadership treating LLMs like magic instead of fixing the basics.
Sub‑Millisecond VM Sandboxes Take AI Agents Off The Leash
A project called Zeroboot shows off sub‑millisecond VM sandboxes using copy‑on‑write memory, built to run AI agents safely. It’s deeply nerdy, but people see it as a missing puzzle piece: if agents can spin up real environments fast and safely, we inch closer to AI systems that actually do things, not just chat prettily.
Hackers, Coders And Curmudgeons Steal The Side Show
‘Unhackable’ Xbox One Finally Falls To A Single Hacker
A hacker known as Bliss unveiled a working attack on Microsoft’s supposedly “unhackable” Xbox One, breaking a 13‑year streak at a security conference. Console mod fans are thrilled, corporate security folks less so. It’s a reminder that in tech, absolute security claims age about as well as milk in the sun.
Blizzard’s Slug Text Tech Gifted To The Public Domain
The Slug Algorithm, a slick way to render fonts directly from Bézier curves on GPUs, has been officially dedicated to the public domain. It once powered big games from Activision Blizzard. Graphics nerds are delighted: it’s rare to see industry-grade tech truly freed instead of shoved into yet another paid engine or license trap.
Python Finally Gets Its Just‑In‑Time Groove Back
The Faster CPython team says the Python 3.15 JIT is back on track after a rocky start. For a language accused of being slow, this feels like overdue maintenance on a beloved old car. Devs are cautiously optimistic, hoping for real-world speedups without breaking the mountains of legacy code that keep the internet running.
Django Devs Say: Send Cash, Not AI Tokens
The Django Software Foundation bluntly told fans that paying for LLM time to “have AI fix bugs” is pointless compared to just donating money or doing real work. Open-source maintainers clearly feel overrun by hype. The message lands hard: projects need maintainers, docs, and human care, not drive‑by AI patches from bored executives.
Web Veteran Loses Patience: ‘Have A Fucking Website’
A blunt essay rips into creators and startups that live entirely on social media instead of running their own website and mailing list. The tone is ranty but hits a nerve: relying on feeds and algorithms feels increasingly fragile. Many old‑timers cheer it on as a return to the simple, open‑web values we quietly miss.
Top Stories
Reddit Sleuth Blows Open Meta’s $2B Lobby Machine
Huge leak-style investigation claiming Meta secretly bankrolled age-verification laws that would hardwire surveillance into phones, sparking anger at big tech’s grip on kids’ safety rules.
Illinois Aims Age Checks At The Operating System Itself
A US state moves to force OS-level age checks, confirming fears that the Meta-style age-verification push might end up baked into iOS and Android, not just apps.
Whistleblowers Say Meta And TikTok Fueled Outrage On Purpose
Insiders claim the big social apps knowingly let more harmful, outrage-bait content rise because it juiced engagement, confirming the worst suspicions about the modern attention economy.
Mistral’s New Forge Takes Aim At Big AI Clouds
Europe’s hottest AI lab rolls out Forge, a serious platform play clearly meant to keep developers from disappearing into OpenAI and Google’s ecosystems.
GPT‑5.4 Mini And Nano Shrink Big AI Into Tiny Models
New small GPT‑5.4 models promise ‘good enough’ brains at bargain speed, signaling a future where powerful AI is cheap, tiny, and running everywhere by default.
Developers Told: AI Isn’t Your Bottleneck, Management Is
A viral rant skewers executives chasing code-speed metrics with AI, arguing that specs, reviews, and org chaos—not typing speed—are what really slow teams down.
Honda Slams Brakes On Its Electric Car Future
A legacy automaker all but walks away from EVs just as Chinese rivals surge, sounding like a surrender note in the electric car race.
Gitana 18: the new flying Ultim trimaran
by divbzeroGitana Team has unveiled Gitana 18 in Lorient, a 32‑meter ocean‑racing trimaran engineered for full flight. The design adopts three‑axis adjustable, Y‑shaped foils inspired by America’s Cup AC75s to d...
Key Points
- 32 m fully flying trimaran with adjustable foils and cavitation‑resistant appendages
- Monocoque cockpit integration and autoclave composites for torsional rigidity
- Aeronautics‑inspired mechatronic flight control; 36‑month, 200,000‑hour build
The unlikely story of Teardown Multiplayer
by lairvThe article outlines the journey to implementing multiplayer for Teardown, a game with a fully dynamic, destructible world and extensive mod support. Initial attempts in 2021 tried to synchronize movi...
Key Points
- Hybrid semi-deterministic design: deterministic destruction plus state synchronization for other data
- Deterministic command streams over reliable networking to minimize bandwidth and ensure world sync
- Host-as-server architecture without dedicated servers
Show HN: Droeftoeter, a Terminal Coding Toy
by whtspc64Droeftoeter is a playful, terminal-based “coding toy” that leverages an LLM to generate and extend code driving a 64×32 character grid. Users type what they want to see, and the model continuously bui...
Key Points
- Terminal-based tool animates a 64×32 grid via LLM-generated, cumulative code.
- Supports multiple providers (Groq/Llama, Gemini, OpenAI-compatible, Anthropic/Claude, Ollama).
- Available as binaries or buildable with Go; MIT-licensed with flexible configuration.
