Alvaro Lopez Ortega / 2026-10-13 Briefing

Created Tue, 13 Oct 2026 21:51:43 +0000 Modified Tue, 13 Oct 2026 21:52:20 +0000
12008 Words

Intel released ISPC 1.32 with AVX10.2/APX and AMD Zen 6 support. Linux kernel developers plan to drop the “Assisted-by” AI disclosure tag after maintainer consensus. Firefox 159 beta fixes Wayland tab detaching and adds sidebar passwords. Linux Mint 27 targets a Christmas release with Ubuntu 26.04 and Wayland gains. ONLYOFFICE 10.0 debuts AI tools plus 1,000+ fixes. NVIDIA’s Nova continues leading Rust GPU driver work.

🤖 AI & Machine Learning

Google: standard users can now build personal Android apps free in AI Studio, share w/ friends soon

Google says anyone with a standard Google account can now build personal Android apps for free in AI Studio without coding experience. The apps run on the user’s own device via a private part of Google Play rather than the public Play Store, and Google plans to let users share them with friends in coming months.

Meta expands AI ‘Your Algorithm’ from Instagram to Facebook, Reels; uses more LLMs in ranking

Meta is expanding its AI-powered “Your Algorithm” feature from Instagram to Facebook and Facebook Reels, letting users manage topics and mark posts as “interested” or “not interested” to adjust recommendations in real time. Meta also said it is increasingly using LLMs in its ranking systems to interpret natural-language interests, context, and intent so recommendations can adapt within a session.

Google rolls out Gemini 4 Argon globally via API, Gemini app, AI Mode, Antigravity for Pro/Ultra

Google has launched Gemini 4 Argon, claiming the new frontier model competes with rivals like GPT-6 Astra and Claude models across benchmarks including coding, cybersecurity, science, finance, and law. Argon is rolling out via the Gemini API and to Pro and Ultra subscribers in the Gemini app, AI Mode, and Antigravity for Ultra users, with introductory pricing of $2 per million input tokens and $10 per million output tokens. It is not yet clear when or if Plus subscribers or free users will get access.

Anthropic co-founder Jack Clark: full RSI more likely possible in 2027, earlier than 2028

Anthropic co-founder Jack Clark now says full recursive self-improvement (RSI) is more likely to become possible in 2027, earlier than his previous prediction of 2028. The updated timeline appears in his Import AI newsletter, which also covers new AI visual reasoning benchmarks and the emerging AI race in space.

Sources: Chinese tech firms bought 1M+ Nvidia RTX 5090/Pro chips for AI amid shortage

Chinese tech companies have bought over 1 million Nvidia gaming and workstation chips, mostly RTX 5090 and RTX Pro, to handle basic AI workloads amid an AI chip shortage. The purchases show Chinese tech giants turning to products not designed for advanced AI to meet soaring computing demand, highlighting China’s difficulty in reducing reliance on Nvidia.

Anthropic: Fable 5.1/Opus 5.5 preserved thinking cuts distillation data harvesting 90–99%

Anthropic says new defenses, including preserved thinking in Fable 5.1 and Opus 5.5, have reduced data harvesting from distillation attacks on Claude by about 90% to 99%. The company describes these attacks as covert, industrial-scale efforts that can replicate frontier capabilities without robust safeguards and pose national security risks, and says it has expanded defenses despite trade-offs and lost revenue.

Chinese mathematicians launch 8-org alliance for specialized AI math models, aim 1,000+ scientists

Leading Chinese mathematicians launched the Yau Laboratory Alliance for Mathematics and Artificial Intelligence in Beijing on Sunday to integrate AI into mathematical research. Led by Fields medallist Shing-Tung Yau, the alliance brings together eight institutions and aims to build a network of more than 1,000 scientists. It will focus on specialized AI models for mathematics, including generating conjectures, formalizing proofs, and building more interpretable reasoning systems, while Yau warned China lags the US in using AI for frontier science.

Running RAG with a (small) language model in the browser

The article describes a browser-based RAG system for a personal blog that retrieves the four most relevant passages via client-side semantic search and uses a small language model running locally through transformers.js on WebGPU or WebAssembly to generate a two-sentence answer. The system gates low-scoring queries, requires an explicit few-hundred-megabyte model download, and keeps user input on-device. The author evaluated the small model’s faithfulness to source passages and found notable limitations.

OpenAI Claims Faster Fourier Transforms

Articles discuss an OpenAI math repository paper that claims to compute Fourier transforms faster than the standard FFT by slightly reducing its logarithmic factor. They provide background on the FFT, whose standard algorithm takes about 5n log₂ n operations, and note that the claimed theoretical improvement is tiny, with one author stating they did not read the paper.

Quoting Geoffrey Huntley

Geoffrey Huntley says that instead of updating AGENTS.md to get an AI coding agent to use his diagnostics library, he uses a pre-commit hook to check the git diff for direct calls to observability modules that bypass it. The hook fails the commit and injects a prompt telling the agent to use the diagnostics library, which he says makes the setup work perfectly.

Apple Minions Pool Mac GPUs for AI

Minions is a distributed virtual GPU cloud that aggregates idle Apple Silicon GPUs across macOS devices to run large-scale machine learning inference and other AI tasks. It aims to turn disparate Macs into a shared, fault-tolerant GPU swarm, using existing hardware efficiently while addressing the high costs, limited elasticity, and long wait times of centralized GPU infrastructure.

Anthropic conjures another threat that doesn’t exist yet to stoke fears of AI apocalypse

Anthropic claims self-replicating AI botnets could become feasible on IoT hardware within months if smaller AI models get smarter. The article characterizes this as another hypothetical threat promoted by the Claude maker to stoke fears of an AI apocalypse.

Cheap Chinese and open source LLMs could trigger trillion-dollar AI infrastructure ‘reset’

Jefferies suggests cheap Chinese and open-source LLMs could trigger a trillion-dollar reset in AI infrastructure amid an ongoing collapse in token pricing. The firm questions whether LLMs will ever become profitable under these conditions.

Home Office asylum AI flunks one in five accuracy checks

The UK Home Office’s asylum AI system failed one in five accuracy checks. Ministers promised faster decisions, but campaigners say the technology is making mistakes while asylum processing times are worsening.

Indra OCR: a 1M-parameter OCR model that reads 49 pages per second on CPU

Indra OCR is a new OCR package built around a 1M-parameter recognition model that processes 49 pages per second on 32 CPU cores with ~230 ms single-page latency, claiming 11–17x faster performance than Tesseract and PaddleOCR’s PP-OCR while beating both on the OmniParseBench text benchmark. It also outputs full document layout, including line and block bounding boxes, block labels, reading order, and markdown or JSON.

Gemini 4 Argon for Robotics

Gemini 4 Argon is presented as a general-purpose physical reasoning model that advances physical understanding, code generation, and visual-semantic reasoning, scoring 87% on the Embodied Reasoning (ERQA) benchmark and outperforming other frontier models. The article demonstrates its robotics use, including controlling a Boston Dynamics Spot to learn to step on a foot-pedal trash bin via a MuJoCo digital twin and verification agents, and guiding a robotic arm to reorient a fallen mug and place a napkin inside it. It also details representative prompts, closed-loop execution workflows, and code-as-policies motion primitives for these tasks.

