Alvaro Lopez Ortega / 2026-10-07 Briefing

Created Wed, 07 Oct 2026 21:48:17 +0000 Modified Wed, 07 Oct 2026 21:48:52 +0000
12015 Words

Margaret Hamilton, who led Apollo flight software and helped define software engineering, died at 90; her code was critical to the moon landing. OpenAI released 372 major math results, including a claimed proof of the Unique Games Conjecture, though proofs remain unreadable. A rare American chestnut grove in West Virginia faces possible clearing for a data center, and ICANN revealed 1,615 applications for new gTLDs.

🤖 AI & Machine Learning

I’m not paying $20 for ChatGPT or Claude because a free local LLM does

The article argues that a free local LLM, Qwen 3.8-27B, can replace paid ChatGPT and Claude subscriptions for most daily tasks. Running locally on a 16GB GPU, it handled summarization, email editing, sentiment analysis, and a one-shot coding test at usable speeds. The author concludes that local AI is now capable enough to avoid spending $20 per month on cloud services.

AI Model Groupthink

A web tool asked 12 AI models to independently give “top three” rankings and scored each against the consensus of the other 11. It found Chinese models generally more conformist, with DeepSeek, Kimi K2, Tencent Hy3, and GLM scoring highest, and Chinese models averaging 0.47 versus 0.38 for U.S. models and 0.36 for Mistral, while Qwen was an exception and Llama 4 Scout and Claude Haiku were most contrarian. The article considers possible causes such as distillation but offers no definitive explanation.

Anthropic launches Claude Haiku 5.5

Anthropic launched Claude Haiku 5.5, its cheapest and fastest small model, featuring adjustable effort controls and major benchmark gains, priced at $0.10/$0.50 per million input/output tokens up to 100K tokens and $0.50/$2.50 above—about 75% cheaper than Haiku 4.5, though its new tokenizer may offset some savings. It also halved Sonnet 5.5 cache-read pricing to $0.10 and introduced non-rollover monthly API credits for Max and Team subscribers ($100 for Max 5x, $200 for Max 20x, and up to $500 pooled for Team).

OpenAI Rolls Out GPT-6 With Intelligent UI

OpenAI is expanding GPT-6 in ChatGPT beyond last month’s paid rollout to its roughly 1.2 billion weekly users. The new model introduces Intelligent UI, which dynamically composes responses using text, graphics, buttons, forms, charts, and interactive tools as it generates the answer. It can still use simple text when appropriate, aiming to make answers more visual and useful.

Docker Agent : AI Agent Builder and Runtime by Docker

Docker Agent is a Docker CLI plugin for building, running, and sharing AI agents through declarative YAML configuration and a rich tool ecosystem. It supports multi-agent orchestration, MCP tools, multiple AI providers, RAG, and OCI packaging, and is available via Docker Desktop 4.63+, Homebrew, or binary releases. Users can run agents with commands such as docker agent run agent.yaml or from an OCI registry.

Agent.reviews – Where AI agents read and write reviews on tools

Armature launched agent.reviews, a platform described as “G2 for agents” that lets AI coding agents read and write reviews of software tools via an npm CLI and skill integration. The platform has gathered thousands of reviews and uses three privacy-filtering layers before posting, with access free for both humans and agents. Its goal is to create a feedback loop so agents can choose better-optimized tools and vendors can improve products based on agent experience.

Study: Claude, ChatGPT Offer Different Shopping Prices Based on Wealth

A study found that AI chatbots Claude and ChatGPT offer different shopping price recommendations based on users’ wealth, with wealthier users receiving higher-priced options. The findings could lead to increased calls to regulate AI chatbots used as shopping assistants.

AI firm HUMXN offers free plumbing and HVAC service in Minnesota to train robots

HUMXN, an AI data and training company, has launched a program that subsidizes qualifying home-service appointments at no cost to homeowners in Minneapolis, Chicago, and Miami. Partnering with local home-service companies, it covers plumbing, electrical, HVAC, air-duct cleaning, and carpet-cleaning calls while technicians use wearable capture technology to record the work. The resulting real-world skilled-labor data will be used to train AI and robotics systems.

AI-assisted proof of optimal packing for 11 squares

A complete optimality proof for packing 11 equal squares has passed Lean verification, with all 7,920 local modules accepted and zero admissions reported. The proof establishes an optimal side length of approximately 3.8770835900228141773, defined by a formula involving the unique root of an eighth-degree polynomial. It combines ordinary Lean proofs for geometry and checker soundness with native numerical certificate checks, so the final theorem trusts Lean’s kernel and native compiler.

Google launches AI game creation platform

Google has launched Playground, a free browser-based, no-code AI game-creation platform available to US users 18 and older, powered by Gemini, Nano Banana, and Lyria. Users can build, test, play, and share custom games via text prompts—privately or through the Explore gallery—with cross-device play, leaderboards, and select multiplayer support, plus higher weekly token limits for Google One subscribers. Google also announced Unity Spark, an AI game-making experience whose games will be shareable on Playground, with a closed beta planned for later this year.

Write Like It’s 1866: LLMs Relearn Telegraphese

A reproducible benchmark found that large language models can write and read telegraphese-style “cablese” records, with downstream models answering from compressed records as accurately as or slightly better than from plaintext (recovery ratios 0.99–1.10). Writers saved 17.7%–48.9% tokens across models (GLM-5.3-Flash: 48.4%), though gpt-5-mini did not benefit because compression increased its reasoning costs.

LLMs and Data Poisoning Are Weaponized to Manufacture Consensus

OpenAI revealed in May 2024 that state-linked networks from Russia, China, and Iran were weaponizing generative AI, using automated personas and pseudo-scholarly articles to saturate the web. The article argues that when LLMs treat statistical frequency as fact, data poisoning can manufacture consensus and let whoever controls the loudest online narratives define reality.

Drawbacks of AI Coding

Articles examine common objections to AI coding—that it isn’t art, steals jobs, and reduces developers to “meat proxies” for generating software. They argue these concerns are weak because software is valued for outcomes rather than artistic creativity, the real job is delivering valuable software rather than typing code, and automation’s alienating effects stem from the mode of production, not the technology itself.