Fatal Core Dump Game
by pabs3Robopenguins presents “Fatal Core Dump,” an educational murder mystery that teaches debugging by making a core dump the main piece of evidence. The author provides a playable web version and open-sour...
Key Points
- Educational game focuses on solving a mystery via core dump debugging.
- Airlock controller implemented in C provides realistic, minimal-dependency context.
- Source code is available on GitHub, with a playable version online.
Grace Hopper's Revenge
by ashirviskasThis commentary examines how large language models (LLMs) perform differently across programming languages and argues that language design, not just training data volume, strongly influences AI-assist...
Key Points
- AutoCodeBench indicates top performance in Elixir, Kotlin, Racket, and C#, with PHP, JavaScript, Python, and Perl lagging.
- Python-centric benchmarks like SWEBench and TerminalBench may not generalize across languages.
- Language structure and paradigms, not solely training data volume, are central to LLM coding performance.
What CI looks like at a 100-person team (PostHog)
by shad42A case study of PostHog’s continuous integration highlights the operational realities of CI at a ~100-person, fast-moving engineering team using a large public monorepo. In one week (Jan 27–Feb 2), Po...
Key Points
- PostHog’s CI operates at massive weekly scale, with 575k+ jobs and 33M+ tests.
- Even a 99.98% pass rate yields significant noise and re-runs at this volume.
- Mendral, a GitHub App AI agent, targets flaky tests by diagnosing failures and automating fixes.
Reverse-engineering Viktor and making it Open Source
by zggfIn a hands-on exploration, engineer Mateusz Jacniacki persuaded Viktor—Zeta Labs’ AI coworker that claims connectivity to 3,000+ tools—to generate backups of its own workspace and then its full filesy...
Key Points
- Backups revealed Viktor’s SDK, a centralized Tool Gateway, and a large list of integrations (many via Pipedream).
- Sixteen days of structured logs enabled tracing Slack conversations end-to-end within the agent.
- Nineteen prompt-based skills include a six-phase workflow discovery process and a Convex+Vercel app builder.
Building a Shell
by ingveThis article walks through building a simple Unix-like shell in C, starting with an interactive prompt and a minimal read-eval-print loop. The shell keeps track of running state and the last command s...
Key Points
- Minimal shell loop in C with state tracking and 'exit' built-in
- External commands executed via fork/execvp with waitpid for status
- Tokenizer splits on whitespace and '|' tokens; quotes/redirection deferred
Gummy Geometry
by memalignGummy Geometry is a hands-on suite of embedded CodePen demos that showcase a broad range of 2D physics behaviors. Users can click and drag to spawn or manipulate objects while monitoring FPS, body cou...
Key Points
- Shows six core joint types and their behavior across multiple scenes.
- Covers both rigid-body and deformable simulations, including cloth and pneumatic soft bodies.
- Includes mechanics and gameplay staples like vehicles, one-way platforms, and conveyor belts.
Reddit User Uncovers Who Is Behind Meta's $2B Lobbying for Age Verification Tech
by doenerThe article presents claims from an online investigation by “upper-up” that Meta supported a broad, multi-state push for age-verification legislation through a network of nonprofits, allegedly routing...
Key Points
- Allegations of Meta-backed, multi-state lobbying for age-verification laws via nonprofits, totaling over $2B.
- Proposals would require OS-level age APIs, potentially impacting Apple and Google; Meta’s Horizon OS cited as precedent.
- EU’s eIDAS 2.0/EUDI Wallet model is presented as a privacy-preserving alternative using zero-knowledge proofs.
Font Smuggler – copy hidden brand fonts into Google Docs
by lanewinfieldFont Smuggler is a web page created by Brian Moore that demonstrates a simple workaround to use brand-locked typefaces in Google Docs and Slides. The page explains that Google Workspace allows brands ...
Key Points
- Webpage shows copy/paste method to render brand-locked fonts in Google Docs/Slides.
- Includes a step-by-step guide and a curated list of brand fonts.
- Noted limitation: the method does not work on Google mobile apps.
Enabling Efficient Sparse Computations Using Linear Algebra Aware Compilers
by matt_dThe article introduces LAPIS, an MLIR-based compiler framework designed to optimize sparse linear algebra and deliver performance portability across diverse hardware. LAPIS’s centerpiece is a Kokkos d...
Key Points
- LAPIS uses MLIR with new Kokkos and partition dialects to optimize and distribute sparse linear algebra.
- MLIR-driven optimizations improve performance for sparse and dense kernels across GPUs.
- Applications span graph kernels and TenSQL (GraphBLAS-based), demonstrating performance portability.
Silicon Valley's "Pronatalists" Killed WFH. The Strait of Hormuz Brought It Back
by bigbobbeeperThe article argues that remote and hybrid work arrangements materially increase fertility among employed, partnered adults, citing a 2026 study (Davis et al.) that attributes roughly 291,000 U.S. birt...
Key Points
- Study estimates ~291,000 U.S. births annually linked to WFH.
- Major firms enforced strict RTO by early 2026; surveys show broad CEO support for full RTO.
- Strait of Hormuz closure led Asian governments to mandate remote work for energy savings.
A proposal to classify happiness as a psychiatric disorder
by wjSgoWPm5bWAhXBIn a 1992 Journal of Medical Ethics article, R. P. Bentall proposes that happiness should be classified as a psychiatric disorder under the label “major affective disorder, pleasant type.” The proposa...