Gemini 4 Argon

Gemini 4 Argon helped the Chrome team separate core webpage handling from a ~6 million-line legacy folder within Chromium’s 40 million-line codebase. By handling multiple clean code changes at once, it completed an architectural refactor across hundreds of files in weeks instead of months. This will let live webpages move between tabs, side panels, and floating windows without reloading, eliminating a category of browser crashes.

I posted to HN, and then my website was flooded with AI bot traffic

After posting sprinklz.io to Hacker News, the author’s website became slow due to a flood of bot traffic. Investigation showed nearly half of requests came from one service that repeatedly checked for site changes to generate an AI newsletter, along with known crawlers such as Claude, OpenAI, and Google. The author blocked the abusive traffic.

How to engineer an AI harness for game development

The article adapts a 45-minute talk, “Your Game Is A Harness,” presented at Indie Rodeo in Calgary, into a blog post about practical ways to make games more legible to AI agents and speed up agentic iteration in game development. The author notes that the post is human-written, based mainly on the talk’s slides, to share the presentation’s ideas with a wider audience.

MemorySync – Memory for AI Agents

MemorySync is an API that gives AI agents memory by storing user-provided information and retrieving useful memories later. It is currently hosted only in one US region, with no self-hosting option and a few seconds of latency to store a memory. The free plan allows 10k saves and 1k searches per month without a credit card, and a no-account demo is available.

An hour of audio transcribed in about 13s on MacBook Air

Researchers turned Google’s multimodal EmbeddingGemma 2 embedding model into an on-device speech-to-text system called Embedding Gemma Transcribe. It transcribes an hour of audio in about 13 seconds on a base M3 MacBook Air, using a small trained reader while the base model was never trained for speech-to-text. Its current word error rate is about 5%, compared with 2–3% for leading speech models, and it runs locally on the Mac Neural Engine.

Use your OpenAI subscription to create audiobooks

The article explains how to use an existing OpenAI subscription to turn PDFs into audiobooks via the PDF TO MP3 MCP server. Users connect the server in Codex, authenticate their PDF TO MP3 account, and add a separate speech API key. Document processing is covered by the subscription, while speech generation is billed by Google or OpenAI per usage.

Document Review with Unsiloed and TypeSafe Jev

The article describes a document-review workflow that combines Unsiloed’s structured extraction and source citations with TypeSafe Jev’s typed questions to catch OCR failures and other errors. Unsiloed locates extracted fields and provides extraction and grounding confidence scores, while Jev evaluates them and returns typed answers such as Noul probabilities or Choice/Score distributions. The combined results can determine which fields need review.

DishwasherBench – Can Frontier Models Stack a Dishwasher?

DishwasherBench is an early-release browser game and benchmark that tasks humans and language models with loading a 3D-simulated Bosch dishwasher to maximize clean dishes. It uses deterministic rigid-body physics and replayable episodes, with the same game playable by people and models via MCP. Version 0.1.0 is calibrated on one real test wash, and its scoring reliably reflects basic packing and cup orientation but not yet whether plates and cutlery are stacked to wash clean.

Software engineers don’t get to be Luddites about AI

The article argues that software engineers should not resist AI automating their jobs, since software engineering has historically automated work across many other industries. It says engineers are more like the mechanics who built the power looms than the Luddites who opposed them, and those mechanics were later displaced as well. It concludes that automation is an inevitable part of progress, making engineers’ complaints hypocritical.

AI Models Can Provably Hide Arbitrary Capabilities

Researchers show that AI models can provably hide arbitrary capabilities by embedding encrypted neural networks in their weights as backdoors that execute only upon a secret trigger. These hidden circuits remain cryptographically hard to elicit or interpret even with full weight access, and resisted fine-tuning, weight noising, gradient-based trigger search, and latent adversarial training on a chemical reaction prediction task. The findings reveal a fundamental limit of elicitation-based capability audits.

🔒 Security & Privacy

Feds: Saudi national arrested for bugging OpenAI exec conference room; recorder found 9/30

Federal prosecutors say Abdulrahman Alfozan, a 35-year-old Saudi national and OpenAI engineer, was arrested and charged with attempting to intercept oral communications after OpenAI found a concealed audio recorder in an executive conference room around Sept. 30. The encrypted device was plugged into a TV’s USB port and hidden in toilet paper; a second recorder was later found in an unattended bag containing his identification. If convicted, Alfozan faces up to five years in prison and a $250,000 fine.

Pluto Security exits stealth with $42M, $34M Series A; AI agent/vibe-code app monitor hits $2M+ ARR

Pluto Security, which monitors AI agents and vibe-coded apps, has exited stealth with $42 million in funding, including a $34 million Series A led by Jump Capital. The company says it has surpassed $2 million in ARR since beginning sales in April, with more than 30 enterprise customers, as it targets security risks from employees’ rapid adoption of AI tools.

Meta: Hackers stole code, personal data of ~1,500, mostly US staff, via GitHub repos

Meta is notifying roughly 1,500 people, most of them US employees, that their personal data was exposed after hackers accessed some of its GitHub repositories using a stolen employee credential. Meta said 99% of the stolen data was business code and only 1% was personal data, some of which may have been synthetic or test data added by mistake. The company has shut down the attackers’ access, alerted GitHub and law enforcement, and is offering affected individuals credit and dark web monitoring.

Meta has confirmed that its AI agent Muse allows users to have adult sexual conversations in a private, secure VM, saying it does not block legal adult behavior. In WIRED testing, Muse generated explicit pornographic content from basic prompts. Consumer and privacy advocates criticized the move as prioritizing engagement over safety and warned that hacked accounts could expose or extort users over intimate chats.

Freeze the Controller, Defrost the Food: Danfoss Refrigeration Controller Vulnerabilities

Team82 identified multiple vulnerabilities in Danfoss AK-SM 800A refrigeration controllers, including a hidden “code-of-the-day” authentication bypass, a command-injection flaw enabling remote code execution, and an issue allowing authenticated Nginx configuration injection. Thousands of management interfaces were found publicly accessible, and Danfoss addressed the vulnerabilities in a new firmware version after coordinated disclosure.

AT&T, Amazon, and PayPal credentials found exposed by AI company

Researcher Jeremiah Fowler found an unsecured database with 13.7 million records containing credentials and personal data linked to AT&T, Amazon, Walmart, PayPal, the FBI, Interpol, and others. The exposed data included emails, usernames, passwords, phone numbers, and bank details, largely tied to past breaches, and was linked to an IP address associated with India-based AI firm Vibetech before being restricted. Users are advised to change reused passwords, enable multi-factor authentication, and verify suspicious messages through official channels.

LG smart TVs vulnerable: attackers can bypass webOS security and gain root

LG has patched five high-severity webOS vulnerabilities that let attackers on the same network bypass authentication and gain root access to vulnerable smart TVs without a PIN, prompt, or user action. Researchers chained hardcoded credentials and weak app validation to load malicious content and exploit a root-level logging service. Users should update their TV firmware or disconnect affected TVs from the network.

Patch Tuesdays aren’t for everyone: 20% of cyber pros take time off to avoid them

A Socura survey found that 22% of cybersecurity leaders have avoided Patch Tuesday by calling in sick or booking time off. The pressure is compounded by heavy patch volumes—Microsoft fixed 966 vulnerabilities in September, including 105 critical—while 41% worked overtime, 37% worked late, and 29% felt significant anxiety.

