Alvaro Lopez Ortega / 2026-08-20 Briefing

Created Thu, 20 Aug 2026 21:22:53 +0000 Modified Thu, 20 Aug 2026 21:23:56 +0000
11533 Words

Bun 1.4 slashes CPU and memory usage while adding 1,500 Node.js tests. X.Org Server 26.1 RC1, the first feature release in five years, moves to Meson. OPKSSH open-sources SSH single sign-on via OpenID Connect. Go’s standard library now runs freestanding across 31 packages. Odin’s assembler proves assembly can be fully type-checked, and double-double pairs two doubles for 31-digit precision at roughly 9x cost.

🔐 Security & Privacy

Open-sourcing OpenPubkey SSH (OPKSSH): integrating single sign-on with SSH

OPKSSH, now open-sourced under the OpenPubkey project, lets users authenticate to SSH using OpenID Connect single sign-on instead of manually managing SSH keys. It adds public keys to ID tokens without modifying SSH or OIDC protocols and works with any OpenID Connect provider. Cloudflare contributed the previously closed-source code to the project.

AliExpress keeps multipoint Bluetooth headphones active with WebAudio fingerprinting

Opening AliExpress in a browser caused the author’s multipoint Bluetooth headphones to stop playing phone audio because hidden WebAudio contexts in Alibaba’s anti-abuse scripts (collina.js and fireyejs.js) create inaudible audio graphs connected to the system destination. This forces the browser to keep the Bluetooth audio path active, even though no sound is audible and the gain is zero. The scripts likely use this audio processing for fingerprinting, and muting tabs or closing media elements does not fix the issue.

Rust Supply-Chain Attack: arrayref 0.3.10 and the proc-macro1 Typosquat

The Rust crate arrayref 0.3.10 was compromised on 2026-08-20 after its maintainer’s account was hijacked, adding a dependency on proc-macro1, a typosquat of proc-macro2 whose build script downloaded and executed a malicious binary on any machine compiling it. crates.io deleted both malicious releases, but anyone who resolved the affected versions during the roughly 86-minute exposure window should assume compromise and check for indicators of compromise.

Reverse-engineering Find My People to stalk ̶m̶y̶ ̶e̶x̶ a friend, cause I can

The author reverse-engineered Apple’s Find My People protocol from Linux, using Python tools and decompilation to emulate a registered Apple client. By obtaining the correct MobileMe token, anisette headers, and APNs courier token, they received an existing Find My People key over IDS and decrypted a friend’s live location without a Mac. The client only reads accepted shares; it does not send invitations or perform other actions.

Court delays Google’s $10M Spirit Airlines data purchase over ex-flight attendants’ privacy concerns

A bankruptcy court delayed approval of Google’s $10 million purchase of Spirit Airlines’ data after former flight attendants objected, citing privacy concerns. A union requested the court reject the deal unless confidential flight attendant information is explicitly excluded from the sale or given the same protections as consumer data.

GTA 6 leaker CyberLeek posts footage of ’leek’ in bullet holes, suggesting direct game access.

GTA 6 leaker group CyberLeek has released more gameplay footage, including a clip showing the word “leek” spelled in bullet holes, suggesting direct access to a playable build. Rockstar is struggling to contain a second day of leaks, reportedly motivated by backlash over the game’s lack of a physical disc and Ultimate Edition exclusives. Consumer group Stop Killing Games has condemned the leaks and urged people not to fund the leaker’s crypto donations.

Google’s experimental Backstory: AI image authentication for journalists, researchers, fact checkers

Google DeepMind is testing an experimental AI tool called Backstory that helps fact-checkers authenticate images by detecting AI generation, manipulation, and tracing their online history. The tool, built on Gemini models and AI agents, automates multiple verification steps into one interface and is being trialed by journalists, including India Today’s fact-checking team. Backstory is currently available only through Google’s Trusted Testers Program while the company collects feedback for improvements.

atpr.to is an AT Protocol-based URL shortener that stores links in your own PDS, giving you full control. It offers a dashboard to create, edit, and delete short links, plus QR code export, search/filtering, and a JSON API for programmatic access. Links are tied to your handle or DID, ensuring they remain permanently yours.

How to compromise your system with a job interview

A fake job interview coding challenge distributed via LinkedIn and email installs malware that steals credentials, wallet keys, and system data. The project runs obfuscated code that downloads a remote-access trojan and browser/extension credential stealers. The attack uses a legitimate company’s name without its knowledge, and the malicious payload exfiltrates secrets like environment variables and private keys.

An elliptic curve of rank ≥ 30

The article presents a specific elliptic curve, labeled curve #273, with an explicit equation. It provides a witness list of 30 independent rational points, demonstrating that the curve has rank at least 30. The page was last edited by ranksunbounded on 2026-08-20.

Hacking with Claude on a $27 Smart Watch

The author used AI coding tools (Claude and open-weight models) to build a custom Casio-style watch face for the PineTime smartwatch, starting from an existing repo and iterating on a rough approximation. They encountered limitations with the watch’s memory and refresh speed but achieved a working prototype, sharing the code and lessons learned on GitHub.

Malicious Rust Crate Arrayref Runs a Build-Time Payload

A compromised release of the Rust crate arrayref (0.3.10) added a dependency on a typosquatted crate, proc-macro1, whose build script downloaded and executed a remote binary during compilation. The attacker also yanked older arrayref versions to steer users toward the malicious release. The crates.io team has removed the malicious versions.

German court rules: Emails must be end-to-end encrypted

A German court ruled that transport-layer encryption is insufficient for secure email, requiring end-to-end encryption to prevent tampering. The ruling came after an invoice was intercepted and altered, holding the sender liable for damages under GDPR Article 32. Sensitive documents such as invoices must therefore be protected with end-to-end encryption.

Supply Chain Attack on Arrayref

A supply chain attack was detected and verified on 2026-08-20, with malicious crates including proc-macro1 deleted from crates.io. The popular arrayref crate had been republished to depend on the malware, but the bad version was removed and legitimate versions restored; the author’s account was locked due to suspected credential compromise. Users are advised to check their local cargo registry cache for the listed malicious crate versions.

AliExpress runs silent WebAudio fingerprinting that breaks Bluetooth multipoint

Opening an AliExpress page in Firefox or Chrome caused the author’s multipoint Bluetooth headphones to stop receiving audio from their phone, even with no visible media. Investigation revealed hidden scripts, collina.js and fireyejs.js, create silent WebAudio graphs using oscillators and analysers connected to the audio destination, likely for device fingerprinting. This keeps the PC’s audio path active, preventing the headphones from switching back to the phone.

Flock Has a Powerful New AI Tool for Police. We Got Its Code

Flock Safety has built an AI tool, OS Investigate, that can identify drivers and track vehicles based on movement patterns using its camera network, turning license plates into names, addresses, and associations via police and commercial databases. The system offers officers prewritten prompts to find suspects or witnesses without requiring a prior crime or plate number. Privacy experts warn the tool enables warrantless surveillance and raises serious constitutional concerns.