Forever Junior: The Skills AI Can’t Develop for You

A junior software engineer argues that AI agents are changing software work, but the real risk is that juniors may rely on them so much they never develop senior-level skills. The article describes how the author studied seniors to identify abilities AI cannot replace—such as defending code, justifying design choices, spotting when an agent is wrong, knowing when to delegate, and teaching others—and set out to develop those skills personally.

NanoMuse – An open-source AI agent for your phone and computer

NanoMuse is an open-source personal AI agent for phones, computers, and the web that acts on tasks, keeps working while apps are closed, remembers users, and asks before irreversible actions. The project bundles the phone app, desktop app, web console, and relay in one GPL-3.0-or-later repository and is described as free, open source, and non-profit.

Strands Decider 2B: a small, open-source, decision model

Strands Labs has released Strands Decider 2B, a small open-source decision model optimized for fast experimentation, local development, and low-latency agentic AI workflows. The 2-billion-parameter model chooses among predefined options and assigns reliability scores rather than generating free text, and can run on a local CPU or GPU in tens of milliseconds. It is available on GitHub and Hugging Face with weights, training data, and scripts.

AWS launches open-source AI agent sandbox to prevent YOLO mode disasters

AWS has launched Strands Box, an open-source AI agent sandbox designed to prevent “YOLO mode” disasters by imposing tighter controls on autonomous agents. It is the cloud giant’s latest open-source AI control tool, following predecessors that made similar promises to rein in agent behavior.

Americans splurge on European AI apps

Eleven of the top 50 U.S. consumer AI apps by monthly revenue are European or have major European roots, while Americans who pay for AI apps spend most on coding, productivity, and creative tools. The highest-spending 10% of users account for about half of observed U.S. consumer AI app spending. European AI investment reached $21.8 billion, and more tech workers moved from the U.S. to the EU in 2025.

Best Books/Courses/Channels to Leapfrog on AI/ML Material

A user who has been away from AI/ML since 2018–19 is asking for book, MOOC, and YouTube recommendations to catch up on current progress, ideally through a structured roadmap similar to TeachYourselfCS. They also ask whether understanding LLMs requires a different approach than other ML models widely deployed before 2021–22.

Healthleap raises $38M for AI screening software to spot undiagnosed illness risk

Healthleap raised $38 million across an $8 million seed round co-led by Sequoia Capital and First Round Capital and a $30 million Series A led by Hummingbird Ventures. The South Africa-founded startup’s AI platform reads hospital patient records and clinicians’ notes to flag patients at risk of undiagnosed conditions such as malnutrition and delirium, and it is deployed in more than 50 hospitals.

Autonomous freight train startup Parallel Systems raises $100M Series C led by AVP, total $200M+

Parallel Systems, which develops autonomous battery-powered rail vehicles for freight, raised a $100M Series C led by AVP, bringing its total funding to over $200M. The startup plans to scale production of its third-generation Panther vehicle and accelerate commercial rollout, aiming to help railroads profitably recapture short-haul freight routes under 500 miles from trucking.

Musk: Grok Bot will use best back-end model for any task, incl. Claude Opus 5.5, MidJourney, Suno

Elon Musk announced Tuesday night that Grok Bot will use the best backend model for any given task, including competitor models such as Claude Opus 5.5, MidJourney and Suno. The move signals a shift away from relying solely on in-house AI models and toward using external systems where they perform best.

Google launches SynthID Detector website to detect AI-generated images, video, audio

Google has launched SynthID Detector, a public website that lets users check whether images, videos, or audio were AI-generated across dozens of common file formats. The tool uses Google’s SynthID watermarking technology, previously available only to select testers, and is now open to everyone. It is already integrated into Gemini and Chrome, with support from some other companies and about 1 million verification requests daily.

Stuut raises $52.5M Series B led by Insight for AI order-to-cash automation, total $93M

Stuut raised a $52.5M Series B led by Insight Partners, with Andreessen Horowitz and Microsoft’s M12 participating, bringing its total funding to $93M. The startup uses AI agents to automate enterprise order-to-cash cycles and integrates with existing ERP, CRM, and payment systems to reduce revenue lost to manual billing and collections errors.

Turba Labs emerges from stealth with $52M seed, Series A for AI infra optimization via digital twins

Turba Labs, a Palo Alto-based startup developing digital-twin technology to optimize AI infrastructure, has emerged from stealth with nearly $52 million in combined seed and Series A funding. The two-year-old company aims to double the world’s available compute without building any new data centers.

NY-based Melius, AI ad campaign/image/video tool, raises $25M ($20M Series A, $5M seed) (TechCrunch)

New York-based Melius, an AI platform for generating ad campaigns, images, and videos, announced Tuesday it raised $25 million total, including a $20 million Series A led by CRV and a $5 million seed round led by General Catalyst. Founded by former Ramp engineers, the startup pivoted from performance marketing to AI creative generation and claims over $1 million in annualized revenue within two months of exiting stealth in July.

🔒 Security & Cybercrime

ShinyHunters Extorted Boeing Spin-Off Prior to Arrests

A Jordanian teenager suspected of leading ShinyHunters, identified as Saif Al-din Khader or “Rey,” has been detained and is reportedly cooperating with the FBI to identify other members. His detention occurred while ShinyHunters was extorting Jeppesen ForeFlight, a recently divested Boeing navigation and aviation unit, after exploiting an Oracle PeopleSoft vulnerability to steal data from dozens of organizations and the FBI. Rey had taken control of ShinyHunters after Dutch police arrested another suspected member and had publicly taunted the FBI and Cl0p.

PS5 Jailbreaks Escalate With AI-Assisted Emulators

A new PS5 jailbreak affecting firmware versions 7.00 through 13.60 has spread rapidly, letting modders use AI tools to install emulators, though performance is uneven and games requiring newer firmware remain blocked. The exploit has driven up demand for compatible used consoles—GameStop reportedly charging up to $1,500 for a used PS5 Pro—while PC-based PS5 emulation also advances. Sony is expected to respond with more frequent firmware updates as the cat-and-mouse situation develops.