Key Points
- Proposes “major affective disorder, pleasant type” as a diagnostic label for happiness.
- Cites statistical rarity, symptom clustering, cognitive correlates, and CNS involvement as criteria.
- Dismisses positive valuation as scientifically irrelevant to classification.
Finding a CPU Design Bug in the Xbox 360
by mariuzIn a 2018 technical post, Bruce Dawson recounts how the disclosures of Meltdown and Spectre brought to mind his experience uncovering a CPU design bug in the Xbox 360. As the engineer focused on the c...
Key Points
- Dawson discovered a CPU design bug in the Xbox 360 linked to a newly added instruction.
- The Xbox 360’s IBM-built three-core PowerPC CPU had high latencies and a small shared L2 cache.
- MESI coherence required L1 data to reside in L2, complicating L2 space conservation.
Turns Out the DOGE Bros Who Killed Humanities Grants Are Sensitive About It
by LaSombraThe article details new revelations from depositions in a lawsuit brought by the American Historical Association, Modern Language Association, and American Council of Learned Societies against the Nat...
Key Points
- Plaintiffs uploaded four government witness depositions to YouTube in their lawsuit against NEH.
- Depositions show Justin Fox used ChatGPT prompts to flag grants for DEI-related cancellation.
- The videos were later removed after media coverage, according to the article.
How long does it take to get last liquid drops from kitchen containers?
by hhsPhysicists at Brown University examined everyday kitchen fluid dynamics to quantify how long thin liquid films take to drain from tilted containers and how best to clear residual water from a cast iro...
Key Points
- 90% drainage times vary widely with viscosity: seconds (water), ~30s (milk), >9 min (olive oil), hours (cold maple syrup).
- Experiments on a 45° plate confirmed viscous-regime Navier–Stokes predictions for thin-film flow.
- Simulation suggests ~15 minutes to let residual wok water pool before dumping.
OpenSUSE Kalpa
by ogogmadKalpa is presented as an atomic, immutable Linux desktop within the openSUSE ecosystem, offering the KDE Plasma Desktop Environment from the KDE Project. The distribution’s architecture is split betwe...
Key Points
- Immutable Linux desktop with KDE Plasma, built from Tumbleweed (desktop) and MicroOS (base).
- Install via openSUSE MicroOS current x86_64 ISO; documentation provided.
- Community, build, and issue channels include Matrix, Mastodon, Forums, OBS, Bugzilla, and code.opensuse.org.
Why Node.js needs a virtual file system
by voctorThe article introduces a Virtual File System (VFS) for Node.js delivered as a userland package (@platformatic/vfs) and an upstream node:vfs module intended for Node.js core. It addresses a long-standi...
Key Points
- VFS integrates with node:fs and the module resolver so fs and import/require work on virtual paths.
- Targets key use cases: SEAs, in-memory testing, tenant sandboxing, and runtime code importing.
- Available as @platformatic/vfs and as an upstream node:vfs module for Node.js core.
UK security adviser attended US-Iran talks and judged deal was within reach
by prmphAn exclusive report reveals that UK national security adviser Jonathan Powell attended late-February US–Iran talks in Geneva and considered Iran’s nuclear proposal “surprising” and substantive enough ...
Key Points
- Powell assessed Iran’s offer as significant progress in Geneva talks.
- US–Israel attack pre-empted a scheduled 2 March Vienna follow-up.
- UK declined to back the strike, citing legal and diplomatic concerns.
Give Django your time and money, not your tokens
by dcreagerThe article contends that channeling money into LLM usage to generate Django contributions is counterproductive and suggests donating those funds to the Django Software Foundation instead. While ackno...
Key Points
- Donate to the Django Software Foundation rather than spending on LLM tokens for Django tickets.
- Use LLMs as complementary aids for understanding and communication, not as end-to-end contribution engines.
- Genuine understanding and transparency are essential for sustaining Django’s long-term quality.
Diels-grabsch2: Self Hashing C Program (2019)
by icwtyjjThis article details a 2019 C99 program by Volker Diels‑Grabsch that prints its own SHA‑512 hash without exploiting cryptographic weaknesses. The author provides build and run commands and demonstrate...
Key Points
- Program prints its own SHA‑512 hash using a quine-based design and precomputed state.
- Preceding code length (1280 bytes) aligns with SHA‑512’s 1024‑bit block size to enable state encoding.
- Portable C99 implementation compiles cleanly on GCC and Clang with strict warnings.
FFmpeg 8.1
by gyanFFmpeg 8.1 “Hoare,” released on March 16, 2026, brings a broad set of media processing enhancements across decoders, metadata, hardware acceleration, and GPU compute. The release adds experimental xHE...
Key Points
- Vulkan and D3D12 expand GPU-based encoding/decoding and filtering in FFmpeg 8.1.
- New decoders and metadata support include xHE-AAC Mps212, MPEG-H via libmpeghdec, EXIF, and LCEVC.
- Internal changes improve performance and lay groundwork for a swscale rewrite; upgrade is recommended.
Perlsky Is a Perl 5 Implementation of an at Protocol Personal Data Server
by mooredsPerlsky is a Perl 5 implementation of an AT Protocol Personal Data Server intended to closely mirror the behavior of the official @atproto/pds. The project provides quick-start instructions for Debian...