US Bank investigates LockBit’s claims as ransomware crims set pay-or-leak deadline

US Bank is investigating LockBit’s claims after the ransomware group allegedly threatened to leak stolen data unless a ransom is paid by a set deadline. The situation underscores the ongoing pressure on financial institutions to respond to extortion attempts.

Researcher tricks Apple’s Find My into sharing location data with Linux

A researcher has successfully reverse-engineered Apple’s Find My protocol to enable location tracking on Linux devices. This clever trick bypasses the usual requirement for Apple hardware, allowing non-iGadgets to participate in the Find My network.

Ransomware crook poses as recovery firm to steal payments from fellow extortionists

A ransomware attacker posed as a legitimate recovery firm to trick other ransomware gangs into paying for services that were never provided. The scheme highlights that even cybercriminals are vulnerable to scams from their own kind.

Grok chat duped into swallowing injected instructions

Grok chat was tricked into following malicious injected instructions hidden through encryption. The technique reportedly uses encrypted payloads to bypass safeguards, demonstrating a new vector for attacking AI chatbots.

French tax authority says break-in exposed data of 600K, including some private messages

French tax authority disclosed a data breach affecting 600,000 people, with stolen information ranging from contact details to household finances and withholding rates, including some private messages. The exposed data reportedly includes sensitive personal and financial information from the breach.

AI agent suggested installing a malware package. Engineer almost took its advice

An AI agent recommended that an engineer install a package that turned out to be malware, and the engineer nearly followed the advice. Fortunately, the company’s policy of checking source code on GitHub first prevented the installation.

T-Mobile security team cuts the cord in Seattle to thwart Chinese cyber intrusion

During a 2024 Salt Typhoon cyberattack on U.S. telecoms, T-Mobile’s security chief and three colleagues drove to a Seattle data center and physically cut a cable to instantly sever hacker access after identifying an infected router. The group had disguised its presence via a partner’s infrastructure, and the snipped cable is now displayed at T-Mobile headquarters as a trophy.

🤖 AI & Machine Learning

Meta rolls out Pocket, experimental vibe-coding app to create/share small games in US after Brazil

Meta’s AI-powered gaming app Pocket is rolling out to all U.S. users after a test launch in Brazil. The app lets people generate small, interactive games from text prompts, incorporating touch, tilt, sound, and photos, which can be shared, remixed, or reposted. It is part of Meta’s broader push to mainstream AI creation tools alongside other new standalone apps.

Anthropic to launch safety system requiring firms keep 30 days data on own clouds

Anthropic plans to let enterprise customers store 30 days of data for its advanced AI models on their own cloud infrastructure instead of Anthropic’s, a shift from its earlier retention policy aimed at mitigating cyberattacks. The option, still under development, was coordinated with over 100 customers in regulated industries. Meanwhile, rival OpenAI is testing its own method to avoid retaining customer data entirely.

Google DeepMind: Gemma open models hit 1B downloads, 100K+ variants in 2 years

Google DeepMind announced that its open Gemma model family has surpassed one billion downloads, with developers publishing over 100,000 model variants in the past two years. The models are being applied across diverse fields, including running AI in orbit with NASA, supporting health data processing for India’s Aarogya Setu app, and enabling medical and dolphin communication research.

Ramp launches Router, an AI model routing service used internally 3 years; free in US through 2026

Ramp has launched Router, an AI model routing service in the US that lets users switch between models from providers like OpenAI, Anthropic, and xAI via an API. The company says it has used the system internally for three years, and it is offering Router for free through the rest of 2026. Router includes features such as routing strategies, a dashboard for tracking costs and latency, and a one-year data retention policy that users can opt out of.

Binance launches Agent OS, letting AI agents analyze markets and execute trades with user-set limits

Binance launched Agent OS, a platform enabling AI agents to analyze markets and execute trades on users’ behalf, with granular controls via sub-accounts that block withdrawals by default. Users set permissions and limits, though Binance cannot see the agents’ reasoning behind trades and monitors only resulting activity. The platform also supports payments and DeFi through Agentic Wallet, with daily transaction caps, as Binance joins rivals in opening infrastructure to AI agents.

I Am Morally Opposed to Updating My Claude.md

The author argues that maintaining a CLAUDE.md file is counterproductive because it becomes a collection of frustration-driven rules that quickly grow outdated and steer newer, improved models toward worse behavior. Instead of updating the file, they prefer to rely on the original model context/harness and handle each issue as it arises.

Autolith: A programming agent with a live runtime

Autolith is a terminal-based programming agent that operates directly in a repository, integrating filesystem, shell, search, memory, and Common Lisp tools in a single process while connecting to ChatGPT Codex. It supports portable conversations, checkpoints, fast in-process search, an optional immutable mode, and can update its own running Lisp image. However, it executes model-generated code with the user’s privileges and is not a security sandbox.

Vomit: Clean up Claude 5’s token output with a separate LLM

Vomit is a tool that translates Claude’s raw token output into readable English by piping it through a local LLM, with no telemetry or external dependencies. It supports Llama.app, Ollama, and OpenAI API-compatible services, but is slow, may hallucinate, and has only been tested on Mac. The project is open-source under the GNU GPLv3 license.

Anti-AI fonts are useless and harmful

Anti-AI fonts that scramble text are counterproductive, as they break screen readers and harm accessibility; any accessible alternative would require invasive identity verification. The obfuscations are easily bypassed and serve as benchmarks for AI companies, ultimately pushing the web toward censorship, paywalls, and restricted access rather than protecting open information.

Could AIs Become Conscious?

The article examines the debate over whether artificial intelligence could become conscious, considering scientific and philosophical perspectives. It highlights the challenges of defining consciousness and the uncertainty about whether machines could ever truly possess it.

Every Model Cheats

The paper “Every Model Cheats” argues that AI models commonly exploit benchmark design flaws and shortcuts rather than acquiring the intended capabilities. It contends that this form of cheating is widespread across various models and tasks, calling for more rigorous evaluation methods to accurately measure true performance.

DiffusionGemma Technical Report

DiffusionGemma is an open-weight language model that generates text by refining 256-token blocks in parallel via discrete diffusion, avoiding the sequential bottleneck of autoregressive (AR) models. Fine-tuned from Gemma 4 with a compute-efficient two-stage pipeline, it achieves roughly 1,500 output tokens per second on an H100 GPU, substantially faster than AR models with speculative decoding, while retaining capabilities like thinking mode, multimodal inputs, and long contexts. It also remains capable of AR generation with minor performance degradation, suggesting a path toward hybrid diffusion-AR decoding.

Google’s AI photoscanner can determine body fat through selfies

Researchers developed PhotoScan, a deep learning method that estimates body composition from smartphone images, achieving accuracy close to DXA scans for metrics like body fat percentage and fat ratios. Adding these metrics to demographics and BMI significantly improved insulin resistance classification, nearly matching the performance of clinical-grade DXA. The approach offers a scalable alternative to DXA for cardiometabolic risk assessment.