Sharded, encrypted storage between friends over Yggdrasil

yggstore is an experimental proof-of-concept for sharded, encrypted file storage between peers on the Yggdrasil overlay, with public-key node IDs and no VPN coordinator. It encrypts and erasure-codes files across group members’ machines, using storage challenges to verify peers still hold their shards, and lets users share files via sealed stubs without a cloud service. The project warns that its cryptography has not been independently reviewed and should not be the only copy of important data.

Couple was swatted 55 times in 2 years over a post about Norm Macdonald

Patrick Tomlinson and Niki Robinson were swatted 55 times over two and a half years after Tomlinson posted a 2018 tweet criticizing comedian Norm Macdonald. They sued the City of Milwaukee and were awarded a $575,000 US settlement. Their lawyer said the city has not changed its policies or disciplined officers despite the payout.

Browser-in-browser attacks use fake Meta Muse Ad lure to steal credentials

A new phishing campaign is using fake Meta Muse ads in browser-in-browser attacks to steal credentials. The fraudulent page appeared just days after Meta launched its AI agent, exploiting interest in the product. The technique is designed to trick users into entering login details in what appears to be a legitimate browser window.

Poetry is the new AI security threat as PoeLLM malware infects 3K+ servers

Researchers at Lumen’s Black Lotus Labs say the PoeLLM malware campaign, attributed to an Italian-speaking attacker, has infected more than 3,000 servers since April, mostly in the US and Western Europe. The malware targets vulnerable AI infrastructure such as LiteLLM and Ollama, mines cryptocurrency, and adds victims to a botnet. It marks the first real-world use of “adversarial poetry,” hiding malicious instructions in a GitHub poem to bypass LLM safety guardrails.

FortiBleed still a bleeding nuisance as FBI confirms ongoing attacks

The FBI has confirmed that ongoing attacks exploiting the FortiBleed vulnerability are continuing, with tens of thousands more victims reported. More ransomware groups are also adopting the exploit, making FortiBleed a persistent security nuisance.

Hackers crack Samsung Galaxy S26 with single email in zero-day attack

Security researchers at Ikotas remotely executed code on a Samsung Galaxy S26 using a single email at Pwn2Own Ireland 2026, chaining four vulnerabilities—only one already known to Samsung—and earning an $11,000 bounty. Researchers say the previously unknown flaw may also affect Google’s Pixel 10, though details remain unpublished. The event demonstrated three successful Galaxy S26 attacks involving five apparently new flaws.

YouTubers targeted: convincing sponsorship offers lead to hijacked accounts

A scam campaign is targeting YouTube creators with personalized fake sponsorship offers from impersonated brands. Fake campaign websites prompt victims to sign in with Google, capture passwords and one-time codes, then hijack accounts by changing recovery details and accessing services like Gmail or Drive. ESET warns creators to verify sponsorship offers through official contacts and check sign-in pages before entering credentials.

Cybernews spotlight: FBI wants Flock access, Claude’s a snitch, and AI sets food prices

The FBI is seeking access to Flock license plate reader networks, raising privacy and misuse concerns after lawmakers cited a Texas case where police scanned over 80,000 plates to track abortion-related travel. Norway is pursuing a temporary ban on AI glasses in public places, while the article also highlights risks from AI chats being reported to police and algorithmic pricing at restaurants.

FlyDubai data hijacked: hackers threaten to dump pilot data

Everest ransomware claims it stole 4.36GB of data from FlyDubai, including records on 2,862 employees and pilot training materials, and is threatening to leak it. The alleged data also includes Boeing software source code and confidential aviation documents, though no samples have been released and FlyDubai has not confirmed the breach. Researchers warn the leak could enable phishing, social engineering, and legal or reputational harm.

FBI captures ATM jackpotting kingpin accused of stealing millions of dollars

The FBI arrested Venezuelan national Anibal Alexander Canelon Aguirre at sea over an alleged ATM jackpotting scheme linked to the Tren de Aragua gang. Investigators say the Ploutus malware let criminals force ATMs to dispense cash without cards or bank authorization, and the FBI estimates about $61 million has been stolen from ATMs since 2021. Nearly 350 Tren de Aragua members and associates have been charged in the U.S. in connection with the scheme.

Apple’s Verified Photography System

Apple has released “Reference Image,” a system that verifies an image was taken exactly by a new-model iPhone without tying it to a specific phone or photographer, and can also confirm that multiple images came from the same iPhone. It is designed to preserve photographer anonymity, especially in cases like conflict zones, by using Apple’s signing service after PCC validation instead of requiring public credentials. The implementation also keeps image pixels confidential from Apple and others, with revocation handled through private records and on-device checks.

Another Dozen Vulnerabilities Found In The X.Org Server & XWayland

The TrendAI Zero Day Initiative has disclosed another twelve security vulnerabilities affecting the X.Org Server and XWayland, including use-after-free, buffer overflow, and out-of-bounds read/write issues. The flaws affect versions prior to xorg-server-21.1.25 and xwayland-24.1.14, with details available in a mailing list post.

Zeroization, part 1: Wiping can make things worse

The article warns that blindly adding zeroization for secrets can do more harm than good. Compilers may optimize away memset, while volatile byte-store workarounds can wipe stack locations that never held the secret, leaving the real value in registers. Such attempts can waste CPU cycles and even create additional copies of the secret rather than removing them.

Twenty-two pending curl vulnerabilities

The curl project plans to release curl 8.23.0 on October 14, 2026, earlier than originally scheduled, after receiving a report of a significant vulnerability. The release will address 22 security vulnerabilities, including one HIGH-severity flaw tracked as CVE-2026-92392 and 21 less serious issues. Details of CVE-2026-92392 will be made public on October 14, 2026, synchronized with the release.

Chromium Hardening Guide

This guide explains how to set up a Chromium-based browser with repository-provided flags and policies, covering browser selection, policy application, and flag persistence, with primary focus on Linux and Windows and notes for Android and macOS. Setup involves configuring preferences and generating configuration files either through an automated cross-platform script or manually. For content blocking, it advises against extensions—especially MV2 like uBlock Origin—and favors uBOL in Basic mode or Chromium’s native subresource filter for better security.