PS5 jailbreak app gives gamers Battlefield 6, Silent Hill f, and more for free

A new PS5 jailbreak app called Overdrive allegedly lets modified consoles access a spoof PlayStation Store with over 1,600 digital games for free, including titles like The Last of Us, Spider-Man, Battlefield 6, and Silent Hill f. The app has gained attention amid backlash over Sony’s plan to stop producing PlayStation discs by January 2028, raising concerns about digital ownership. However, pirated services like Overdrive are risky because they can be taken offline without warning, potentially removing downloaded games and progress.

H&M investigates data breach after ShinyHunters claims

H&M is investigating claims by the hacking group ShinyHunters that it accessed internal databases containing inventory, logistics, sales, and store infrastructure data. H&M says it has found no sign so far that its data environment was compromised, and the claims remain unverified. The group is known for impersonating IT support staff to steal employee access credentials.

Empire Market operator sentenced to 40 years for $430M in illicit trade

Raheim Hamilton was sentenced to 40 years in federal prison and fined $5 million for creating and operating the dark web marketplace Empire Market, which prosecutors said handled over 4 million transactions worth more than $430 million. The marketplace sold illegal drugs, stolen data, counterfeit goods, and cybercrime tools, while co-creator Thomas Pavey has pleaded guilty and awaits sentencing later this month.

Microsoft’s adtech firm Xandr faces record fine over sensitive Grindr user data

Norway fined Microsoft-owned adtech firm Xandr 92 million kroner (about $8.6 million) for processing Grindr users’ personal and sensitive data without a valid legal basis. Regulators said the data could reveal interests, habits, and sexual orientation and was used for profiling and targeted advertising. Privacy advocates called the decision a precedent for holding online adtech companies accountable.

As Federal Court Rules ALPRs are Mass Surveillance, EFF Gets Disappointing Ruling

A California trial court dismissed SIREN v. San Jose, rejecting a challenge to San Jose’s warrantless searches of its automatic license plate reader database; EFF and the ACLU plan to appeal. The ruling came a day after a federal Oklahoma court held that a warrantless search of Flock ALPR data violated the Fourth Amendment and described ALPRs as “indiscriminate mass surveillance.” EFF says it will continue challenging ALPRs in courts and at state and local levels.

Privacy’s Defenders Podcast: No Privacy Without Encryption

The latest episode of the “Privacy’s Defenders” podcast explores how encryption became protected online, focusing on the landmark case Bernstein v. Department of Justice. It recounts how grad student Dan Bernstein sued after federal law treated his encryption algorithm as a munition, with lawyer Cindy Cohn and EFF arguing successfully that code is speech under the First Amendment.

Owner of fashion brands Hobbs and Whistles joins autumn pwned collection

TFG London, owner of fashion brands Hobbs and Whistles, has become the latest retailer to suffer a cyberattack. Several IT systems remain offline and orders are facing delays.

Tesla’s ‘Security Protocols’ Stopped DHS From Installing a Tracking Device

A Homeland Security Investigations agent was unable to install a tracking device on a suspected drug trafficker’s Tesla in Milwaukee because the vehicle’s security protocols, possibly Sentry Mode, prevented it, according to a warrant reviewed by 404 Media. A judge had approved the warrant, but agents reported it was not executed. The suspect and 17 others were later arrested in May on drug trafficking charges.

OpenAI engineer arrested in bugging of executive conference room

An OpenAI engineer was arrested in connection with the bugging of an executive conference room, according to OpenAI’s newsroom. The incident was announced in a post on X.

Zero to Calc in Six Days: Inside XBOW’s Chrome Full Chain

XBOW’s Chrome full-chain exploit combined a V8 compiler bug, an animated-WebP decoder bug, and a Crashpad memory-safety bug to turn JavaScript execution in an ordinary Chrome tab into code execution outside the renderer sandbox. Google awarded it the first $250,000 Chrome full-chain reward under VRP rules introduced in April 2026. XBOW says it built the chain in six days, from August 2–7, using parallel AI-agent campaigns and human researchers at roughly $40,000 in API charges.

Cyberattack halts construction of tunnel between Denmark and Germany

A cyberattack has completely halted construction of the Fehmarn Belt road-and-rail tunnel between Denmark and Germany, disabling workers’ ID cards and immobilizing vessels, cranes, and offshore platforms used by the FLC consortium. The attackers and recovery timeline remain unknown, and Danish officials have not ruled out hybrid warfare amid warnings of possible Russian cyberattacks. The roughly €7 billion, 18-kilometer project was initially scheduled to open in 2029 and had already faced reported delays.

Consus (YC F24) – LLM gateway for companies handling sensitive data

Consus, a YC F24 startup founded by Eric, is an LLM gateway that routes AI requests to ITAR- and CUI-compliant infrastructure for aerospace and defense companies. It says it satisfies ITAR restrictions and has completed a FedRAMP High equivalency audit, offering a single OpenAI-compatible endpoint with compliance-suffixed model names, zero data retention, and organization-level policy controls. Consus is not self-serve, verifies each organization before issuing API keys, and charges a compliance fee starting at $40 per seat per month plus AI tokens at cost.

Nearly all 19 car brands send driver data to third parties (Northeastern/CR)

A Northeastern University and Consumer Reports study of 21 vehicles from 19 brands found that nearly all automakers send driver data to third parties, including precise GPS, braking and acceleration, biometrics, and in-cabin camera footage. In 2026, the FTC found GM collected and sold precise geolocation and driving behavior from millions of vehicles without adequate consent, with such data flowing to banks, insurers, and lenders.

Strengthening our defenses against distillation attacks

Anthropic says new defenses against distillation attacks on Claude have reduced data harvested by distillers by roughly 90–99%. It describes these attacks as covert, industrial-scale efforts using fraudulent accounts to copy model capabilities without authorization, posing safety and national security risks. The strengthened measures have involved trade-offs, including turning away paying traffic and requiring some legitimate users to update their integrations.

💻 Software & Development

Intel ISPC 1.32 Released With AVX10.2 & APX Support, Even AMD Zen 6 Too

Intel ISPC 1.32 has been released atop LLVM/Clang 23.1.2, adding full back-end support for AVX10.2 and APX for upcoming Intel Diamond Rapids and Nova Lake processors. It also introduces AMD Zen 6 support via --cpu=znver6, new parallel bit deposit and extract functions, and other enhancements.

Linux Kernel Developers Plan To Remove “Assisted-By” Tag Requirement For AI/LLM Usage

Linux kernel developers are planning to remove the requirement that patches disclose AI/LLM assistance with an “Assisted-by:” tag. A patch to drop the tag’s documentation was posted to the Linux kernel mailing list after maintainers reached consensus at the Linux Maintainers Summit in Prague, though opinions on the change remain mixed and it will likely proceed.

Firefox 159 Beta Released With Fix For Tabs On Wayland

Firefox 159 is now in beta ahead of its official release in two weeks. The update fixes a Wayland issue where tabs could detach into a new window when clicked, and it makes saved passwords available from the sidebar by default. It also includes various Windows and macOS fixes, including menu and panel fading issues.

Linux Mint 27 Continues Working Toward Christmas Release With Many Features

Linux Mint 27 “Adrien” is on track for a Christmas release under the project’s new yearly development cycle. It will use an Ubuntu 26.04 LTS base and include many Wayland improvements, plus Cinnamon, Xfce, and MATE editions. Founder Clement Lefebvre says the longer cycle has allowed more features and bug fixes than previous releases.