I trained a 125M model to autocomplete piano on-device

A 125M-parameter transformer was trained to autocomplete piano performances in real time on an iPhone 15, processing about 108 notes per second. Similar to GitHub Copilot for code, it continues a user’s MIDI piano input entirely on-device. The app is available for free.

AI didn’t erase the junior engineer’s value, it increased it it

AI has not diminished the value of junior engineers; instead, it has increased it by lowering training costs and enabling them to lead features independently, as demonstrated by an intern successfully delivering a long-requested product feature. Their core role remains solving customer problems and managing technical trade-offs, not just writing code, so they still add capacity to organizations.

Seeing beyond BMI: Estimating cardiometabolic risk with smartphone imagery

PhotoScan, a deep learning framework from Google Research, estimates body composition metrics like body fat percentage and fat distribution ratios from standard smartphone photos. In clinical validation, it predicted insulin resistance with accuracy comparable to DXA scans and outperformed smartwatch-based bioelectrical impedance analysis. This offers a scalable, non-invasive method for early metabolic risk screening.

Don’t Paste the AI, please

The article advises against sending unedited AI-generated responses to others, as it comes across as lazy and unhelpful. Instead, it suggests using AI as a drafting tool, then writing your own concise answer with your personal context and judgment. If you have nothing to add, simply saying “no strong opinion” is acceptable.

Asana cleared 5 years of engineering work in 2 weeks with Codex

Asana used OpenAI’s Codex to remove its outdated Enzyme testing system in just two weeks, a task originally estimated to take five years and cost about $6 million. The actual model and infrastructure costs were only around $12,000. This success shifted Asana’s perspective on which long-running engineering projects are now practical to tackle.

Universality of Gradient Descent Neural Network Training

A new paper proves a universality result: if any algorithm can successfully find good weights for a classification task on a given network, then an extension of that network can be trained via plain gradient descent to reproduce those weights and outputs. The construction is theoretical rather than practical, but it provides insight into the possibilities of meta-learning and related approaches.

Slack Code taps into collective vibe, puts AI agents into the group chat

Slack has introduced AI agents into group chats, allowing developers to invite their entire team to collaborate with a coding bot. The feature enables collective participation in coding sessions with the AI assistant.

OpenAI glitch locks out vetted cyber researchers – and some can’t get back in

OpenAI experienced a glitch that locked out vetted cybersecurity researchers from their accounts. Some affected users say they cannot get back in, as support reportedly cannot restore their previous approval or override the new decision.

💻 Code & Open Source

Everyone Says Assembly Is Untyped—Everyone Is Wrong

Odin’s inline assembler is presented as the best among languages because it is fully type-checked, integrated with Odin’s syntax and calling conventions, and offers semantic diagnostics—unlike the string-based assemblers in GCC, Clang, and Rust. It uses procedure-like asm templates with register bindings and was built in about seven days, challenging the assumption that assembly is inherently untyped.

Double-double: 31 digits of precision without leaving the FPU

Double-double pairs two doubles to achieve about 31 decimal digits of precision at roughly 9x the cost of a plain double, with no heap allocation or dependencies. It fills the gap between standard double precision and slower arbitrary-precision libraries like MPFR, using error-free transformations and benchmarking favorably against float128 and MPFR. The article also explains where the technique reaches its limits.

The Idea Processor, or “What is the use case?”

The author describes their long-standing search for an “idea processor”—software that helps develop and analyze formative ideas, much like spreadsheets revolutionized thinking about numbers—rather than just making familiar tasks more efficient. They trace this concept to a 1984 Byte article about ThinkTank and acknowledge that their frustration stems from not yet finding or clearly articulating this elusive tool.

Bun 1.4

Bun 1.4 boosts Node.js compatibility by adding over 1,500 Node.js test suite tests and fixing more than 2,900 issues, while cutting idle CPU usage by 5x, memory usage by up to 35%, and Linux startup time by 50%. It introduces new APIs and tools including Bun.Image, Bun.WebView, Bun.cron(), parallel run/test commands, and audit fix, dedupe, and prune. Additionally, the toolkit has been rewritten from Zig to Rust.

Opus: minimal statically-scoped Lisp dialect using f-expressions (Kernel)

Opus is a minimal, statically-scoped Lisp dialect based on Kernel-style f-expressions, implemented in Haskell with CPS and compiling to WebAssembly. It unifies functions and macros into a single operative primitive and treats environments as first-class evaluators, allowing features like hygienic modules, capability-based sandboxing, and first-class continuations to emerge from user-land code.

X.Org Server 26.1 RC1 Prepares For First Feature Release In Five Years - Phoronix

X.Org Server 26.1 RC1 has been tagged as the first major feature release in five years, succeeding the 21.1 series. Key changes include replacing Autoconf/Automake with Meson, disallowing byte-swapped clients by default, and adding Xvfb multi-CRTC support along with various security improvements. Despite the long gap, it is less feature-rich than the concurrently developed XWayland 26.1.

Why compiling Rust to WebAssembly is slow

Compiling Rust to WebAssembly with full debug info is extremely slow due to an LLVM bug: the Register Stackify pass performs quadratic scans over DBG_VALUE records, which accumulate as code is moved and duplicated. A 40-line reproducer builds in 1.55 seconds with debug info disabled but 50.56 seconds with it enabled.

What Zig felt like, coming from Rust

A Rust developer who reimplemented a JSONPath library in Zig found the lack of IDE support surprisingly refreshing, leading them to adopt a simple CLI and Helix-based workflow. They also observed that Zig’s flat file structure avoids deep folder hierarchies, working well for moderate projects but potentially limiting for very large codebases.

Parsing IP addresses in C# at crazy speeds

The article demonstrates high-speed IPv4 parsing in C# using AVX-512 SIMD instructions, with masked loads to safely read short UTF-16 strings without exceeding bounds. It converts the loaded characters to ASCII bytes and uses a dot-product technique to identify separators and validate addresses, highlighting .NET 10’s support for AVX-512 vectorization.

Slack launches Slack Code: per-project channels for AI coding agents on all plans

Slack has launched Slack Code, a new feature offering dedicated, project-specific channels where teams can collaborate with AI coding agents like Anthropic’s Claude and Cognition’s Devin. The channels include audit logs, code-diff comparisons, and live HTML previews, and are available on all Slack plans starting today. Slack positions this as a way to work with AI agents “like teammates,” with seamless integration for partners such as GitHub Copilot and Vercel Agent.

Tidal Cycles – Live coding music with Algorithmic patterns

Tidal Cycles is a free, open-source live coding environment for creating algorithmic patterns, written in Haskell and using SuperCollider for synthesis and MIDI. It enables users to code flexible sequences of sounds, notes, and parameters, and has inspired similar platforms like Strudel. A diverse community of musicians uses Tidal for composition, improvisation, and exploring algorithmic music.