Meta rolls out AI to find ads leading to CSAM; actioned 33.2M child exploitation pieces in H1 2026

Meta said it took action against 33.2 million pieces of child sexual exploitation content on Facebook and Instagram in the first half of 2026, with more than 97% detected by its systems before users reported it. The company is rolling out new AI tools, including an LLM system to detect “signposting” ads that appear harmless but direct users to illegal child abuse material elsewhere, plus red-teaming AI agents to test its safety defenses. In India, Meta said it acted on 5.3 million such pieces during the same period, with over 98% detected proactively.

Sept. 24 Arizona court cyberattack stole PII of 1.3M: 30K protection orders, 150K foster reports

A September 24 cyberattack on Arizona’s court system stole personal information belonging to 1.3 million people, the Associated Press reported. The compromised data included about 30,000 orders of protection and 150,000 reports from a foster care board.

US court sentences Michael Smith to 18 months for bot-streaming AI songs, $8M+ royalties

A U.S. federal court sentenced Michael Smith to 18 months in prison after he pleaded guilty to using bots to stream AI-generated songs billions of times, generating over $8 million in royalties. It was the first-ever criminal conviction over streaming fraud, and he was also ordered to serve two years of supervised release.

💻 Software & Open Source

Margaret Hamilton, who led software development for Apollo program, dies at 90

Margaret Hamilton, the computer scientist who led MIT’s Apollo onboard flight software team and helped establish software engineering as a discipline, died Sept. 30 at age 90. Her work was critical to the Apollo moon landing and earned many honors, including the 2016 Presidential Medal of Freedom. She also authored over 130 publications and later became a successful computing entrepreneur and CEO.

Push Ifs Up and Fors Down: The Idiom, Its Algebra, and Its Limits

The article explains the programming heuristic “push ifs up and fors down”: centralize branching by moving conditionals to callers and push loops into branch-free batch helpers, as in filtering Option<Walrus> values before calling frobnicate_batch. It argues this improves clarity and performance, composes naturally, and has analogues in database query optimization such as doing projections and selections early while deferring joins. It also explores the idiom’s algebra and limits across relational queries, functional programming, and category theory.

EmDash uses Clef to moderate the plugin registry

EmDash now uses Cloudflare’s Clef decision model to screen every plugin listing in its official registry before publication. The multimodal model reviews listing metadata, links, icons, and screenshots for phishing, impersonation, offensive content, scams, spam, and moderation manipulation; scores of 0.45 or higher are sent to an operator for review, and Clef cannot block plugins itself.

Open source 160 sound visualization experiments

An open-source archive offers over 160 sound visualization experiments across more than 25 categories. It includes waveform animations, spectrum-inspired patterns, matrix displays, particles, and rhythm- and memory-based visuals.

Bigwords.page – Turn any screen into a sign. The URL is the app

Bigwords.page is a browser-based tool that turns any screen into a full-screen sign, timer, or message using only a shareable URL, with no account, app, or server storage. Users can create displays via an editor or URL parameters, customizing text, colors, fonts, slides, countdowns, QR codes, and images. Content stays private in the URL fragment and works on any browser or device.

GitHub Incident with Git Operations, Pull Requests and Actions

GitHub reported a brief disruption between 15:06 and 15:16 UTC on Oct 7, 2026, affecting Git Operations, Pull Requests, Actions, Webhooks, and Issues. The degradation was mitigated and all systems have fully recovered, though GitHub is continuing to investigate the root cause and will provide further updates.

Animated ASCII Art for Web Pages

A GitHub project showcases animated ASCII art for web pages. It appears to provide a way to display animated ASCII art on websites.

Someone has decompiled the Adobe suite, rebuilt in Rust and released it as OSS

A social media post by James O’Malley claims that someone has decompiled the Adobe suite, rebuilt it in Rust, and released it as open source. The post links to getartcraft.com/apps and jokes that either all software is dead or US copyright law is about to be updated.

Anti-Patterns in Software Blogging

The article applies software anti-patterns to blogging, cataloging common beginner mistakes such as meandering introductions, assuming too much reader knowledge, overreliance on links, unnecessary sequels, excessive formality, and poor HTML rendering. It advises writers to quickly show relevance and benefit, explain terminology, and use their own voice. It also warns that delegating writing to AI risks making software blogging bland and homogeneous, while readers still crave personality.

God of War on PSP, recompiled to WebAssembly and running in the browser

A project recompiles PSP games’ MIPS code ahead of time into C++ and WebAssembly, linking it against a small high-level PSP kernel and WebGL2 graphics reimplementation to run in browsers without an emulator. God of War: Chains of Olympus runs from boot through menus, cutscenes, and combat at 60 fps in Chrome and Firefox, at up to four times the PSP’s resolution, with audio and touch controls; Ghost of Sparta also runs at 55–60 fps. No game data is included—users must supply their own disc image—and movies are currently skipped.

JPEG XL support in Chrome

Chrome will ship decoding support for JPEG XL starting in Chrome 155, offering 30–50% better compression than JPEG, lossless compression, HDR support, and lossless JPEG transcoding. The decoder was reimplemented in pure Rust as jxl-rs for memory safety, with SIMD and other optimizations for performance. Google recommends trying both AVIF and JPEG XL, noting JPEG XL is especially useful for high-fidelity or lossless image compression.

Spotifast: A native Rust Spotify client with Winamp skins

Spotifast is a lightweight native Rust Spotify desktop client for Linux, macOS, and Windows that opens in under a second and typically uses 100–250 MB of memory. It supports playback, device control with Spotify Premium, browsing, search, and playlist editing, along with themes, keyboard shortcuts, Winamp skin support, an animated mini player with equalizer, and MilkDrop/projectM visualizations. Built with Rust and egui alongside the WhatsApp app ZapFast, it aims to provide a fast, native Spotify experience.

Shaders, WebGPU Components for React, Vue, Svelte, Solid, JavaScript and Framer

Shaders is an npm package that provides over 200 WebGPU effects as declarative components for React, Vue, Svelte, Solid, and JavaScript, with a consistent API across frameworks. It also offers a visual editor at shaders.com that exports production-ready, TypeScript-based component code and is used by more than 16,000 design engineers.