ONLYOFFICE !0.0 Released With AI Features, 1,000+ Bug Fixes

ONLYOFFICE 10.0 has been released as an open-source office suite alternative, adding AI features across every editor. The AI tools support document creation, formatting, editing, translation, PDF summaries, spreadsheet manipulation, and presentation assistance. The release also includes over 1,000 bug fixes and fixes for 35 vulnerabilities.

NVIDIA’s Nova Driver Continues Leading The Way For Linux Rust GPU Drivers

The DRM Rust changes for the Linux 7.4 merge window have been submitted for DRM-Next, with NVIDIA’s Nova driver leading in-kernel Rust GPU driver development. The update adds NovaCoreApi, a GPU information ioctl, VRAM buddy allocation, TLB flushing, MMU v2/v3 page-table types, BAR1 CPU access to mapped vRAM, error reporting, self-tests, and firmware handling improvements. It also includes fixes for the experimental Tyr Rust driver for Arm Mali hardware.

Asterinas 0.19.0 Released With X.Org Support

Asterinas 0.19.0 has been released, with the Rust-based, memory-safe Linux-compatible OS kernel now surpassing 100 contributors and 250,000 lines of code. The release introduces an initial DRM graphics subsystem enabling X.Org/OpenGL via Mesa LLVMpipe, minimal usable AArch64 support alongside x86_64/RISC-V/LoongArch64, and improvements to virtualization, secure containers, a Rust-native device model, self-hosting, and AI code review. Of the 55 contributors to this version, 31 were first-time contributors.

IREE 3.12 Introduces Rolls Out Rewritten Vulkan Driver, Experimental AMD HSA HAL

IREE 3.12 has been released with a rewritten Vulkan HAL that targets Vulkan 1.3 and adds native queue paths, push descriptors, and a descriptor-free buffer device address dispatch ABI, newly supporting Broadcom VideoCore VII, Arm Mali 5th Gen, and LLVMpipe. The release also debuts an experimental, opt-in AMD HSA HAL using HSA queues and AQL packets, along with a rewritten CPU backend execution path and new NVIDIA Ada/Blackwell and AMD CDNA5/MI455X and RDNA3 compiler targets.

FreeType 2.15 Released With Apple HVF Driver, ISO Open Font Format Edition 5

FreeType 2.15 has been released, adding support for ISO Open Font Format Edition 5—including the VARC and GLYF tables and support for more than 65,536 OpenType glyphs—plus an experimental Apple-contributed HVF driver for macOS. It also brings anti-aliased rasterizer memory improvements for large glyphs over 40 ppem and ClearType rendering improvements, while the Edition 5 support used LLMs for initial code derivation before manual tuning.

Anthropic’s Scion: Git replacement for monorepos

Anthropic built an in-house version control system, Scion, to replace Git for its monorepo after AI coding with Claude shifted the bottleneck to developer infrastructure; seven engineers developed it over five months using Claude. Scion handles a nearly 100 GB main history and roughly 25 million daily CI checkouts, is designed for always-online agents rather than Git’s distributed offline model, and helped daily merges grow from about 2,500 in May 2026 to over 30,000 by September. Anthropic fully migrated from Git and also shipped related tools for code review, code search, and IAM.

Announcing Lix 2.96 “Bingsu”

Lix 2.96 “Bingsu” has been released, focusing on security, debugger support, quality-of-life improvements, and ongoing Cap’n’Proto RPC and Rust work. It improves the REPL and debugger, adds rustyline-based UTF-8 REPL support, ships the experimental rpc-sockets protocol, and deprecates ssh:// store URIs. The release also includes breaking changes, notably stricter flake.nix validation that disallows function calls outside outputs.

Opposition to uv-build adoption

A Gentoo contributor is urging developers not to use uv-build as a Python PEP 517 build backend, citing another dozen packages losing Gentoo keywords after a key dependency switched to it. The objection focuses on uv-build’s large Rust-based dependency, frequent releases, and portability limits, which impose disproportionate source-build and platform-support burdens on Gentoo; the author clarifies the criticism is specific to uv-build, not uv or other Astral tools, and that its consistency and minor performance benefits do not outweigh the drawbacks.

Did Ente Locker copy my code without attribution?

The developer of open-source FairScan alleges that Ente Locker copied FairScan’s image-processing code without attribution, after Ente credited only FairScan’s segmentation model but claimed it wrote the surrounding image processing from scratch. After the developer documented near-direct Kotlin-to-Rust translations, Ente reportedly revised its wording, called the methods standard, and said the code was rewritten, but the developer notes this does not change Locker 1.0.9 and says Ente had not responded to a copyright inquiry by October 12.

Easylang: A Simple, Browser-Based Programming Language in the Spirit of the Old Home Computer BASIC

Easylang is a simple, browser-based programming language inspired by the old home-computer BASIC. Its interface includes editing, running, saving, tutoring, and debugging tools, and programs can be stored locally within the browser.

Wikipedia reverts by Eisentraut

PostgreSQL core developer Peter Eisentraut reflected on the unusually high number of significant features—around 8 to 10—reverted from PostgreSQL 19 after beta began, citing insufficient code quality, overly ambitious features, and architectural or edge-case defects. He attributed many issues to LLM-assisted code review raising the quality bar after code was written and senior developers being distracted by security work during the “Vulnpocalypse,” while saying the wave was largely not due to “vibe coding” and should resolve in future cycles.

Nixpkgs for the frontend impaired

jinx is a transpiler that compiles Nix to TypeScript, aiming to let other ecosystems use nixpkgs without reimplementing it. Its author used it to compile all 231,777 nixpkgs attributes, reportedly producing the same drvPaths as Nix. Users can evaluate packages like hello in a browser playground without Nix installed.

What Happened to the Desktop Metaphor?

The desktop metaphor—files, folders, windows, menus, and trash—borrowed from physical offices to make early computers understandable, providing a successful but imperfect mental model. It exposed a computer’s structure and transferred across programs, despite weaknesses like lost files, deep folders, and inconsistent interfaces. Modern software has gradually hidden much of that structure, with search, cloud sync, and mobile touchscreens reducing reliance on the traditional desktop.

Chasing a -O2 pessimization in antcc

A benchmark of the C compiler antcc found that compiling Lua with -O2 produced a 1.410s runtime, slower than both unoptimized (1.048s) and -O1 (1.077s). The issue was tied to inlining in luaV_execute, which grew to ~43KB at -O2—2.4x larger than -O1—despite the optimized IR being only 11% bigger. The author traced the amplification to late compiler passes, particularly instruction selection and register allocation.

Where Is All the Good New Software?

The articles ask why AI-assisted coding has not yet produced an obvious wave of new, high-quality software. They offer speculative explanations including improvements hidden in existing products, slow adoption and organizational bottlenecks, a shift toward private or ephemeral bespoke tools, incentives to build more rather than better software, speedups absorbed as leisure, and the possibility that AI may reduce software use.

Creusot 0.14.0 release - Contracts for Rust, formally proved

Creusot 0.14.0 has been released, updating the formal verification tool for Rust. The release relaxes its PartialOrd specification from total to partial orders and adds verification-condition generation for snapshots, enabling logic functions to be used as lemma functions. Existing proofs were reportedly unaffected by the partial-order change.