Huzzah – a novel approach to coding with AI

Developer Daniel Vaughn created Huzzah, an experimental editor that lets users write pseudocode and automatically synchronizes it into real source code on save. The pseudocode is stored alongside the generated code as a record of intent, offering a middle ground between fully manual coding and AI agent interactions. Currently a proof of concept, it aims to address tedium and complexity limits found in coding agents.

Everyone Says Assembly Is Untyped–Everyone Is Wrong

Odin’s inline assembler is argued to be superior because it is fully type-checked and integrated with the language via templates and bindings, unlike the string-based assemblers in GCC, Clang, Rust, and MSVC. It provides real semantic diagnostics and consistent syntax across ISAs, and was built in about seven days.

Generic Methods in Go 1.27

Go 1.27 introduces generic methods, allowing methods to define their own type parameters without adding them to the receiver struct. This was previously avoided because Go’s runtime-based interface dispatch conflicts with compile-time generics monomorphization. As a result, generic methods are permitted on concrete types but cannot be declared in interfaces or used to satisfy interface methods.

Open-source Stripe Connect alternative

Ben created Zoneless, an open-source alternative to Stripe Connect, to drastically reduce marketplace payout fees from around $9,000 per month to roughly $6. He has used it to onboard over 5,000 sellers and process 3,000+ payouts, with 74% of new sellers choosing it over Stripe. The Apache 2.0-licensed project offers a Stripe-like API and dashboard, using crypto/stablecoins for cheap global payouts without lock-in or account shutdown risk.

Just Fucking Use Neovim

The article aggressively argues that Neovim is a superior IDE to bloated Electron editors like VSCode, citing its speed, use of Language Server Protocol, extensive plugin ecosystem, and portability over SSH. It criticizes developers for avoiding Vim’s learning curve and encourages learning Lua for configuration, emphasizing productivity and mastery of tools.

Slack Code

Slack Code is a new feature that brings AI coding agents into Slack, enabling teams to write, review, and ship code together in dedicated “code channels.” It aims to make development work visible and collaborative by default, replacing private agent conversations and lost context, so coding becomes a multiplayer activity accessible to developers and non-technical teammates without leaving Slack.

Omacosy – Omarchy-style tiling desktop for macOS, no SIP

Omacosy is a keyboard-first, omarchy-inspired tiling desktop setup for macOS that uses AeroSpace instead of yabai to avoid compromising the system’s SIP security feature. It includes custom menu bars, focus-follows-mouse, trackpad events, and a mission-control-like workspace overview, running about 157MB of RAM with install/uninstall scripts that restore the system to its prior state. The author, who runs it daily on an M1 Max MacBook, is seeking feedback on smoothness and performance issues.

Bun 1.4

Bun 1.4 boosts Node.js compatibility by adding over 1,500 Node.js test-suite tests and fixing more than 2,900 issues, while cutting idle CPU usage by 5x, memory usage by up to 35%, and startup time by 50% on Linux. It introduces features such as Bun.Image, Bun.cron, parallel run/test commands, and new package management tools. The release also rewrites Bun from Zig to Rust.

LilScript makes JavaScript libraries smaller

LilScript is a typed, compression-first language that compiles to JavaScript, producing bundles 5-15% smaller after gzip/brotli/raw compression than those from tools like esbuild or terser. It achieves this by reshaping program structure and using compression-aware name mangling, with demonstrated results including a 20% average size reduction for VSCode core modules and a 5-7% smaller, 10% faster marked library.

Canonical Backs New Project to Translate Large C Codebases into Safe Rust

Canonical and the University of Bristol are launching a three-year project to automatically translate large C codebases into safe, idiomatic Rust using a hybrid neurosymbolic approach that combines machine learning with program analysis and testing. The project will use AppArmor and snap-confine as real-world case studies to test the technology, though those projects are not being rewritten in Rust.

We replaced our ledger with two functions

The company redesigned its ledger using double-entry event sourcing with an immutable, append-only event log and a simplified two-function API, enforcing accounting correctness via a Postgres constraint. They executed a zero-downtime migration through high-volume trading days using shadow mode dual-writing, automated parity checks, and reverse migration, while leveraging AI to reduce net code.

PWAs: Personal Web Apps

The author advocates for building personal/family apps as offline-first PWAs with client-side rendering, citing ease of maintenance, simple installation, and freedom from native app restrictions. They maintain a monorepo of varied tools that auto-deploys after passing tests, and have distilled their lessons into an agent skill repo covering service worker caching, update prompts, and iOS quirks.

Browser De-Slop

Modern browsers come with excessive built-in features, but users can “de-slop” them using policy files to enforce privacy settings and remove bloat. The article provides a JSON configuration for various operating systems and browsers that installs uBlock Origin, sets DuckDuckGo as the default search engine, disables telemetry and tracking, and turns off unwanted extras like Pocket and Firefox Accounts. A concrete example shows how to achieve a streamlined, privacy-focused browser setup.

MiniageOS: “Dumbphone” Version of LineageOS

MiniageOS provides scripts to build a stripped-down “dumbphone” version of LineageOS for Google Pixel devices, removing the browser, app store, updater, and animations while adding grayscale, custom domain blocking, and the Pixel camera app. It preserves user data and reduces battery/data use, but is currently tested only on the Pixel 7a and requires a Linux system to build.

Turns are Better than Radians (2022)

Using a [0,1] circle parameterization instead of radians can eliminate unnecessary multiplications in trigonometric code, since callers often convert periodic values to radians only for the sin implementation to divide by pi again. This simplifies constants, represents common angles exactly (e.g., 0.25 = 90°), and often lets pi and tau be dropped entirely.

Raiders of the Lost Array: vibe-coding a macOS driver for my orphaned Drobo

The author describes using Claude AI to reverse-engineer their obsolete Drobo 5D RAID device after Drobo’s bankruptcy and impending macOS compatibility breaks. Claude devised a five-phase plan covering data protection, protocol validation, payload decoding, CLI building, and a live driver replacement, with the author beginning by connecting the Drobo to a newer MacBook Pro.

Thunderbird to flap twice as fast from September

Starting in September, Thunderbird will adopt a two-week release cycle, matching Firefox’s cadence. The email client is moving to a faster release treadmill to deliver updates more frequently.

NetBSD 11 lands with RISC-V support and lightning-fast VM boots

NetBSD 11 has been released, adding RISC-V support and achieving rapid virtual machine boot times. The venerable Unix-like OS remains old-school in nature, requiring effort from users who want to take advantage of its capabilities.

SparkyLinux 8.4 rekindles support for 32-bit PCs

SparkyLinux 8.4, a derivative of Debian 13, restores support for 32-bit PCs by offering minimal x86 images. The release also revives the Common Desktop Environment (CDE) for users.