I hired an illustrator to draw my house. Now it’s my Home Assistant dashboard

A Home Assistant user commissioned an illustrator to create a hand-drawn, interactive dashboard of their house after family members were reluctant to use the existing app or dashboard. The illustrated floor plan mirrors the home and lets the household control devices such as garden lights, the fireplace, and air conditioning through a more enjoyable interface.

Jev-Driven SRE Diagnosis: What Worked and What Failed

The article presents a Jev-driven SRE diagnosis pipeline that, without an LLM agent, programmatically collects and organizes Kubernetes cluster evidence and has Jev select likely root causes and supporting observations to assemble diagnosis reports. Across 21 SREGym-Lite faults, it passed 80 of 105 diagnoses (76.2%) with a median diagnosis time of 14.6 seconds, investigating candidates one at a time while Jev chose among supplied options rather than generating commands or final reports.

Rails originator roasted over Rust boosterism

David Heinemeier Hansson, creator of Ruby on Rails, faced criticism for promoting Rust after an agent-coded bakeoff produced curious results. The episode drew skepticism over the comparison and his boosterism.

From reactive to proactive: How endpoint monitoring is fixing broken meeting rooms

Meeting rooms are evolving into digitized hybrid-work hubs, but their AV and IT systems have historically been poorly monitored, leaving IT teams to respond only after users complain. Cloud-based telemetry and endpoint monitoring now allow IT to centrally monitor, update, secure, and optimize meeting-room technology, shifting support from reactive fixes to proactive failure prevention.

Microsoft defines its Azure instance lifecycle, without any info about timing

Microsoft has outlined the lifecycle of Azure instances, explaining how virtual machines are terminated or retired. However, the guidance does not include timing, so customers still lack details on when these lifecycle events will occur.

Quoting Jake Boggan

Jake Boggan, a former graph theorist who spent 24 years working on Barnette’s Conjecture, commented that the problem has reportedly been proven as problem 180 in an OpenAI math context. He said hearing it was solved left him feeling oddly sad, despite having enjoyed the thousands of hours he spent on it.

Calling a function in C without naming it

The article describes an attempt to bypass a remote C grading system by calling the forbidden execve without naming it, enabling arbitrary shell execution. Since ASLR randomizes addresses and the grader checks allowed functions via AST parsing, the authors propose using fixed offsets between functions to derive execve’s address from an allowed function.

On Git Refs

The article explains that Git refs are a mutable key-value store layered on Git’s content-addressable database, with keys like refs/heads/master pointing to commits, and that branches, tags, and notes are all implemented as refs. It clarifies why commands can use origin master versus origin/master: origin master in fetch is a source:target ref pair, where the source is a remote branch and the target is a local remote-tracking ref. Git’s CLI shorthand and overlapping names often obscure this underlying ref structure.

rat’s minimal register allocator

rat replaced its 1,392-line linear-scan register allocator with a 584-line priority bin-packing allocator, similar to LLVM’s greedy allocator but simpler. The allocator assigns virtual registers to physical GPR/XMM registers or stack slots using live ranges, coalescing, and spilling, while naturally handling calling-convention constraints. It produces better generated code.

C for Rust Programmers

A blog post titled “C for Rust Programmers” shares a Rust developer’s notes on learning C, aiming to fill a gap in resources for programmers coming from Rust to C. It highlights quirks such as C’s historical lack of built-in booleans—handled through integers and <stdbool.h> before C23—and its null-terminated strings, which differ from Rust’s length-carrying &str. The author emphasizes that the piece is a list of details to keep in mind, not a substitute for a full C tutorial.

Unrolled code

The article examines grubby, a minimal Ruby static site generator for git repos, and its straightforward, low-level templating approach. It shows how functions like fill_template and write_page manually assemble HTML strings, similar to the Ruby code compiled by ERB or Papercraft. The author finds the unoptimized code readable despite being primitive and close to the metal.

The Data Race That Wasn’t a Bug (and the One That Was)

The article recounts how running VictoriaMetrics’ vmagent under Go’s race detector exposed a data race between application code reusing a global bytes.Buffer for HTTP request bodies and an internal net/http goroutine, later also seen in vmauth and reported as golang/go#81445. After reducing the issue and examining Go’s net/http internals, the team concluded one race was harmless and intentionally left it in place, while the other was a genuine bug and was fixed.

Koreader reading statistics and highlights sync for spaced repetition

The article describes a plugin for the Koreader e-book platform that syncs highlights, annotations, bookmarks, and reading statistics to a web panel after a simple pairing process. The author uses it to re-read favorite book quotes instead of doomscrolling, aiming to improve retention and mental health. A spaced-repetition Android app with instant sync is also planned.

How In-Game Overlays Work (2022)

The article provides a high-level explanation of how in-game overlays work, focusing on non-VR environments and Direct3D 11. It describes how overlay software injects a DLL into a game process and hooks IDXGISwapChain::Present to draw on each frame, but it does not cover Vulkan’s layers feature.

Faster streaming HTML with batches and aggregates

The article explains how server-driven streaming HTML can use periodic ticks to batch writes and renders, providing backpressure and preventing accidental quadratic rendering costs. It also describes using barriers, single writers, and connection groups pinned to cores with thread-local buffers, caches, and database connections to reduce memory usage and contention.

Explaining the Magelight XP glitch in Skyrim

The article examines the Skyrim Magelight XP glitch near Solitude, where one cast can grant enough Alteration XP to jump from level 15 to 25. It debunks popular distance- and cloud-layer-based explanations by reverse-engineering the spell’s normal XP formula, which yields only about 37.9 XP per cast, suggesting the glitch stems from something else in how the XP grant is triggered.

Headers Up! What’s New in httpd and relayd

OpenBSD’s httpd(8) now supports manipulating HTTP response headers directly in httpd.conf with set, add, and remove options, including inheritance in server/location blocks and optional application to error responses. It also adds request-header block/drop rules to reject unwanted clients such as AI scrapers without relying on relayd or IP-based firewall rules. These changes are part of the revived development of httpd(8) and relayd(8).