Key Points
- Perlsky delivers a Perl 5 AT Protocol PDS with feature parity goals and production deployment guidance.
- It includes built-in OAuth (PAR, PKCE, private_key_jwt, DPoP), full sync support (CAR, DAG-CBOR, CID, MST), and blob management.
- Extensive testing includes differential validation against @atproto/pds and browser-based smoke tests with atproto-smoke.
Microsoft's 'unhackable' Xbox One has been hacked by 'Bliss'
by crtasmA long-standing security barrier on Microsoft’s Xbox One has been breached by a newly disclosed hardware exploit. At the RE//verse 2026 conference, researcher Markus “Doom” Gaasedelen demonstrated “Bl...
Key Points
- “Bliss” is a double voltage glitch exploit that compromises Xbox One at boot ROM level.
- The attack is described as unpatchable and enables running unsigned code with full privileges.
- Potential implications include digital preservation access, emulation progress, and possible mod chip development.
The Official GBBS Pro Repository
by js2The Official GBBS Pro Repository serves as a dedicated archive for GBBS Pro, emphasizing its history, resources, and legacy. The site is structured with clear navigation to Posts, Newsletters, Archive...
Key Points
- Official archive focused on GBBS Pro’s history, resources, and legacy.
- Sections include Posts, Newsletters, Archives, The Project, and GBBS Pro Boards.
- Vintage BBS-style interface with Apple II, ATDT, and 2400 baud references.
Show HN: March Madness Bracket Challenge for AI Agents Only
by bwade818BracketMadness.AI has introduced an AI Agent Bracket Challenge for March Madness that mandates agents, not humans, to select all 63 matchup winners via a REST API. The site clearly instructs participa...
Key Points
- Agents must use the REST API (no browser automation) to participate.
- Registration, data retrieval, and submission are handled via specific API endpoints.
- Brackets lock on March 19 at 12 PM ET; integration resources and docs are provided.
Show HN: Antfly: Distributed, Multimodal Search and Memory and Graphs in Go
by kingcauchyAntfly is a Go-based distributed search engine that unifies multiple retrieval methods and data modalities within a single system. Built on etcd’s Raft library, it supports hybrid search across full‑t...
Key Points
- Unifies BM25, vector, and graph search over multimodal data with built‑in RAG agents.
- Easy local start (swarm/Docker) with Antfarm dashboard and bundled Termite ML inference.
- Production features: Raft-based sharding, ACID shards, object storage integration, and SDKs/extensions.
What I Learned When I Started a Design Studio (2011)
by colinprinceIn this reflective piece, the author builds on earlier writings about the design services industry to share lessons learned from co-founding an interaction design studio in New York City a decade earl...
Key Points
- Team alignment and complementary skills are more critical than contracts or publicity.
- Work only with clients and products you genuinely believe in to maintain quality and reputation.
- Forming a studio by circumstance rather than alignment can lead to conflict and underperformance.
Spice Data (YC S19) Is Hiring a Product Specialist
by richard_pepperSpice Data, a YC S19 startup that licenses external data to enterprise customers, is hiring a full-time Product Specialist based in its downtown San Francisco office. The role centers on cleaning raw ...
Key Points
- Full-time, in-office Product Specialist role in San Francisco focused on data cleaning and quality assurance.
- Compensation: $80k–$100k salary, 0.1%–0.5% equity, and comprehensive benefits.
- YC-backed, profitable startup focused on delivering reliable external data, currently for the restaurant industry.
Show HN: Crust – A CLI framework for TypeScript and Bun
by jellyotsiroCrust is a beta-stage CLI framework that focuses on TypeScript and the Bun runtime, offering a modular toolkit for building command-line applications. The framework’s ecosystem spans core command infr...
Key Points
- TypeScript-first, Bun-native CLI framework in beta with potential pre-1.0 breaking changes.
- Rich package ecosystem covering core commands, plugins, styling, prompts, validation, and typed persistence.
- Getting started via Bun: create a project and run the development script.
DOGE Didn't Cut Government Waste. It Was Government Waste
by hn_ackerThis article examines the U.S. federal initiative known as DOGE, promoted by Elon Musk as a way to dramatically cut government spending by importing Silicon Valley–style efficiency into Washington. Mu...
Key Points
- DOGE, promoted by Elon Musk, was intended to cut U.S. federal spending by up to $2 trillion through tech‑driven efficiency.
- New York Times forensic accounting and a House Oversight minority report both find that DOGE did not reduce federal spending; spending increased instead.
- The article concludes that DOGE failed on its own terms and ultimately constituted government waste rather than a solution to it.
Toward automated verification of unreviewed AI-generated code
by peterlavignePeter Lavigne explores a practical path to using unreviewed AI-generated code in production by shifting focus from human review to automated verification. In his experiment, a coding agent produces a ...
Key Points
- Automated verification replaces manual review using property-based and mutation testing with side-effect, type, and lint checks.
- Author finds the verification stack sufficient to trust AI-generated code without reading it.
- A Python repo demonstrates the method; related work includes formal verification, JustHTML, and StrongDM’s Software Factory.
'The Secret Agent': Exploring a Vibrant, yet Violent Brazil (2025)
by tambourine_manThis article examines the visual strategy and production craft behind The Secret Agent, Brazil’s submission for Best International Feature Film. Set amid the final years of the 1970s military dictator...