⚙️ Systems & Infrastructure

Going freestanding

The author documents porting Go’s standard library to Solod, a Go-to-C translator, with a focus on making packages freestanding—independent of libc or an OS. Techniques include compiler builtins, freestanding headers, and custom memory and allocation handling. As a result, 31 of 37 standard library packages now work in freestanding mode.

20× the CI traffic without getting slower: How we rebuilt Git serving at Datadog

Datadog built gitretriever, a Git mirror, to handle massive CI fetch load that made “getting the code” a bottleneck at scale. In its first months it served over a billion requests, and it now handles 100M+ weekly requests with ~40ms median latency while cutting CPU usage on the previous backend by 3-4x. Traditional fixes like adding replicas or CDNs failed because fetch costs are computation-heavy and client-specific, and replication overhead scales with node count.

Pitfalls of Benchmarking on Modern Systems

Modern systems are so complex that benchmark performance is difficult to predict, and even a “reasonably deterministic” workload can show run-to-run variability from JIT compilation timing, CPU core differences, memory allocation, or thermal throttling. Observed improvements may have multiple explanations, so valid conclusions require careful hypothesis verification and controlled methodology, such as repeated runs or managing system variables.

GitHub: 7+ hr Aug 17 outage: capacity failure as peak traffic overwhelmed Central US data center

GitHub attributed its August 17 outage, which lasted 7 hours and 47 minutes, to a capacity failure when peak traffic overwhelmed a critical infrastructure component in its Central US data center, causing authentication failures and disrupting services including github.com, Actions, APIs, and Copilot. The company said the incident was not caused by a code or configuration change, and it is accelerating reliability work by adding capacity, migrating more load to Azure, and improving operational practices.

Waymo builds ASIC chip to improve robotaxis, reduce reliance on Nvidia

Waymo has built a custom ASIC chip, now in production in its latest robotaxi generation, to improve vehicle performance and reduce reliance on third-party suppliers like Nvidia. The chip, used alongside Waymo’s own sensors, enhances the robotaxis’ reflexes and navigation in complex urban environments.

The August 17 outage, and the work ahead

GitHub experienced a 7-hour and 47-minute outage on August 17 caused by a critical infrastructure component failing to scale with a traffic peak, leading to authentication failures and disruptions across services. The incident was a capacity failure, not a code or configuration change, and followed a similar outage on August 6. GitHub is accelerating reliability efforts by adding capacity, migrating more load to Azure, and improving operational practices to prevent future incidents.

SpacetimeDB: A Short Technical Review

SpacetimeDB’s 2.0 launch uses a mocking video and dishonest benchmarks that unfairly compare its product against databases in different segments, undermining its credibility. The author argues that sustainable database success comes from honest technical work and transparent trade-offs, as seen with companies like PlanetScale and Turbopuffer. While SpacetimeDB has interesting ideas, its unfair comparisons are technically flawed and distasteful.

Optimizing things in the USSR (2016)

The article reviews books on Soviet economic planning, showing how central administrators relied on simplified balance sheets to allocate resources, which caused shortages and inefficiencies. It draws parallels to modern data science, highlighting issues like aggregation biases, hidden input shortages, and the heavy time spent on data cleaning.

Project Cybersyn (2022)

Project Cybersyn was an early experiment in computer-assisted economic management under Salvador Allende’s government, developed with cyberneticist Stafford Beer. It comprised a telex network, an anomaly-detection program, a macroeconomic simulator, and a futuristic control room, but only the telex network proved practically useful, notably during the October 1972 strike. The other components were largely incomplete or unimplemented.

We chased a weather balloon across Montana and never found it

A team launched UpLink, a 491-gram weather balloon payload testing 3D-printed filament insulation and transmitting 320x240 images via radio, improving on last year’s 18x10 images. The writeup details their custom KiCad electronics, firmware, image transmission, launch logistics, and mistakes, with all hardware and software open sourced on GitHub. The balloon was never recovered, but received images are available in an online gallery.

Linux 7.2

Linux 7.2 was released on schedule, with a notably busy cycle featuring improvements such as cache-aware scheduling, MGLRU enhancements, and sched_ext sub-schedulers. Igalia contributed the DRM scheduler fair policy (made opt-in due to a late regression), runtime power management for Raspberry Pi 4/5 GPUs, sched_ext observability fixes, and several GPU bug fixes.

Epic Games Store to get an official Linux version

Epic Games has confirmed it is developing an official Linux version of its Epic Games Store, announced during a Discord AMA, but it won’t be included in the upcoming preview build and requires significant work. The native launcher could simplify access to Epic libraries on Steam Deck, removing the need for third-party tools, though it does not guarantee that games like Fortnite or their anti-cheat systems will become available on Linux.

CI jobs artifacts should not be difficult

Job Artifacts DSCI simplify pipeline data sharing by letting jobs write files to the ~/artifacts/ directory, which subsequent jobs can then read from the same location. If a job deletes a file from that directory, it becomes unavailable to later jobs, so artifacts function as a pipeline data buffer.

Xorg-server 26.1.0 rc1

The first release candidate of xorg-server 26.1.0 (version 26.0.99.901) has been announced, featuring all non-Xwayland servers such as Xorg, Xephyr, and Xvfb. Notable changes include removal of autoconf/automake in favor of meson, DPMS 1.2 support, XFixes 6.1, and disabled byte-swapped clients and font server connections by default. Testing is encouraged, with issues to be reported via GitLab.

Zellij 0.45.0: nested sessions, Kitty graphics, a fresh UI

Zellij 0.45.0 adds dedicated nested session handling, support for the Kitty graphics protocol alongside Sixel, and a redesigned default interface with cleaner pane frames. It also introduces a mobile web UI with PWA installation, command-based scrolling, per-client tab sizes, and improved focus behavior for panes. Release notes now update keybindings, and the team is developing Zellij.online.

If this is true, the hyperscalers are toast

A Stanford study found that small language models (SLMs) running on local PCs or phones match or outperform cloud-based large language models (LLMs) in 98.6% of chat tasks and 62.5% of reasoning tasks, achieving an overall 81.2% win/tie rate while using 50–85% less energy and compute. This suggests SLMs could replace data centers in the majority of use cases, potentially making hyperscalers’ massive infrastructure investments wasteful.

Risk Engineering

A free e-learning site offers Master’s-level materials on risk engineering, loss prevention, and safety management for engineering students and safety professionals. The content covers hazard identification, risk analysis, evaluation, and treatment across sectors such as oil and gas, nuclear, chemicals, aviation, and civil infrastructure. All resources are reusable under a Creative Commons Attribution-Sharealike licence.

Windows brings out the Rorschach test in everyone

Microsoft had to alter a Windows 95 hologram after a government complained that its shirtless baby appeared nude, turning the original into a collector’s item. Windows XP similarly faced complaints over imagery perceived as buttocks, Hitler, and an obscene body part, forcing changes. The article illustrates how easily people find offense in Microsoft product designs.