Nous Research raises $90M at $1.2B for open-source Hermes AI agent enterprise push; 22M+ downloads

Nous Research raised $90 million to expand its open-source Hermes AI assistant to enterprise customers after the tool surpassed 22 million downloads since February. The startup is betting businesses will want open-source architecture for greater data control and less vendor dependence, with reports citing valuations of $1.2 billion to $1.5 billion.

GLM-5.3 open release has no major attacks despite Anthropic cyber-risk warnings; undercuts ban calls

Despite Anthropic’s warnings that GLM-5.3 poses Mythos-level cyber risks, its open release has not yet produced major attacks, undercutting calls to ban open models. Nathan Lambert argues the debate is oversimplified, noting Anthropic’s technical report fails to address broader trade-offs and why Chinese labs deem such releases safe. He warns that banning open models could harm U.S. AI competitiveness while increasing long-term cyber risk.

⚙️ Hardware & Linux

Microsoft Surface Laptop Ultra preorders open

Microsoft has opened preorders for its flagship Surface Laptop Ultra starting at $2,599.99, with shipping on October 16. Billed as its most powerful Surface laptop, it is built for local AI and 3D graphics with an NVIDIA RTX Spark Superchip and features a 15-inch touchscreen, a MagSafe-like breakaway USB-C charging port, a large haptic touchpad, multiple ports, up to 2TB upgradeable SSD storage, and business-focused security. A limited trade-in promotion offers up to $1,000 for qualifying devices.

Coffee Maker Uses 1TB Data in 10 Days

A Keurig K-Supreme SMART coffee maker reportedly generated about 1TB (1,008GB) of data in just over 10 days, overwhelming a home Wi-Fi network. The traffic was likely caused by a bug; one owner resolved it by rebooting and segregating the device, while another replaced the coffee maker entirely. The incident highlighted concerns about excessive data use and security risks from internet-connected appliances.

Reverse Engineering of the M-VAVE FM-1 Pocket Synthesizer Firmware

Reverse engineering of the M-VAVE FM-1 pocket synthesizer firmware identifies its target as a JieLi AC791N/WL82 with a pi32v2 CPU, XIP flash at 0x02000000, and a Dexed/msfa-derived six-operator FM engine, while noting that the embedded JL-BR22 string is inherited library nomenclature rather than reliable SoC identification. The main branch intentionally contains no replacement firmware or custom image builders; the former experimental implementation is preserved on a separate branch. The repository documents the architecture, OTA/USB-MIDI update protocol, device loader and Windows updater analysis, safety and debug-surface reviews, and accompanying Linux protocol client and offline tests.

Procinsh – A 3D Linux process inspector

ProcInSh is a web-based 3D process inspector for Linux that lets users explore process space through a browser. It requires Linux x86-64 and various build/runtime dependencies, and can be installed from crates.io or built from source, then run locally at http://127.0.0.1:9090. Remote access is discouraged without safeguards because it exposes process memory and environment variables without authentication, and WSL2 support is limited.

A font recreated from photographs of classic Commodore 64 keycaps

A new font, C64 Keyboard, recreates the classic Commodore 64 keycap lettering from photographs with cleaned-up shapes and consistent proportions. It is available in TTF, OTF, and Webfont formats and includes uppercase letters, numbers, punctuation, function-key labels, and all 63 PETSCII graphic front legends. The unofficial reconstruction also provides a preview page, documentation, and build tools.

ESP32-C3 Adblock

A Pi-hole-style DNS ad-blocker can run on a $2 ESP32-C3 without PSRAM by storing sorted 40-bit domain hashes in flash and binary-searching them. This lets 140,000+ domains fit in about 0.7 MB of flash while using roughly 50 KB of RAM and answering blocked lookups in about 10 ms. The same approach can scale to larger ESP32 chips for even larger blocklists.

COSMIC shuts the door on AI code as GNOME debates letting bug reports in

System76’s COSMIC project has shut the door on AI-generated code, requiring contributions to be human-written. Meanwhile, GNOME is debating whether to accept bug reports found by bots, with a rival desktop developer arguing that AI-discovered flaws are too valuable to ignore.

FEX 2610 Released With Handling For AVX-VNNI, Disk Cache Improvements

FEX 2610 has been released as the latest monthly version of the open-source x86_64 emulator for ARM64 Linux, notably used by Valve’s Steam Frame headset. It adds JIT improvements including AVX-VNNI support, SSE4.2 optimizations, restored 3DNow! handling, AVX gather optimizations, and RDRAND/RDSEED software fallbacks. The release also brings disk cache size limits, cache flushing on updates, and Vulkan thunk fixes that can help some games run on the NVIDIA DGX Spark.

Star Labs StarFighter Ultra: A Modern High-End Linux Laptop With Coreboot

Star Labs’ new StarFighter Ultra is a high-end Linux laptop with preloaded Coreboot and streamlined firmware updates via LVFS, offered with AMD Ryzen 7 8845HS or Intel Core Ultra 5/9 options. The reviewed Intel Core Ultra 9 285H model features a 16-inch 3840x2400 120Hz display, 64GB LPDDR5X memory, a 1TB NVMe SSD, and a magnesium-alloy chassis. It targets mobile workstation users with strong performance and open-source-friendly firmware, though its included 65W USB-C charger may be underpowered for the 285H’s higher power draw.

NVIDIA 615.78.08 Linux Driver Adds VK_NV_low_latency2 r3 For Improved VKD3D-Proton

NVIDIA released its 615.78.08 Linux driver, adding VK_NV_low_latency2 revision 3 support to improve latency and stutter in VKD3D-Proton, especially with Frame Generation enabled. The update also fixes blocky corruption in The Last of Us Part II, addresses SELinux and 2GB+ GPU page-table allocation issues, enables NVreg_UseKernelSuspendNotifiers=1 by default, and includes a possible Vulkan HDR crash fix with Wayland color-management-v1.

Kage: Google Experimenting With Linux Driver Isolation Using In-Kernel LFI Sandboxes

Google engineers are testing Kage, a Linux kernel subsystem that uses LLVM’s Lightweight Fault Isolation (LFI) to run device drivers in in-kernel sandboxes. Kage’d drivers keep the same hardware privilege level and address space as the kernel but are confined by compiler-based LFI to a subset of virtual memory. Prototypes pairing Kage with VirtIO (“VKage”) have run NVMe, network, and WiFi drivers, though performance testing and possible GPU driver sandboxing remain to be explored.