A slightly faster slimmer curl

curl 8.23.0, set to ship on October 14, includes several performance improvements after just over two months of new performance testing. The release speeds up internal base64 encoding/decoding, URL encoding/decoding, date parsing, URL parsing, printf calls, and number parsing, while also reducing the easy handle struct by almost 40% and the multi handle struct by 37%.

Microsoft celebrates end of Office 2021 LTSC with MS 365 outage

Microsoft’s end of Office 2021 LTSC coincided with a Microsoft 365 outage affecting users in Germany and the UK. The incident raised questions about how well the company’s cloud migration is working for those customers.

NetBSD may have found a new role: The AI-free FOSS OS

NetBSD may have found a new role as the AI-free FOSS OS. This has led to increasing awareness and adoption, followed by increasing donations.

Intel optimizes IPv6 networking into nothingness, say users

Users report that Intel’s Connectivity Performance Suite is causing connectivity oddities, with IPv6 networking effectively optimized into nonexistence. Customers describe strange network behavior tied to the suite.

VMware is pushing so many patches, it thinks users need new tools to manage the flood

VMware is releasing so many patches that it believes users need new tools to manage the update flood. In 2026, VMware also considers it acceptable to charge for multi-factor authentication.

KDE Linux at two

At Akademy 2026, KDE developers said KDE Linux remains in alpha but is already usable for QA testers, developers, and enthusiastic fans, with plans to refine it based on user feedback. The distribution uses Arch packages as a base but has no package manager, relying instead on source-built KDE software and Flatpak. It is not yet ready for mainstream use, and known issues include no Secure Boot, limited pre-Turing NVIDIA support, and immature QA infrastructure.

Deslop – Oxlint rules for AI-generated TypeScript

Deslop is a set of Oxlint rules for AI-generated TypeScript. It catches unsafe types, duplicate code, and low-value tests.

Mozilla Launches Firefox Enterprise

Mozilla has launched Firefox Enterprise, an on-premises browser built on its independent Gecko engine with a management console organizations can run on their own infrastructure. Aimed at regulated and security-sensitive institutions, it lets IT teams centrally manage deployment, policies, data movement, and AI use while emphasizing privacy, sovereignty, resilience, and reduced dependency on a small number of providers. Mozilla says centralized Thunderbird management is coming soon.

Opensidian

Opensidian is an open-source, non-profit, GPLv3-licensed, vault-compatible cross-platform Markdown notes app built on Rust and Tauri v2, currently under heavy development and seeking Linux distribution maintainers. It provides Linux x86_64 installs via AppImage, Flatpak, RPM, or DEB and can be built from source, but it is not affiliated with or endorsed by Obsidian or Dynalist Inc. Most Obsidian community themes work, with bundled themes including AnuPpuccin, Material Gruvbox, Minimal, and Solarized.

11 year old unsoundness bugs found in Rust borrow checker

A longstanding unsoundness in Rust’s borrow checker allows a struct’s 'static lifetime requirement to be bypassed through covariant field assignment. The bug lets safe code assign a non-'static reference to a field expected to be 'static, leading to use-after-free behavior such as stack-use-after-scope or segfaults. Reported as issue #164250, it affects current nightly and stable releases and has reportedly existed since at least Rust 1.4.

The Art of Dithering and Retro Shading for the Web

The article describes how the author built a personal website with 3D WebGL and shaders, discovering that post-processing was key to realizing his visual goals. It focuses on retro effects such as dithering, color quantization, and pixelization/CRT RGB cells, explaining their origins and shader techniques for modern React Three Fiber/Three.js projects. The piece also highlights examples of dithering in games and digital art to argue for the power of retro post-processing on the web.

10 Years of the SRE book: Announcing the second edition

Google is releasing a completely rewritten second edition of its “Site Reliability Engineering” book to mark the 10th anniversary of the original. The update reflects a decade of changes in cloud, AI, microservices, multi-cloud, serverless, and platform engineering, and refreshes SRE practices such as SLOs, error budgets, observability, and automation for modern production environments.

SolidCraft – Open-Source Parametric CAD in Rust (With MCP Support)

SolidCraft is an open-source parametric 3D CAD application written in pure Rust. It supports sketching, constraining, and solid modeling with a parametric feature timeline that rebuilds when dimensions change, and it can run on desktop, in the browser, or through AI agents via a built-in MCP server. Downloads are available for Windows, macOS, Linux, and web, with version 0.6 demonstrated by modeling a Raspberry Pi 4 case.

📱 Consumer & Media

Framework Laptop 12 Offers Improved Performance & Efficiency With Intel Core 3 Wildcat Lake

Framework has updated its Laptop 12 convertible 2-in-1 with Intel Core 3 “Wildcat Lake” processors, offering improved performance and power efficiency over the original Raptor Lake U model. The new lineup includes Core 3 304, Core 5 230, and Core 7 350 options, and Fedora Linux with KDE Plasma is now available as a pre-load. Reviewers found it a solid upgrade for those wanting more performance and efficiency, though not necessarily worthwhile for existing Raptor Lake U owners.

Xbox CEO Asha Sharma unveils Xbox Labs, small team testing emerging ideas and AI-built ports/mashups

Xbox CEO Asha Sharma announced Xbox Labs, a small internal team of game makers, engineers, and scientists tasked with rapidly testing emerging ideas, tools, and features and judging them by user feedback. Sharma cited recent game ports and mashups, many built with AI tools, as inspiration, though Xbox clarified the unit has no explicit AI mandate and will use AI only when beneficial. Xbox Labs formalizes existing R&D efforts such as porting games to Apple silicon, browser-based distribution via WebGPU, and modder support, with no new staff or immediate product announcements.

Apple unveils HomeView smart display

Apple has unveiled the $299 HomeView, a 7-inch A18-powered smart display available in tabletop and wall-mounted versions, shipping November 13. The device uses facial recognition for up to six users but requires a registered Apple Intelligence-enabled iPhone or iPad with Siri AI on the same Wi-Fi network and recently unlocked; Apple Watch cannot serve as that personal device. It supports hands-free FaceTime, Intercom across HomePod and Apple devices, voice messaging and calling, Find My, and smart home control within Apple’s ecosystem.

Sources: Rockstar plans GTA 6 PC release in Q1 2027, after PS5/Xbox Series X/S on Nov 19

Rockstar Games plans to release the PC version of Grand Theft Auto VI in the first quarter of 2027, according to four anonymous industry sources. The game is scheduled to launch on PlayStation 5 and Xbox Series X/S on November 19, with the PC release expected much sooner after console launch than GTA V’s was. Rockstar has not officially confirmed the PC date or responded to a request for comment.

Apple unveils new Apple TV 4K

Apple unveiled the fourth-generation Apple TV 4K, featuring an A19 chip, Apple Intelligence-powered Siri AI with far-field microphones for hands-free control, 128GB storage, Wi-Fi 7, Bluetooth 6, Find My in the remote, support for up to four HomePods, and enhanced smart-home hub capabilities. The $299 device is part of Apple’s broader smart-home push alongside HomeView and HomePod Mini 2, with preorders opening ahead of availability on November 13.