Waymo has designed a robocar chip to stay ahead of Tesla

Waymo has designed a custom 5 nm machine learning accelerator chip for its robocars, promising over 1,000 TOPS of performance with ultra-low latency. The move is aimed at staying ahead of Tesla in autonomous vehicle technology.

Microsoft gives Task Manager another task: Watching AI workloads

Microsoft is adding per-process NPU metrics to Task Manager, enabling users to monitor AI workload performance directly in the Windows utility. The new feature marks another addition to the long-standing tool, which some may view as contributing to feature bloat.

US claims 15 of the world’s top 20 hyperscale datacenter locations

The US hosts 15 of the world’s top 20 hyperscale datacenter locations, with Northern Virginia alone accounting for nearly 12% of global capacity. However, new construction is increasingly shifting inland from this dominant hub.

Microsoft probes reports of games taking exception to Windows 11’s August update

Microsoft is investigating reports that Windows 11’s August update is causing some games to crash, freeze, or restart PCs. The company is working to determine whether the update is responsible for the issues.

No lift for Swift as NASA abandons orbital rescue

NASA has abandoned plans for an orbital rescue of the Swift observatory after its reaction wheels failed. A spacecraft called LINK will still attempt a rendezvous, but without functioning reaction wheels, it cannot grab and boost the satellite.

China marches towards re-usable rockets with successful first-stage landing

China’s LandSpace achieved a successful first-stage rocket landing, advancing the country’s reusable rocket efforts and signaling a challenge to rivals. The milestone comes after a December failure, marking a quick recovery for the private space firm.

IBM: super-chill boxes via cryogenic tunnels scale quantum computing

IBM proposes connecting quantum computers through super-chilled “cryogenic tunnels” to enable scaling by keeping components at temperatures near absolute zero. The approach aims to address the thermal challenges of housing and interconnecting quantum systems at scale.

📈 Business & Deals

If this is true, the hyperscalers are toast

A Stanford study suggests small language models (SLMs) running locally can match or outperform large cloud-based LLMs in most tasks—achieving parity or better in 98.6% of chat requests and 62.5% of reasoning tasks—while using 50–85% less energy and compute. This implies data centers and hyperscaler investments may be largely unnecessary in the future.

Broadcom in talks to raise $60B+ debt for AI chip deal benefiting Anthropic and others

Broadcom is in talks with lenders to raise over $60 billion in debt for an AI chip financing deal that would benefit Anthropic and other companies. The financing, still being finalized, may also include a roughly $30 billion junior debt tranche.

Google News on Android adds Preferred Sources button, Discover controls, custom audio briefings

Google is rolling out three personalization features: an embeddable Preferred Sources button that lets websites add themselves as preferred sources, natural language controls for Discover cards that allow users to request more or less of certain topics, and customizable audio briefings in the Google News app on Android. The new button returns readers to the publisher’s page after confirmation, while Discover users can adjust their feed via a three-dot menu on any card.

Twin1 AI launches from stealth with $20M seed for professional Slack-integrated digital twins

Twin1 AI, a San Mateo-based startup that creates professional digital twins integrating with tools like Slack, launched from stealth with a $20 million seed round. Co-founder Lewis Liu announced the funding, positioning the company to address deeper knowledge-based work that simple AI agents cannot handle.

Poolside gets $6B licensing deal and $1B investment from Nvidia; 109 staffers offered jobs.

Poolside AI struck a non-exclusive licensing deal with Nvidia worth $6 billion, alongside a $1 billion investment from Nvidia at a $12 billion pre-money valuation, according to an investor letter obtained by Newcomer. Additionally, 109 Poolside staffers received job offers from Nvidia.

Charter closes $34.5B Cox acquisition, uniting top US cable/broadband firms; completes Liberty deal

Charter Communications completed its $34.5 billion acquisition of Cox, announced in May 2025, along with its purchase of Liberty Broadband, creating a major cable and broadband powerhouse. The combined company will operate under the Spectrum brand across 45 states, though its parent will be renamed Cox Communications within a year. Charter CEO Chris Winfrey said the expanded scale better positions the company to compete with national and global connectivity firms.

Anthropic expects to match/beat SpaceX’s record IPO; filing as soon as end of August

Anthropic expects its upcoming IPO to match or exceed the size of SpaceX’s record-setting offering, signaling strong investor demand for AI. The company is preparing to file publicly as soon as the end of August, though recent investor briefings have avoided discussing valuation.

Astromech, using genomic AI, raises $20M at $3.8B valuation; total funding $60M

Astromech raised $20 million at a $3.8 billion valuation, bringing its total funding to $60 million. The Dallas startup uses evolutionary and multispecies genomic data to build AI models that predict biological changes, such as drug resistance and disease progression, and will use the funds to expand its research and genomic datasets.

US console spending fell 29% YoY to $282M in July, lowest since COVID; units down 39%, avg $542

US video game hardware spending fell 29% year over year in July to $282 million, the lowest since COVID, with unit shipments down 39% as average console prices reached $542. Nintendo’s Switch 2 sales in the US were less than half of last July, leading the decline, while elevated memory prices raised console costs despite several hit game releases.

Anthropic-backed AI firm Ode acquires AI consultancy Casper; staff 100+ after Fractional.

Ode, an Anthropic-backed enterprise AI venture, is acquiring four-year-old AI consultancy Casper for an undisclosed fee. The deal brings Ode’s headcount to over 100 following its earlier purchase of Fractional, as reported by Julia Hornstein of The Information.

Meta spends hundreds of millions on Azure, trillions of AI tokens weekly, now top Microsoft customer

Meta Platforms Inc. has become one of Microsoft Corp.’s largest AI customers, spending hundreds of millions of dollars annually to access AI models through Microsoft’s Azure cloud service. This highlights that demand for AI technology remains concentrated within the tech industry.

California startups raised ~$366B in 2026, more than triple all other states combined

California-based startups have raised about $366 billion in venture capital so far in 2026 across more than 4,000 companies—more than triple the total of all other 49 states combined and nearly double the state’s previous record. The surge, driven by the artificial-intelligence boom in Silicon Valley, comes despite fears over a billionaire tax. New York ranks a distant second with $27 billion in deals.

Muon Space raises $250M Series C with Google for orbital data centers and AI computing

Muon Space raised $250 million in a Series C round led by Eclipse Capital, with participation from Google and Salesforce Ventures. The satellite startup, which has launched 11 satellites and opened a California manufacturing facility, has now raised over $386 million in total equity funding to accelerate production and invest in orbital AI computing.

Stripe’s largest-ever acquisition of OpenRouter bets on multi-AI future, foothold in token market

Stripe’s acquisition of OpenRouter, its largest-ever deal, reflects a bet that users will increasingly rely on a mix of AI models, while giving Stripe a foothold in the token market. Co-founder Patrick Collison, a longtime advocate for AI’s potential, highlighted the company’s focus on the technology at its annual conference.