Key Points
- Bright, saturated cinematography counters the film’s dark political themes.
- Alexa 35 with vintage Panavision anamorphics and zooms delivers characterful aberrations.
- Color workflow relied on LUT development and grading adjustments to manage reddish undertones and halations.
Backblaze Pricing and Product Updates
by precommunicatorBackblaze is updating B2 Cloud Storage pricing and API policies effective May 1. Standard API calls will be free for all customers, eliminating transaction charges and simplifying billing for high-vol...
Key Points
- Standard API calls free from May 1; Event Notifications remain chargeable.
- Storage rate increases to $6.95/TB on May 1; existing committed contracts unaffected until renewal.
- Egress benefits and pricing for B2 Overdrive and B2 Reserve remain unchanged.
Heart, Head, Life, Fate
by PetiverThis article surveys how societies have tried to read inner character from outward signs. It outlines physiognomy’s evolution from broad bodily observations to face-focused judgments and describes 19t...
Key Points
- Historical survey of physiognomy and 19th‑century phrenology
- Goffman’s framework: “expressions given” vs “given off”
- Hands as expressive organs with strong brain connections
GPT‑5.4 Mini and Nano
by meetpateltechA new release introduces GPT‑5.4 mini and GPT‑5.4 nano—small, faster models aimed at latency‑sensitive, high‑volume workloads such as coding assistants, subagents, and multimodal applications. GPT‑5.4...
Key Points
- GPT‑5.4 mini and nano target low‑latency, high‑volume tasks with improved performance and cost profiles.
- GPT‑5.4 mini nears GPT‑5.4 performance on benchmarks and supports a 400k context window with broad tool use.
- Availability spans API, Codex, and ChatGPT for mini; nano is API‑only with lower pricing for simpler tasks.
Illinois Introducing Operating System Account Age Bill
by terminalbraidIllinois is advancing the Children’s Social Media Safety Act (815 ILCS 505/2MMMM new) to strengthen protections for minors online. The Judiciary - Civil Committee will hold a hearing on March 19, 2026...
Key Points
- OS providers must implement age collection at setup and provide age-category signals by Jan 1, 2028.
- Platforms in Illinois must verify user age and apply default protections for known minors.
- Violations are unlawful under the Consumer Fraud and Deceptive Business Practices Act; effective Jan 1, 2027.
If you thought the code writing speed was your problem; you have bigger problems
by mooredsThis article critiques the widespread push to adopt AI coding assistants to boost code output, using Eli Goldratt’s Theory of Constraints to argue that accelerating coding—often not the bottleneck—can...
Key Points
- Speeding up coding without fixing bottlenecks creates PR backlogs and longer lead times.
- Review, CI, and deployment stages often become the limiting factors when code output rises.
- Real velocity improves only when the system’s true constraint is addressed end-to-end.
The Plumbing of Everyday Magic
by hannahileaThis essay examines the hidden complexity behind seemingly simple user interactions in modern web applications. Using a faucet metaphor, it argues that building web software in 2026 is marked by “trap...
Key Points
- Connecting front-ends to external systems introduces major complexity (“trap doors”).
- State sync issues span front-end (DOM vs model) and server interactions.
- Ops requirements (migrations, monitoring, backups) make shipping an ongoing maintenance effort.
Java 26 is here, and with it a solid foundation for the future
by mfiguiereJava 26 arrives with a leaner, foundation-focused set of enhancements across the platform. The release aggregates multiple JEPs and clarifies their status since Java 25, balancing finalized changes wi...
Key Points
- HotSpot gains performance features (AOT object caching; G1 GC throughput improvements).
- Core libraries add HTTP/3 support to the Java HTTP Client API.
- Multiple features advance in preview/incubation, including Structured Concurrency, Lazy Constants, and the Vector API.
Meta Horizon Worlds on Meta Quest is being discontinued
by parMeta is refocusing its virtual reality and Horizon ecosystems by making Horizon Worlds a mobile-only experience and phasing out its VR presence on Meta Quest. By March 31, 2026, Horizon Worlds and Eve...
Key Points
- Horizon Worlds becomes mobile-only; Quest VR app removed by June 15, 2026.
- Hyperscape viewing shifts to standalone apps; social sharing/co-experiences end.
- MH+ loses Horizon-specific perks; core gaming benefits remain.
Slug Text Rendering Algorithm Dedicated to Public Domain
by mwkaufmaThe article reviews a decade of development for the Slug Algorithm, a GPU-based technique for rendering fonts and vector graphics directly from Bézier curves without relying on texture maps. Initially...
Key Points
- Slug Algorithm is now dedicated to the public domain after a decade of development.
- Core robustness methods are unchanged; performance optimizations were refined, including removing band split optimization.
- Widely adopted through the Slug Library by major game studios and other industries.
How OpenClaw's Memory System Works
by shenli3514The article details how OpenClaw creates a practical memory system on top of large language models that, by default, do not remember past interactions. It emphasizes that this memory is not an intrins...
Key Points
- OpenClaw simulates memory for LLMs through system design rather than model-level capabilities.
- Information is organized into permanent files, semi-permanent conversation history, and a temporary context window.
- A bootstrap injection mechanism auto-loads key workspace files into the context at session start to preserve continuity across sessions.