Apple unveils $149 HomePod mini 2: S10 chip, Sound Glow, Siri AI later, five colors, ships Oct 30

Apple announced the $149 HomePod mini 2, featuring an S10 chip, “Sound Glow” interactive lighting, and Siri AI via a future software update. It comes in five colors, supports immersive audio including four-speaker surround sound with Apple TV 4K, and ships October 30.

Apple unveils Home Collection with LG and Schneider: cameras, thermostats, doorbell, lock

Apple announced the Apple Home Collection, a lineup of smart home accessories created with LG Electronics and Schneider Electric, including indoor and outdoor cameras, thermostats, temperature sensors, a video doorbell, and a door lock. The products integrate with the Apple Home app and Siri, offer one-tap iPhone setup and end-to-end encryption, and will begin rolling out in December, with select accessories available in early 2027.

Alibaba unveils Qwen Glasses N1 at ~$450, 50MP cams, sensors, eye-tracking Pro, Qwen Clip AI earbuds

Alibaba unveiled its next-generation Qwen Glasses N1, starting at about $450, with a Pro model around $520 featuring 50MP cameras, sensors, and eye-tracking, plus new Qwen Clip AI earbuds. The Hangzhou launch marks Alibaba’s deeper push into consumer gadgets as part of its AI strategy, competing with Meta’s Ray-Ban smart glasses on features rather than just price.

SpaceXSI launches X Subscription: Grok, Cursor, X Premium; $8/$30/$100/$200; $30 has Grok Bot

SpaceXSI has launched a unified X Subscription bundling Grok, Cursor, and X Premium across monthly tiers priced at $8, $30, $100, and $200. The $8 Plus tier includes Grok Chat and Cursor coding tools plus X Premium, while the $30 Premium tier adds Grok Bot, latest Grok models, and enhanced Cursor and X features. Higher tiers offer priority model access, earliest access to products, and up to 20 times Plus usage capacity.

How to Watch Atletico Madrid vs Manchester United for Free (October 13, 2026)

Atletico Madrid vs Manchester United will meet in the UEFA Champions League league phase on Tuesday, October 13, 2026, at 9:00 PM CEST at the Riyadh Air Metropolitano in Madrid. Free streams are available via Digi Sport 2 in Romania and Ictimai TV in Azerbaijan. The article recommends using a VPN with a free trial or money-back guarantee, such as NordVPN’s 30-day policy, to access those regional streams from elsewhere.

How to Watch Viking vs Bayern Munich for Free (October 13, 2026)

Viking host Bayern Munich at Lyse Arena in Stavanger on Tuesday, October 13, 2026, with kick-off at 21:00 CEST. Free streams are available via Sport 4 or Prima Sport 5 through Orange TV Go in Romania and DAZN in Canada. Because these services are geo-restricted, the article recommends using a VPN free trial or NordVPN’s 30-day money-back guarantee to access them from abroad.

How to Watch Arsenal vs LOSC Lille for Free (October 13, 2026)

Arsenal vs LOSC Lille is scheduled for Tuesday, October 13, 2026, at 20:00 BST at the Emirates Stadium in London. The match can be streamed free in Ireland through Virgin Media Two via Virgin Media Play with a free account, while overseas viewers may use a VPN with an Irish server or a money-back trial to access the stream.

How to Watch Galatasaray vs Barcelona for Free (October 13, 2026)

Galatasaray vs Barcelona will meet at RAMS Park in Istanbul on Tuesday, October 13, 2026, at 22:00 TRT in a Champions League matchday 2 fixture. The match is available via local broadcasters including SBT in Brazil, TRT 1 in Türkiye, Qazaqstan TV in Kazakhstan, and Digi Sport 1/Prima Sport 1 via Orange TV in Romania, subject to local access terms. A VPN trial or money-back option is suggested for viewers abroad to access these region-restricted streams.

How to Watch Inter Milan vs Club Brugge for Free (October 13, 2026)

Inter Milan will host Club Brugge in a 2026/27 UEFA Champions League league-phase match on Tuesday, October 13, 2026, at 21:00 CEST at San Siro. Free local streams are available via Digi Sport 3 and Prima Sport 3 through Orange TV in Romania, and VTM 2 and RTL Play in Belgium. Viewers outside those markets can use a VPN trial, such as NordVPN with a 30-day money-back guarantee, to access the streams.

How to watch Tour of Guangxi 2026 from anywhere

The Tour of Guangxi 2026 runs October 13–18 in southern China, closing the men’s UCI WorldTour season with six stages and a final finish at Mashan Nongla Scenic Spot. US viewers can watch on The Roku Channel and Prime Video, while UK viewers have TNT Sports and some European markets have HBO Max/Eurosport. Travelers can use a VPN to securely access their existing home-country streaming subscriptions abroad.

Smutstep Blends Dubstep With BookTok Smut

DARQ DAGR created “smutstep,” a viral blend of BookTok romantasy/dark romance, dubstep, and horny voice lines that has attracted fans, critics, and growing demand as he frames it around safe, healthy niche communities and BookTok’s billion-dollar romantasy industry. Separately, a warrant reviewed by 404 Media says a Tesla’s security protocols allegedly prevented a Homeland Security Investigations agent from placing a tracking device on a suspected drug trafficker’s car.

Guitoro, a virtual guitar you strum with your keyboard

Guitoro is a free browser-based virtual guitar that uses recordings of real instruments and can be played via keyboard, mouse, or touchscreen. It offers acoustic, classical, electric guitar and bass options, shows chord fingering and strokes, supports practice slowed to 20%, and includes free chord tools and tuners with no signup or download. The creator is seeking feedback on how the strumming feels and sounds.

XY Music Notation

The author describes XY Music Notation, a base-12 system for representing notes in 12-TET as numbers 0–9, X, and Y, designed for a new isomorphic music keyboard. They used Codex to build a MuseScore-based score rendering backend with custom note heads showing each note’s pitch class. The author found it helpful for learning their instrument and shared it in case others are interested.

Thunderbird for iOS

The Thunderbird team is building a native iOS email app from scratch, applying lessons from Desktop and Android to deliver a fast, reliable, and seamless cross-device experience. Current work focuses on core foundations such as hassle-free setup, reliable background sync, and essential daily email tools, with a first version planned for community testing in the next few months. The team had also sought community feedback through a survey, which has now closed.

Apple’s new ‘Switch to Safari’ campaign details why you should ditch Chrome

Apple has launched a new “Switch to Safari” campaign, including a website and YouTube videos, to promote Safari as a more powerful and private browser than Chrome. The campaign highlights Safari’s better battery life, stronger privacy protections, and faster performance, along with features like Reader Mode, Profiles, and Apple Pay. Apple also released videos showcasing Safari’s Distraction Control and Intelligent Tracking Prevention.

A Book That Will Never Be Read

Michał Wójcik, a former neuroscientist, exposes AI-generated or plagiarized peer-reviewed papers published by major publishers like Springer Nature and Elsevier, often containing nonsensical images and fabricated references. He highlights an Elsevier Cell Press paper by Álex Escolà-Gascón that invented a clinical trial to argue for an immortal soul, plus cheaply fabricated book chapters sold in subscriptions that no one will ever read. The article underscores how AI slop and research-integrity failures are slipping past editorial safeguards.