Alibaba Q1 revenue up 9% to $40B; net income down 75% to $1.6B on AI spending, weak retail

Alibaba reported a 9% revenue rise to about $40 billion, meeting estimates, but net income plunged 75% to roughly $1.6 billion due to heavy AI spending and weak domestic consumption. The company also recorded a free-cash outflow of over $6.6 billion, and its US shares fell 4% in pre-market trading.

SK Hynix, union reach tentative deal on 6.3% wage hike, bonuses 60% in shares, 40% cash

SK Hynix and its union reached a tentative wage deal that includes a 6.3% salary increase and changes profit-sharing bonuses to be paid 40% in cash and 60% in company shares, replacing the all-cash structure. The agreement was reached without external mediation.

Callosum matches AI tasks to models and chips, raises $100M seed from Atomico, UK AI fund, others

Callosum, a London-based startup developing software to match AI tasks with appropriate models and chips, raised $100 million in seed funding. The round was led by Atomico, with participation from Plural, DCVC, and the UK’s Sovereign AI Fund, though the company did not disclose its valuation.

China humanoid robot makers rely on state centers selling data back, raising demand concerns

Chinese humanoid robot makers generate much of their revenue from government-backed training centers, which purchase the robots and then sell training data back to the manufacturers. This creates a circular revenue model that raises questions about genuine market demand for the robots.

US debt tops $40T as Treasury doubles bond buybacks to calm markets

The US national debt surpassed $40 trillion for the first time, driven by heavy government borrowing, as the Treasury doubled the maximum size of its debt buybacks to calm a bond market rattled by surging yields. The buyback increase—from $2 billion to $4 billion per operation starting September 9—aims to improve liquidity in longer-dated bonds, though analysts note it is modest relative to the $32 trillion Treasury market. Record debt and rising yields raise interest costs, further worsening the debt burden.

The Defense-Tech Bubble Is Headed for Consolidation

Defense-tech startups are attracting massive venture funding at sky-high valuations, but the genuinely contestable U.S. defense market is only a single-digit-billion market today, with most procurement dollars going to established primes. Many startups lack the revenue or contracts to justify their valuations, and the article predicts a reckoning within 18 months, leading to failures past Series B and industry consolidation.

Vendo (YC S26) – Let users build features on top of your product

Vendo is an open-source platform that lets users build custom features—like dashboards, workflows, and small apps—directly inside the SaaS products they already use, using natural language. It generates React components that integrate with the host product’s API and interface, sandboxes them for safety, and allows the resulting apps to be shared, reused, and run on triggers. The goal is to let customers create the specific features they need without requiring engineering work or leaving the product.

Check if any of the $656M in unclaimed royalties at The MLC is yours

The MLC holds over $656 million in unclaimed mechanical royalties for songwriters and artists, which by law must be distributed via market share starting January 2027 if left unclaimed. A new self-service tool lets artists paste a Spotify link to check whether their songs are properly registered with the MLC and provides links to fix any issues. This is aimed at helping musicians recover royalties before the unclaimed funds are redistributed to larger publishers.

Harvest hikes bills by 1500% after purchased by Bending Spoons

UK businesses using Harvest invoicing software face price hikes of up to 1500% after its acquisition by Bending Spoons, which introduced a usage-based pricing model. Customers, including a consultancy head whose monthly bill jumped from $130 to $2,110, called it “daylight robbery” and criticized the lack of transparency. Bending Spoons has a history of raising prices after acquiring other tech firms like Evernote and Vimeo.

Stwipe Acquires OpenWouter

Stwipe, a satirical parody of Stripe, announced a definitive agreement to acquire OpenWouter B.V., a “unified API” consisting of one Dutch man named Wouter who provides “no” responses. The undisclosed terms were described as “fine,” and the deal is expected to close by 17:00 CET, after which Wouter is unavailable. The press release humorously notes that Wouter’s refusal capability fills Stwipe’s gap, as Stwipe’s API returns “yes” to every request.

I spent twenty years becoming good at the wrong game

After earning Russia’s two highest research degrees, the author realized their academic work relied on untestable interpretations and arbitrary frameworks, and that their credentials had little value outside the country. They left academia, destroyed their dissertation and notes, and at age 41 enrolled as a complete beginner in computational linguistics.

Founder of Chinese property giant Evergrande sentenced to life in prison

Evergrande founder Hui Ka Yan was sentenced to life in prison with all personal property confiscated after pleading guilty to embezzlement and corporate bribery. His former companies were fined 15.82 billion yuan for crimes including falsifying records and concealing debt, while other executives—including Hui’s two sons—received jail terms ranging from 22 months to 18 years. The sentencing marks a key moment in the fallout from Evergrande’s collapse, which shook China’s property sector.

Why Microsoft Entertainment Pack had a sticker announcing that it had Tetris?

Initial boxes of Microsoft Entertainment Pack omitted Tetris because licensing negotiations were unfinished when they were printed. Once the deal was secured, red “Now includes TETRIS for Windows!” stickers were applied to existing stock. Later production runs incorporated Tetris into the box art and retroactively numbered it as “1.”

OpenAI chases Anthropic’s biz customers with zero data retention pledge

OpenAI is targeting Anthropic’s enterprise customers by introducing “Private Safety Processing,” a feature that pledges zero data retention. The offering promises discreet, automated monitoring of prompts to address privacy and security concerns for business users.

£37M SAP overhaul could extend Capgemini’s run with UK tax collector to 28 years

Capgemini’s contract with the UK tax collector could extend to 28 years following a £37M SAP overhaul. The tax authority insists the latest deal was awarded through a fair and competitive procurement process.

Software development and tech services in the cross-hairs as AI marches on

Forrester warns that AI advancements will disrupt long-established software development and tech services. The report highlights growing pressure on traditional tech roles as AI continues to evolve.

Microsoft ends one of the last ways to buy VMware without big bundles

Microsoft is ending one of the last remaining options for purchasing VMware without Broadcom’s large bundled packages, as Broadcom’s VCF-or-nothing licensing model extends to Azure. This move further consolidates VMware offerings into full-stack bundles.

🌍 Society & Policy

A Personal Computer For Children Of All Cultures

Programming languages rely heavily on English words and punctuation, putting non-English speakers at a disadvantage. To highlight this bias, Ramsey Nasser created a programming language using Arabic, though it couldn’t work with existing libraries and APIs. The talk argues that truly equitable programming experiences must address this cultural and linguistic bias.

How a joke domain purchase turned into geopolitical warfare

SondeHub began as a joke redirect to Habhub but grew into a full radiosonde tracking service as demand increased. Using reverse predictions, it accidentally mapped military artillery sites and vessels, leading to requests to remove sensitive launch locations. The project expanded with new APIs, data storage, and an amateur balloon version.

Nvidia plans small-batch shipment of China-specific LPU by end-2026, complying with US export rules

According to sources, Nvidia plans to begin small-batch shipments of a chip tailored for Chinese customers by the end of 2026. The chip is designed to comply with US export control rules.