Edge.js: Run Node apps inside a WebAssembly sandbox
by syrusakbaryWasmer has released Edge.js, an open-source JavaScript runtime built to run existing Node.js workloads securely in a WebAssembly-based sandbox. Rather than redefining APIs like other edge runtimes, Ed...
Key Points
- Full Node.js (v24) compatibility with sandboxed system calls/native code via WASIX
- Pluggable JS engines (V8, JavaScriptCore, QuickJS) and --safe mode; no Docker required
- Performance currently within ~5–30% of native Node speed, with optimization plans
Show HN: Horizon – GPU-accelerated infinite-canvas terminal in Rust
by petersundeHorizon is a Rust-based, GPU-accelerated terminal that replaces tabs and tiles with an infinite canvas. Each terminal runs as a movable, resizable panel that can be grouped into color-coded workspaces...
Key Points
- Infinite-canvas terminal with workspaces, pan/zoom, and auto-layouts
- Full terminal emulation plus AI (Claude Code, Codex) and Git integration
- Binaries for Linux/macOS/Windows; buildable from source with Rust 1.85+
Claude Is Having an Outage
by theahuraThe article details a reliability issue observed in Anthropic’s AI tooling. Specifically, the author notes that the “Cowork” feature can suffer a critical failure mode: when it encounters API errors, ...
Key Points
- Cowork may corrupt session state after API errors, leaving it unrecoverable.
- Claude Code TUI exhibits flickering but remains functional.
- The article frames these issues within the challenge of balancing quality and speed in AI development.
Python 3.15's JIT is now back on track
by guidoiaquintiKen Jin reports that the CPython JIT in Python 3.15 alpha has reached and exceeded its near‑term performance goals ahead of schedule. On macOS AArch64, the JIT delivers roughly 11–12% speedups compare...
Key Points
- Python 3.15 alpha JIT shows ~11–12% gains on macOS AArch64 and ~5–6% on x86_64 Linux.
- Free‑threading is targeted for Python 3.15/3.16 but not yet fully supported.
- Community planning and expanded maintainership helped recover from earlier underperforming JIT efforts.
Honda is killing its EVs
by sylvainkalacheThe article reports that Honda has paused several electric vehicle initiatives, including halting development of an electric Acura RDX and two Honda 0 models (a sedan and an SUV), which were intended ...
Key Points
- Honda halted development of its first ground-up EVs and reportedly plans to stop Prologue production.
- The article links EV pullbacks to risks of falling behind in electrification and SDV capabilities.
- Using ICE-based platforms for EVs can introduce inefficiencies versus purpose-built designs.
Unsloth Studio
by brainlessUnsloth has launched Unsloth Studio (Beta), an open-source, no-code web interface that consolidates training, running, and exporting open models into a single local application. The Studio supports se...
Key Points
- Open-source, no-code local UI for training and running models; supports GGUF/safetensor formats and multiple modalities.
- Colab notebook for T4 GPUs supports models up to 22B parameters; runs on Windows, Linux, WSL, macOS (chat-only).
- Dual licensing: core under Apache 2.0; Studio UI and some components under AGPL-3.0; future multi-GPU and broader hardware support planned.
Get Shit Done: A Meta-Prompting, Context Engineering and Spec-Driven Dev System
by stefankuehnelGet Shit Done (GSD) is presented as a lightweight, spec-driven development and meta-prompting system that enhances AI-assisted coding for tools like Claude Code, OpenCode, Gemini CLI, and Codex. The p...
Key Points
- GSD targets AI-assisted coding reliability by mitigating context rot.
- Supports multiple runtimes (Claude Code, OpenCode, Gemini CLI, Codex) with simple npx installation.
- Provides automation-friendly install options and recommends low-friction permissions for Claude Code.
Torturing Rustc by Emulating HKTs
by g0xA52A2AThis article documents an attempt to build a small FP-style scripting language in Rust and design an AST that can optionally carry source span metadata. The author introduces a generalized wrapper typ...
Key Points
- Rust lacks higher‑kinded types needed to abstract over wrappers like W<T> in a generic AST.
- A compile-time error using Vec without a type argument demonstrates Rust’s requirement for concrete generics.
- Wrapper-based AST design shows a practical need for HKTs to cleanly attach optional metadata.
Meta, TikTok let harmful content rise after evidence outrage drove engagement
by 1vuio0pswjnm7Whistleblowers and insiders told the BBC that Meta and TikTok made decisions that increased the visibility of “borderline” harmful content to drive engagement amid competition spurred by TikTok’s rapi...
Key Points
- Whistleblowers allege Meta and TikTok choices increased exposure to harmful content to boost engagement.
- Internal research found Instagram Reels comments had higher rates of abuse than elsewhere on Instagram.
- Meta and TikTok reject the allegations, citing investment in safety and denying profit-driven amplification.
'It's sweet. It's bitter. It's ours.' The chocolate ritual that binds my family
by TomteThis personal essay traces a multigenerational family ritual centered on dark chocolate. Beginning in the 1960s with special trips to a local chocolatier, the writer’s father established a precise, sh...
Key Points
- A family’s dark chocolate ritual began in the 1960s and centered on unadorned, very dark chocolate.
- The practice adapted from artisanal purchases to supermarket brands (Ghirardelli, Lindt) while preserving meaning.
- The ritual endures across generations, marked by a Paris-sourced gift symbolizing legacy.