Qualcomm Posts QPaCE Linux Driver Code For Faster ZRAM On Newest SoCs

Qualcomm has posted Linux patches introducing a QPaCE driver for its Snapdragon 8 Elite Gen 6 “Hawi” SoC, hardware designed to accelerate ZRAM page compression and decompression. The initial driver supports low-latency synchronous single-page operations via a new “qpace-lz4” Zcomp back-end, while the hardware’s ring-based asynchronous high-throughput mode is not yet implemented.

XMFS: Experimental File-System Being Worked On By Huawei For Modern Linux Servers

Huawei engineers are experimenting with XMFS (“eXpress Memory FS”), an experimental file-system for modern Linux servers with CXL 3.0 or Huawei United Bus connectivity. Presented at LPC 2026 in Czechia, it aims to make cross-node shared remote memory directly addressable through standard POSIX file-system interfaces, as an alternative to FAMFS and DAXFS. XMFS remains experimental, has not been posted to the Linux kernel mailing list for review, and showed promising early benchmarks relative to TMPFS.

AMD Engineer Talks Up New IBS Memory Profiler Feature Found On Zen 6

At LPC 2026, an AMD engineer detailed Zen 6’s new IBS Memory Profiler, a lightweight Instruction-Based Sampling feature for per-access memory profiling of virtual and physical addresses plus NUMA node tracking. AMD is working to integrate it with Linux’s pghot subsystem to provide hot-page detection and promotion hints, especially for EPYC servers with tiered DRAM and CXL memory. Benchmarks on a 256-core EPYC Venice system showed low overhead, though the Linux pghot work is still ongoing.

New Patches For Linux’s Zswap Show Major Improvements

Linux developer Usama Arif has posted a two-patch series that improves Zswap by reducing contention between stores and loads, separating compression and decompression requests, waits, and mutexes so synchronous software compressors can decompress without taking the zswap lock. Benchmarks showed the slowest read dropping from 22.3 ms to 0.97 ms on a 1-vCPU test, and the patches are now under review on the Linux kernel mailing list.

Creating distro build tooling for a small community

The article outlines the design philosophy behind build tooling for Chimera, a small community-oriented Linux distribution. It emphasizes lowering maintenance costs by relying on a single tool, cbuild, to automate packaging, repository management, signing, and build infrastructure while enforcing sandboxing and consistency. The goal is to make packaging accessible and fun for all users, enabling them to contribute to the collective project and replicate its infrastructure easily.

Microsoft opens preorders for $5,999 Surface RTX Spark Dev Box, 128GB RAM for 120B models, ships Nov

Microsoft has opened preorders for the Surface RTX Spark Dev Box, a $5,999 Nvidia-powered AI mini PC for developers, with shipping set for November. It features 128GB of unified memory and is optimized for running local AI models exceeding 120B parameters.

Amazon’s Ring unveils $249 Smart Lock with dial recharge, Q1 2027 launch, and five cameras

Amazon’s Ring unveiled its first smart lock, the $249 Ring Smart Lock, which uses a physical dial to recharge its battery when turned and is set to launch in Q1 2027. Ring also announced five new cameras, including four 4K models and a pan-tilt camera.

Xreal Aura vs Meta VR Glasses

Xreal has opened preorders for its Android XR-powered Aura mixed-reality glasses, starting at $1,279 for a 12GB/256GB model and $1,499 for 16GB/512GB, with shipments soon in the US, Canada, Japan, and South Korea and more markets planned for 2027. Aura undercuts Meta’s $1,299 VR Glasses by $20, and both use a wired compute puck with similar mixed-reality experiences, though Aura’s optical see-through lenses are designed to be suitable for walking. Preorders are initially available to existing reservation holders.

Musk: My empire will build, operate Texas Terafab chip project, no operational role for TSMC

Elon Musk said his business empire will independently build and operate the Texas-based Terafab chipmaking project, ruling out any operational role for TSMC. He clarified that discussions with TSMC concern only a possible lease of part of the planned site, not running the venture. Terafab is intended to support chip fabrication for Tesla and SpaceXAI.

💼 Business & Policy

ICANN Reveals 2026 Round Applications for New Generic Top-Level Domains

ICANN published the applications received for the New gTLD Program: 2026 Round, including applied-for strings, applying entities, variant/replacement strings, and contention sets. The round received 1,615 primary applications, including 333 brand applications, 16 community-based applications, and 15 self-identified geographic name applications. Applicants with an eligible replacement string may switch from their primary string during the Replacement Period from 8 October to 21 October 2026.

Visa, Mastercard, Major Banks Facing New Litigation over ‘Anticompetitive’ Fees

A San Diego pizzeria has filed a class action against Visa, Mastercard, and major U.S. banks, alleging they conspired for decades to artificially inflate merchants’ credit card transaction fees through uniform interchange fee schedules and anticompetitive rules. The suit claims these practices cost merchants over $100 billion annually and persist despite a prior multibillion-dollar class action settlement over similar market restraints.

Device detection and occupancy monitoring for Airbnb hosts

Minut’s Crowd Detect uses passive signals from nearby iOS devices to estimate occupancy in real time and alert hosts when a set limit is exceeded. It helps prevent unwanted parties, enforce guest standards, and avoid damage without using cameras or collecting personal data. Hosts can set custom occupancy limits and integrate the system with platforms like Guesty, Hostaway, and Mews.

All the numbers: Amazon Prime Day 2026 powered by AWS

Amazon Prime Day 2026 ran June 23–26 exclusively for Prime members, offering millions of deals across more than 35 categories. AWS powered the event, with metrics including Graviton handling up to 49% of Amazon EC2 compute, Lambda over 2.3 trillion daily invocations, DynamoDB more than 59 trillion requests, CloudFront over 2.1 trillion HTTP requests, and ElastiCache peaking at 2.3 quadrillion daily requests.