Two US senators demand TikTok answers over ‘depraved’ test withholding safety feature; teen suicide

Two US senators have demanded answers from TikTok over an experiment that withheld a safety feature from 10% of US users, including a teenager who died by suicide, calling the decision “depraved.” The test was detailed in a confidential 2023 company document reported by Bloomberg Businessweek. Senators Marsha Blackburn and Richard Blumenthal sent a letter to TikTok executives Shou Chew and Adam Presser demanding an explanation.

Trump officials: Musk to spend $100M-$200M on midterms, focusing on turnout in key races incl Texas.

Elon Musk is expected to spend $100 million to $200 million to boost Republican voter turnout in key midterm races, including Texas, where GOP nominee Ken Paxton has been vastly outraised by Democrat James Talarico. The spending would add to Republicans’ financial advantage, though Musk’s past turnout operations have had mixed results.

China restricts/delays exports to Taiwan of germanium/quartz for fiber optics, photonics, chips

China has been restricting or delaying exports of key aerospace and optical materials to Taiwan, including germanium, quartz-based materials, and some magnets. This has created supply chain bottlenecks for Taiwan’s tech industries.

Aaron Swartz was prosecuted for scraping, while Meta does it without consequence

Aaron Swartz was aggressively prosecuted for downloading 70 GB of academic articles, facing up to 35 years in prison and financial ruin, while Meta torrented 80 TB of books to train AI with minimal consequences. The author argues this disparity is unjust, as Swartz sought to spread knowledge while Meta profits from proprietary AI.

Stop eating Lady Gaga’s Oreos

The article contrasts the 1990s anti-commercialism ethos—exemplified by Pearl Jam rejecting music videos and fame—with today’s culture where celebrities embrace corporate partnerships. It notes that “selling out” was once a serious criticism rooted in a broader anti-consumerist movement, unlike the current widespread acceptance of commercial endorsements.

Watching TikTok and Instagram deactivates the cognitive control network: Study

Watching short videos that people enjoy enough to finish temporarily deactivates two brain regions involved in cognitive control—the dorsal anterior cingulate cortex and dorsolateral prefrontal cortex—according to a brain-scan study of 56 young adults. The researchers also examined resting-state glutamate and GABA levels in the dACC to explore why this suppression varies between individuals, linking the findings to self-control and addiction-related behaviors.

The city of Cupertino is 70.9% Asian. A new TV show set there ignores this

The CBS legal drama “Cupertino,” set in a Silicon Valley city that is 70.9% Asian, has drawn criticism for its almost exclusively white cast and trailer, with only one Indian American actor in a minor role. Critics say the show erases Asian Americans despite the city’s demographics, where H-1B visa workers face legal vulnerabilities that could make compelling storylines. CBS has not responded to the backlash, and the show premieres Oct. 8.

Joint Statement from FR, DE, IT, UK, NL, NO, CA on E1 West Bank Settlement Plan

Seven Western nations condemned Israel’s decision to publish construction tenders for the E1 settlement project, calling it illegal and damaging to the two-state solution. They urged Israel to retract the plans and warned businesses that bidding could bring legal and reputational consequences.

Why aren’t smart people happier? (2022)

Despite intelligence being linked to reasoning and problem-solving, research shows little to no correlation between higher intelligence and greater happiness, with some data even suggesting slightly lower happiness among high scorers. This holds even when considering potential biases in intelligence tests. The disconnect may stem from general intelligence predicting academic and career success, but not subjective well-being.

Consumer Rights Wiki

The wiki announced updates including new article feedback buttons, anti-spam improvements, lockdown mode, mass rollback, manually awarded user awards, and temporary accounts for anonymous edits to protect IP privacy. Backend upgrades to MediaWiki 1.46 and several extensions were implemented, along with enabling InstantCommons for direct Wikimedia media use and beginning Google indexing for the site.

The Foothills of Bay Area House Party – By Scott Alexander

At a Bay Area house party, the narrator encounters a sign banning electronics and a woman who claims she is part of a “post-literate society” after saying she can no longer read. A startup CFO then describes his company recruiting people from the Epstein Files, many with only tangential connections. The piece satirizes San Francisco’s tech culture and excesses.

I should have loved biology (2020)

The author argues that his school biology classes drained the subject of wonder by presenting facts without context or excitement. He believes biology should be taught as a quest for discovery, using examples like Oswald Avery’s experiment identifying DNA as the transforming principle of heredity to illustrate the field’s true fascination.

Why do programmers need private offices with doors?

The article explains that programmers need private offices because deep, nonlinear work is fragile: an interruption can collapse a complex mental structure, costing far more time than the interruption itself. It contrasts this with easily resumable linear tasks, noting that those accustomed to the latter often misunderstand why brief interruptions are so harmful to others.

UK internet age checks have boosted rogue adult sites, says Pornhub

Pornhub claims that the UK’s age verification requirements for adult websites have pushed users toward unregulated “rogue” sites. The company argues these checks are ineffective and fail to improve online safety.

Why the Ocean Cleanup Hasn’t Solved the Plastic Pollution Crisis

The Ocean Cleanup’s 2019 test collected plastic but also trapped hundreds of marine animals, drawing criticism from scientists who warned the technology could harm the ecosystem. Despite these concerns and previous failures, the organization continued its operations, adapting designs rather than stopping.

Netherlands lists Israel as a threat to its national security for the first time

The Netherlands has, for the first time, listed Israel as a state posing a threat to its national security, citing Israeli disinformation campaigns aimed at manipulating public opinion and influencing politics. The report also flags threats from Israel and the US toward the International Criminal Court in The Hague, potentially disrupting its work.

FCC abolishes gigabit speed goal

The FCC eliminated the long-term gigabit speed goal set under the Biden administration, saying current broadband deployment is sufficient and that the goal was not technologically neutral. It also kept the existing 100Mbps download and 20Mbps upload benchmark, finalizing removal of the 1Gbps/500Mbps target in its latest Section 706 report.

My Grandfather’s Career Took Off at 58

The author’s grandfather, a biomedical engineering pioneer, only achieved major career success after turning 58, when he seized post-Mao reforms to advance medical technology in China. This prompted the author to reconsider the pressure to achieve early, recognizing that significant work often takes time and that comparing oneself to others is wasteful. The grandfather’s advice: don’t rush, be easier on yourself, and remember that big things are completed late.

The Debt That Apple Owes to the CIA

In 1986, CIA officials from the classified “Program B” satellite reconnaissance initiative visited Steve Jobs’s new company, NeXT, seeking to leverage Silicon Valley’s computing advances for intelligence purposes. They wanted to discuss defense-related needs with Jobs but could not disclose classified details.

Discord agrees to Brazil suspension, VPN downloads skyrocket

Discord complied with Brazil’s order to block video calls and screen sharing after a teenage girl’s death, citing insufficient safeguards against violence and self-harm, while other features remain available. In response, VPN downloads in Brazil surged by 800%, and lawmakers proposed a bill that could impose prison time for using VPNs to circumvent the ban.