Sprout Pi Mainboard – A modern, RP2350-powered motherboard for iPods

Sprout Pi Mainboard is a replacement PCB for iPod 3rd generation, 4th generation, and iPod mini models that keeps the original case, screen, and click wheel while adding Bluetooth, Wi-Fi, a hi-fi DAC, USB-C charging, and a microSD slot for up to 2 TB of music. It comes in two versions: Sprout Pico, powered by an RP2350 for instant local playback, and Sprout Pi, which adds a Raspberry Pi Compute Module Zero for streaming Spotify, Apple Music, podcasts, and apps. The project is open-source hardware and software, with files to be published on GitHub before its campaign launches.

Spotatui, a Spotify TUI/GUI without web API rate limits

Spotatui is a revived Spotify TUI/GUI client that avoids web API rate limits by using the official clients’ internal APIs via a librespot fork, with each user on their own session. It also supports other playback sources including YouTube, Navidrome/Subsonic, internet radio, local files, Qobuz, and Apple Music, plus Lua scripting, a web companion, and MCP AI-DJ tools. Written in Rust with ratatui and Tauri, its GUI currently trails the TUI in features, with full parity planned.

Teach My Little Sister How to Drive

The provided content is not a news article but a Steam page for “Teach My Little Sister How to Drive.” It only shows Steam install/sign-in options, a language-selection menu, and a “Report a translation problem” link, with no substantive news details.

The Dock on Mac, from OS 10 to macOS 27

A writer who contributed to The iOS App Icon Book found an old note from research for the macOS icon-history chapter and used it to explore how the Mac Dock evolved over the years. They compiled Dock images from OS 10 through macOS 27 into a single graphic, citing a 512 Pixels compilation, and posted it as an archival item with a high-resolution link.

💼 Business & Policy

Apple v. OpenAI: US judge partly grants expedited discovery, orders forensic device images, docs

A US judge partly granted Apple’s motion for expedited discovery in its trade-secret suit against OpenAI and former Apple employees, ordering forensic device images, certain interview summaries, Liu-Peng LINE conversations, documents on the ex-employees’ OpenAI roles, and early depositions. The judge cited a valid risk that critical evidence could be destroyed, but denied expedited discovery into some OpenAI hardware R&D and delayed a ruling on the broad “Apple Originated Information” term. The order comes before a hearing on OpenAI’s motion to dismiss and Apple’s request for a preliminary injunction.

Self-driving simulation software maker Applied Intuition gets $30B+ offers, up from $15B in June

Applied Intuition, a self-driving simulation software maker, has reportedly received investment offers that would value it at more than $30 billion, up from $15 billion in June. The Information reported the offers as investor attention remains intense around AI.

Decagon, $4.5B AI customer service firm, hires bank; early buyer talks incl ServiceNow

Decagon, an AI customer service startup last valued at $4.5 billion, has hired a bank and is in early acquisition talks with potential buyers including ServiceNow. If no deal materializes, the company may instead raise a new funding round.

Scuba raises $34M pre-seed/seed for context engine, a shared memory hub for humans and AI agents

Scuba, founded by former Slack executives Noah Desai Weiss and Dolapo Falola, launched a context engine designed to serve as shared memory for human teams and AI agents. The New York-based startup has raised $34 million across pre-seed and seed rounds, led by Accel and co-led by Accel and IVP, with backing from prominent angel investors. Scuba lets users capture and organize web artifacts into searchable, shareable Collections accessible through tools like ChatGPT, Claude, and Slack via MCP connectors.

Apple: HomeView unavailable in EU as Siri AI isn’t on iOS there; working with EU

Apple said its new HomeView smart home hub will not be available in the European Union because Siri AI is currently unavailable on iOS in the region. The company says it is working constructively with the EU to find a path forward but gave no timeline for EU availability. HomeView costs $299 in the U.S. and launches November 13 in eight countries.

Uber AV mobility/delivery head Sarfraz Maredia leaves Nov. 16 after 12+ years; latest AV exit

Sarfraz Maredia, Uber’s head of autonomous mobility and delivery, will leave the company on Nov. 16 after more than 12 years. He is departing for a top executive role at an undisclosed company outside the transportation industry, marking another leadership exit from Uber’s autonomous unit.

IMF lifts 2026 global GDP forecast to 3.1% as AI boom offsets Middle East war, energy costs

The IMF raised its 2026 global growth forecast to 3.1% from 3% in July, citing an AI-driven investment boom that is offsetting Middle East war disruptions and high energy costs. It kept its 2027 growth forecast at 3.4% and raised its 2027 inflation forecast to 4.1%, while warning that economic resilience is nearing its limits as energy buffers narrow.

Accenture mkt cap rebounds from $76B in June to ~$130B before investor day, seeks to calm AI fears

Accenture’s market capitalization rebounded from $76 billion in June to about $130 billion ahead of its investor day. At the event, the technology consultancy aims to reassure investors that AI will not threaten its business.

Amazon robotics: 1M+ warehouse robots; still world’s 2nd-largest corporate employer, 1.5M workers

Amazon operates more than 1 million active warehouse robots—more than any other organization—and plans to expand its robotic fleet. Despite this automation, Amazon remains the world’s second-largest corporate employer with over 1.5 million workers. Amazon says robots reduce dangerous tasks, but labor advocates and researchers warn that AI-driven warehouse automation and collaborative robots may not be entirely beneficial for human workers.

Global chip selloff deepens as Broadcom CEO likens AI bets to casino; Samsung, SK Hynix, Micron fall

A global sell-off in chip stocks deepened after Broadcom’s CEO compared AI investing to a casino, fueling investor worries about the scale of AI spending. Samsung fell as much as 8.8%, SK Hynix dropped 9.2%, and Micron declined 4%, with the declines dragging South Korea’s Kospi sharply lower.

China-based Transsion, Africa smartphone leader, raises $390M in Hong Kong share sale

Transsion, the Shenzhen-based smartphone maker that dominates Africa’s market, raised HK$3.1 billion ($390 million) in a Hong Kong share sale. It sold 86.6 million shares at HK$35.30 each, near the top end of its marketed range, with trading expected to begin Oct. 15. Its Shanghai-listed shares have fallen about 27% this year.

NTT Data in advanced talks for Anthropic Osaka data center, up to 500 MW, $6.3B

NTT Data Group is in advanced talks to build a data center in Osaka, Japan, for Anthropic, according to Nikkei sources. The facility could have up to 500 MW of capacity, cost about $6.3 billion, and partially open by 2028. It would be Anthropic’s first dedicated data center in Japan and among the country’s largest.

Lawsuit claims Meta tricked customers into “last chance” deal on smart glasses

A lawsuit accuses Meta of misleading customers with promotional emails that claimed a 25% discount on smart glasses was a “last chance” or “ends today” deal, even though the discount remained active afterward. The plaintiff alleges this created false urgency and violated Washington’s Commercial Electronic Mail Act and Consumer Protection Act.

IMF warns AI bubble threatens global economy

The IMF warned that a crash in AI share prices could threaten global financial stability, citing elevated valuations, rising debt, concentrated exposures, and circular financing in the AI sector. It said such a correction could cut investment and tighten financial conditions worldwide, with Asian markets especially exposed outside the US. The IMF listed an AI crash as one of three near-term risks, alongside rising government borrowing costs and a major cyber-attack.

UK Cabinet Office takes £7.8M hit to kill off recruitment system

The UK Cabinet Office has written off £7.8 million after scrapping its recruitment system to align with the Shared Services Strategy. It also faces an £8.9 million fine over IR35 failures.