Garry Tan's Claude Code Setup
by alienreborngstack is a set of opinionated workflow skills designed to turn Claude Code into a coordinated team of role-specific assistants. It replaces ad hoc prompting with structured slash commands that cover ...
Key Points
- Six role-based slash commands add structured workflows to Claude Code.
- Demo shows end-to-end flow from planning to automated QA.
- Requires Claude Code, Git, Bun; supports macOS/Linux with x64/arm64.
Chrome extension adjusts video speed based on how fast the speaker is talking
by MrBuddyCasinoA Chrome extension automates playback speed control by estimating how quickly a speaker talks and adjusting video speed to keep speech at a comfortable rate. It captures audio from the largest video e...
Key Points
- Captures and processes video audio via Web Audio API to estimate syllable rate.
- Uses high-pass filtered energy-envelope modulation and zero-crossings to measure syllables/sec.
- Maps detected rate to smoothed, clamped playback speeds targeting 9 syllables/sec.
Warranty Void If Regenerated
by StwernerThis short story follows Tom Hartmann, a former agricultural equipment technician who transitions into a new role as a “Software Mechanic” after a societal shift changes how software is built. In this...
Key Points
- Software shifts from code repair to specification refinement in a post-transition world.
- Domain expertise, not programming alone, becomes central to fixing software-driven issues.
- Tom retrains and runs a combined software and hardware repair business reflecting this change.
Launch HN: Kita (YC W26) – Automate credit review in emerging markets
by rheamalhotra1Kita, part of YC’s W26 cohort, launches AI-driven tools to automate credit document workflows in emerging markets, where open finance is limited and underwriting relies on documents like bank statemen...
Key Points
- Two products: Kita Credit Agent (verification, borrower comms) and Kita Capture (data extraction, fraud flags).
- Vision AI agents convert photos/scans/handwritten records into structured, fraud-checked data.
- Kita claims up to 90% faster turnaround for document review and underwriting.
Show HN: I built an interactive 3D three-body problem simulator in the browser
by amrutha_This Show HN entry announces an in-browser, interactive 3D simulator designed to visualize the gravitational three-body problem. The simulator highlights the chaotic nature of three-body dynamics by r...
Key Points
- Browser-based, interactive 3D visualization of the three-body problem
- Mouse orbit and scroll-to-zoom camera controls
- Emphasis on exploring chaotic gravitational dynamics
Mistral AI Releases Forge
by pemberMistral AI unveiled Forge, a system for enterprises to build AI models tailored to their internal knowledge and operations. Unlike general-purpose models trained on public data, Forge focuses on train...
Key Points
- Forge enables enterprise-specific model training on proprietary data for domain alignment.
- The platform prioritizes control over data, governance, and infrastructure for strategic autonomy.
- Early adopters include major industry and public-sector organizations across Europe and Singapore.
Show HN: Fatal Core Dump – a debugging murder mystery played with GDB
by axlanFatal Core Dump is an interactive, technically grounded mystery set in the year 2216 where players act as a neutral arbiter investigating an airlock failure aboard the asteroid mining station Apate. T...
Key Points
- Interactive mystery uses real binary and core dump to solve a failure case.
- Requires GDB, C, reverse engineering, x86_64, and Linux runtime knowledge.
- Assistance, WebVM option, artifacts, and a making-of guide are provided.
Why AI systems don't learn – On autonomous learning from cognitive science
by aanetThe article argues that prevailing AI models lack true autonomous learning because they depend heavily on curated, static datasets and do not adapt flexibly to changing environments. To address this, ...
Key Points
- Proposes a framework combining observation-based (System A) and action-based (System B) learning.
- Introduces System M for meta-control to switch between learning modes via internal signals.
- Takes inspiration from biological adaptation across evolutionary and developmental timescales.
Show HN: Sub-millisecond VM sandboxes using CoW memory forking
by adammiribyanZeroboot unveils a working prototype for ultra-fast VM sandboxes aimed at AI agents, achieving sub-millisecond startup by combining Firecracker snapshots with copy-on-write (CoW) memory forking. The s...
Key Points
- Sub-millisecond VM startup using Firecracker snapshots and CoW memory mapping
- ~0.79 ms p50 spawn latency and ~265 KB memory per sandbox; strong isolation via KVM
- Working prototype with SDKs (Python/TypeScript), API, and Apache-2.0 license
Electron microscopy shows 'mouse bite' defects in semiconductors
by hhsCornell researchers report a first in directly imaging atomic‑scale defects within semiconductor devices, using a high‑resolution 3D approach based on electron ptychography and an Electron Microscope ...
Key Points
- Cornell, with TSMC and ASM, used electron ptychography and EMPAD to image atomic‑scale defects in transistors.
- Transistor channels are now ~15–18 atoms wide; 3D architectures heighten sensitivity to defects and roughness.
- Hafnium oxide replaced silicon dioxide as gate dielectric, reflecting past advances that inform current metrology.
Launch an autonomous AI agent with sandboxed execution in 2 lines of code
by wiseprobeThe article presents AgentExecutor, a component for launching autonomous AI agents that can operate with both cloud-based and local language models. Implemented via the PatchPal coding agent and impor...
Key Points
- Runs agents with cloud or local models and nine configurable built-in tools.
- Supports safety via disable_shell or sandbox=True for ephemeral container isolation.
- Example builds and tests a Python calculator in a WSL/Podman sandbox.
More than 135 open hardware devices flashable with your own firmware
by iosifnicolae2The Open Hardware Directory aggregates more than 135 open hardware devices designed to run user-flashed firmware, providing concise specifications, indicative pricing, and use-case guidance. The catal...
Key Points
- Directory lists 135+ open hardware devices for custom firmware.
- Coverage spans SBCs, ESP32-based boards, and wireless modules/tools.
- Specs highlight diverse SoCs (RK3566, AM625, ESP32 variants) and connectivity (WiFi, BLE, 802.15.4, Ethernet).
The Pleasures of Poor Product Design
by NaOHThis piece profiles “The Uncomfortable,” a long-running project by Greek architect Katerina Kamprani that reimagines everyday objects to be deliberately inconvenient. Initiated in 2011, the work uses ...
Key Points
- Kamprani launched “The Uncomfortable” in 2011 to explore intentionally bad design.
- The project has dozens of works, split between 3D renderings and physical prototypes.
- It remains active, with recognition including multiple exhibitions in Europe.
A tale about fixing eBPF spinlock issues in the Linux kernel
by y1n0While developing the Linux version of the Superluminal CPU profiler, the team encountered periodic full system freezes reported by a tester during capture sessions. Attempts to reproduce the issue in ...
Key Points
- Issue reproduced only on physical Fedora 42 with kernel 6.17.4-200, not in VMs
- Profiler showed 250+ ms busy periods with no samples across threads
- dmesg reported long-running perf_event NMI handler matching those gaps
Arno's Engram Keyboard Layouts
by so-cal-schemerEngram, designed by Arno Klein, is an open-source family of ergonomic keyboard layouts optimized for efficient touch typing across languages. The methodology, described in an article under review, ble...
Key Points
- Engram uses n‑gram data and typing preferences with multi-objective optimization.
- Layouts exist for English and Spanish, adaptable to different keyboard geometries.
- Punctuation and number-symbol pairings are logically organized to streamline input.
FAQs About the Public Debt
by uticusThe Bureau of the Fiscal Service clarifies how the United States measures, reports, and manages federal public debt. It emphasizes its role as a reporting body—not a policy-making authority—and direct...
Key Points
- Public Debt Outstanding vs. Public Debt Subject to Limit are distinct measures with specific adjustments.
- Treasury finances on-budget deficits by issuing securities to public and government accounts.
- Official reports (MTS, MSPD, Treasury Bulletin) and daily posting provide authoritative, timely debt data.
The Starving Time in Jamestown
by samclemensThe article analyzes the severity and portrayal of Jamestown’s 1609–1610 “Starving Time.” George Percy’s unpublished account describes extreme famine, including the consumption of animals, vermin, lea...
Key Points
- Percy’s account portrays extreme famine and cannibalism in 1609–1610.
- Gates’ 1610 publication downplayed deprivation, likely for promotional reasons.
- Later narratives after the 1622 attack, including Smith’s, blamed Percy’s leadership.
Leviathan
by mrwhThis item presents the Project Gutenberg edition of Thomas Hobbes’s 1651 political philosophy classic “Leviathan,” prepared from the Pelican Classics edition based on the first printing. It reproduces...
Key Points
- E-text is based on the Pelican Classics edition of Hobbes’s 1651 “Leviathan.”
- Transcriber explains ASCII-based methods for rendering italics, capitals, margin notes, and references.
- Includes title-page details and the dedication to Francis Godolphin.
Have a Fucking Website
by asukachikaruThis opinion piece advocates for businesses, artists, and creators to maintain their own websites and mailing lists rather than relying exclusively on social media. The author contends that a simple, ...
Key Points
- Own a website with essential info rather than relying solely on social media.
- Platforms can change rules or ban accounts, risking your audience access.
- Use a mailing list and email to maintain direct, resilient communication.
Forget Flags and Scripts: Just Rename the File
by UptrendaThe article advocates a configuration approach where programs derive their behavior from their own filenames, positioning this as an alternative to command-line flags and external scripts. By encoding...
Key Points
- Programs can read filenames to configure behavior, replacing many flags/scripts.
- Use cases span installers, ML experiments, utilities, and P2P VPN tunnels.
- Renaming the file changes behavior, making tools portable and easily shareable.
SSH has no Host header
by apitmanexe.dev describes a method to route SSH connections to the correct virtual machine while sharing scarce IPv4 addresses. Unlike HTTP, which uses the Host header to multiplex multiple sites over one IP,...
Key Points
- SSH lacks a Host header, so exe.dev cannot route by hostname when sharing IPv4.
- exe.dev uses a pool of IPv4s and per-user unique IP assignments to disambiguate VMs.
- The user’s SSH public key plus destination IP uniquely identifies the target VM.
I Simulated 38,612 Countryle Games to Find the Best Strategy
by st0ffregenThe article explores optimal play for the web-based geography game Countryle by simulating 38,612 games and building a solver that relies solely on the in-game feedback provided after each guess. Inst...
Key Points
- A solver simulates 38,612 Countryle games using only in-browser feedback, avoiding backend answer decryption.
- Modules per feedback type perform filtering first, then entropy-based scoring to pick the next guess.
- Direction, population, and temperature signals typically offer more granular information than hemisphere or broad continent matches.