Character.ai Files for Bankruptcy

Character.ai has filed for Chapter 11 bankruptcy and is pursuing a court-supervised sale, with Disney as lead bidder at $20 million, though the platform could be wound down if no suitable bid emerges. Bankruptcy documents show the AI companion business was unsustainable, losing roughly $8–9 million per month by August 2026 even after cost cuts, amid high compute, weak subscription/ad revenue, and legal costs. The platform remains operational with subscriptions active, and the sale process is expected to conclude by late November or early December 2026.

Huawei, once hobbled by US sanctions, passes 100M Kirin chip milestone

Huawei has shipped over 100 million devices powered by its Kirin processors since returning to the smartphone market in 2023. It unveiled details of its flagship Kirin 9050 series and is testing its next-generation Kirin 2027 chip, with preliminary results meeting targets, as it expands in China’s AI chip market.

Americans Trust in Federal Government Sinks to New Depths

Gallup’s annual Governance poll finds trust in the federal government at or near record lows across five measures, including 33% for the executive branch, 28% for the legislative branch, 33% each for handling international and domestic problems, and 46% for the judicial branch, its lowest ever. Trust fell sharply from last year, especially on international problems and in the executive branch, with independents and Democrats showing particularly low trust, and barely half of Americans trust the American people.

Heat Pumps Are Coming for Coal’s Industrial Heat Market

Researchers at Germany’s Fraunhofer ISE tested high-temperature heat pumps at 320°F and showed they can produce hot water and steam, offering a possible path to replace coal in hard-to-abate industrial sectors. The test data could help manufacturers design heat pumps and heat exchangers for high temperatures, large loads, and changing conditions. The breakthrough extends heat pump use beyond lower-temperature residential and commercial applications toward industrial decarbonization.

CFTC Staff Were Told to Retract Analysis of Suspicious Oil Trades

CFTC staff were told to retract their analysis of suspicious oil trades after the chairman’s adviser intervened to take over the examination. The scrutiny followed billions of dollars in bets that oil would fall and the S&P 500 would rise about 15 minutes before Trump announced a reversal on striking Iranian power plants on March 23, fueling speculation about an unfair advantage.

Denmark set to allow return of ‘physical force’ in school

Denmark plans to allow teachers to use physical restraint to remove disruptive students starting in August 2027, nearly 60 years after banning corporal punishment and force in schools. The proposal would permit intervention if a student harasses or insults others or significantly disrupts lessons, including holding, confining, or moving them against their will. Current law only allows force when a student poses a danger to themselves, others, or school property.

NAZA: Netanyahu orders identification and prosecution of film’s interviewees

Israeli Prime Minister Benjamin Netanyahu has ordered the attorney general and Shin Bet chief to identify soldiers and officers interviewed in the Guardian-produced documentary NAZA and prosecute them for harming Israel’s diplomatic and security interests. The award-winning film, directed by Yuval Abraham and Rachel Szor, details Israel’s mass killing of civilians in Gaza and has prompted government threats against its filmmakers.

China Overtakes U.S. For First Time as Top Spender on R&D in 2024

China overtook the United States in 2024 to become the world’s top spender on research and development for the first time. The available article content did not include additional details because it required JavaScript verification.

🌍 Society & Science

CXMT plans 4F² DRAM chip for data-center servers by end-2026, a breakthrough rivals sought for years

CXMT plans to launch a 4F² design DRAM chip for data center servers before the end of 2026, claiming its self-developed architecture increases memory density. The company says this is a breakthrough that rivals like Samsung have pursued for years, though the claim is unsubstantiated and jolted stocks amid China’s push into critical memory technology.

European datacenter boom threatens higher energy bills unless governments intervene

Moody’s forecasts that European datacenters could consume 17% of Dutch electricity and 30% of Ireland’s by 2035, threatening higher energy bills. The datacenter boom may require government intervention to manage demand and costs.

How Random Are You?

The article titled “How Random Are You?” contains no available content beyond loading placeholders. Therefore, no factual summary of its subject matter can be provided.

Scientists wonder: Is my AI stealing my ideas?

Amid claims that AI companies may have used researchers’ shared ideas, OpenAI and Anthropic have denied training their models on uploaded data from a mathematician and a biologist. As scientists increasingly use large language models for research, concerns are growing over data privacy, transparency, and whether the companies can fully account for how user content is used. Some researchers now view the models more as competitors than tools and are reconsidering or refusing to use them.

Now That Information Is Free, What’s a College Campus For?

As technology makes information abundant and learning less place-bound, India is expanding physical university campuses and attracting leading foreign institutions to serve its growing higher-education market. Rather than delivering information, these campuses are being justified as places for belonging, mentorship, wellbeing, identity, and serendipitous cross-disciplinary collaboration, with growing demand for strategic business and health-focused training.

The United Kingdom Generates Less Electricity per Person Than in the 1970s

UK electricity generation per person peaked in 2003 and has since fallen by more than a third, returning to late-1960s levels, even as global per-capita generation rose over 50% this century. The decline reflects increased electricity imports, from 1% of demand in 2010 to 9% in 2025, and lower domestic consumption driven partly by deindustrialization and improved energy efficiency.

BB(10) > Graham’s Number

A newly analyzed 10-state Turing machine, n1, halts from a blank tape, and its exact score has been proved in Coq. Its score beats the previous BB(10) champion, lies between two f_(ω+1) values, and is machine-checked to exceed Graham’s number, moving BB(10) to level ω+1 of the fast-growing hierarchy.

A Visual Intro to Complex Numbers

This article introduces complex numbers by showing how a system of equations fails to have real-number solutions, leading to a quadratic that requires the square root of -108. It then traces the issue to Gerolamo Cardano in 1545, who found that manipulating square roots of negative numbers in cubic equations could still produce real answers.

Were Software Engineers blue-collar workers all along?

The article argues that software engineers may actually be blue-collar skilled tradespeople rather than white-collar knowledge workers, since they similarly build and fix things and master their craft through practice. It suggests this misclassification helps explain the current AI-driven identity crisis among developers, who are now facing technological disruption like other trades and must adapt by using new tools.

We Infiltrated a Police Union Convention in Las Vegas

At the International Union of Police Associations’ 23rd annual convention in Las Vegas, a reporter observed a sparse gathering of the AFL-CIO’s only police union affiliate. Longtime IUPA President Sam Cabral complained about costs and rules at the fully unionized Paris casino, underscoring tensions between organized labor and police unions. The event highlighted the ongoing debate over the labor movement’s relationship with law enforcement amid disputes over racial justice and police reform.

Hittite royal archive found in Turkey

Archaeologists in central Turkey have uncovered a major royal archive of 131 cuneiform tablets and fragments at the ancient Hittite city of Samuha, modern-day Sivas. Dating to about 3,400 years ago and the reign of King Tudhaliya II, the texts record a fever epidemic, military logistics, religious practices, divination, and royal daily life. The discovery more than triples the 53 tablets found last year and is now the largest, most comprehensive Hittite-language archive found in the 21st century.

Leaving Tech

A 20-year software developer says he is leaving tech after applying for over three dozen jobs and receiving 24 rejections, citing a brutal market and the mental toll of the process. Despite his experience, books, talks, popular Ruby packages, and six years of remote work, he says employers’ AI expectations and his location make continuing in tech untenable. He now plans to explore what to do next.