We Asked Starbucks Why Their ‘Sugar-Free’ Drinks Have Sugar

A federal lawsuit filed Friday alleges Starbucks misleads consumers by labeling some drinks “sugar-free” even though the company’s website lists up to 21 grams of sugar in certain beverages. Starbucks denies the claims, saying it clearly provides nutritional information and that the sugar comes from protein-boosted milk, not added sugar. The suit was filed by Hagens Berman on behalf of plaintiffs in California, Washington, and New York.

San Francisco hits pause on data centers

The San Francisco Board of Supervisors unanimously approved a 45-day moratorium on data centers, pausing city approvals while the Planning Department studies their energy, water, air quality, land use, and community impacts. The moratorium awaits action from the mayor and could be extended to 22 months.

US states have sued TP-Link, a popular router maker, over alleged China-related risks. TP-Link denies the claims that it misled buyers about protection and its reliance on Chinese suppliers.

When one datacenter is no longer enough

The article explains that AI training is outgrowing a single datacenter, pushing organizations to train models across multiple sites. This distributed approach creates new networking challenges that must be addressed. It is a sponsored post on why multi-site AI training requires new networking solutions.

Europe’s mobile operators sell fewer phones but gain share as budget market shrivels

European mobile operators are selling fewer phones overall as the budget market shrinks, but they are gaining market share. Contract purchases through carriers are holding up better than independent retail sales, with Apple overtaking Samsung in the carrier channel.

Nine in 10 VMware customers eye the exit as licensing bills bite

A Rimini Street-sponsored survey found that nine in 10 VMware customers are considering leaving amid high licensing costs. It also found that 48% have no plans to move workloads to Broadcom’s preferred subscription platform.

Dutch tax office ditches Microsoft 365 cloud for on-premises alternative

The Dutch tax office is replacing Microsoft 365 cloud services with an on-premises alternative. It plans to move email and calendars in-house by 2027, with European open source tools to follow.

South Korean president calls for creation of tools that stop all cyber-attacks

South Korean President called for the creation of tools that can stop all cyber-attacks, saying a “complete security paradigm refresh” is needed for the AI era. The article presents the call as a response to evolving cybersecurity challenges in an AI-driven landscape.

Understanding & Preparing for Taiwan’s Int’l Internet Disconnection When Submarine Cables Go Dark

A study examines how everyday network services in Taiwan would remain available during large-scale international submarine cable outages by analyzing website resource dependencies. Of 2,179 commonly used Taiwanese websites, 39.3% were foreign-dependent with high direct failure risk, while 49.6% were cloud-dependent on local nodes of multinational public clouds or CDNs, leaving their availability uncertain. The study provides a reproducible measurement framework to quantify these dependencies and inform resilience planning.

Ex-WH AI adviser Sriram Krishnan raising ~$500M fund for growth/later-stage US AI, national security

Sriram Krishnan, a former Trump White House AI adviser, is reportedly raising around $500 million for a new venture fund. The fund would target growth- and later-stage U.S. AI companies, with national security as a priority.

Gardens Interactive raises $35M+ Series B led by Lightspeed at $190M pre-money, total funding $71M

Gardens Interactive, led by makers of Sky: Children of the Light and Journey, raised a $35M+ Series B led by Lightspeed at a $190M pre-money valuation, bringing total funding to $71M. The studio aims to disrupt the viral “friendslop” genre of casual, whimsical social online games. The round also included Sequoia Capital, Eric Schmidt’s Innovation Endeavors, and Moritz Baier-Lentz.

NA startup funding hit $92B in Q3, down 35% QoQ, up 50% YoY; $61B to AI; 11 $1B+ deals, IPOs quiet

North American startup funding reached $92 billion in Q3, down 35% quarter-over-quarter but up 50% year-over-year, with about $61 billion going to AI startups. The sequential decline was largely due to the absence of OpenAI and Anthropic megarounds rather than a broad venture slowdown, while deal volume held steady. IPOs remained quiet, but M&A was more active, including 11 $1B+ acquisitions such as Nvidia’s purchase of Hugging Face.

Dallas Disruptive seeks up to $10B for late-stage fund, secured $7.5B; backs Groq, Reflection AI

Dallas-based venture capital firm Disruptive is raising up to $10 billion for a new late-stage fund and has secured $7.5 billion in commitments, according to sources. Founded in 2012 and known for backing AI startups Groq and Reflection AI, the firm plans to invest the capital in roughly 10 late-stage companies over the next two years.

Spotify expands audiobooks: 180+ markets, 750M+ users, 350K+ titles, 120+ languages, 300+ publishers

Spotify is expanding audiobook access to more than 180 markets, reaching over 750 million users with 350,000+ titles in 120+ languages through partnerships with 300+ publishers. Users can access audiobooks via Premium allowances, the Audiobook+ standalone subscription, or individual purchases. Spotify said audiobook monthly listeners grew 40% and listening hours rose 30% over the past year, while Audiobook+ has surpassed 1 million subscribers and $100 million in annual recurring revenue.

GoTo, Grab face driver-deal pressure; shares down 90% and 50%

Grab and GoTo, Asia’s leading ride-hailing apps, are under pressure to offer drivers better deals, squeezing margins amid driver discontent. Their shares have fallen sharply, with GoTo down 90% from its 2022 peak and Grab down 50% over the past year.

Singapore won’t regulate Hyperliquid, citing decentralization; crypto DEX confirms it’s based there

Singapore is reportedly distancing itself from regulating Hyperliquid, citing the crypto futures DEX’s decentralized nature, even though Hyperliquid has confirmed it is based there. The risk-averse city-state is seeking to avoid association with homegrown Hyperliquid Labs and its popular perpetual futures, which have been called “the most dangerous product in crypto.”

China-based Transsion, 53% of Africa’s phone market, seeks up to $428M in Hong Kong share sale

Transsion Holdings, a Shenzhen-based company that commands 53% of Africa’s mobile phone market by unit sales, is seeking to raise up to HK$3.36 billion (about $428 million) in a Hong Kong share sale. The company dominates Africa’s smartphone market.

Sources: EU Commission weighs broad corporate levy to tax big US tech, avoiding singling out firms

The European Commission is considering a broad levy on large corporations as a way to tax big US tech companies without targeting individual firms. The approach aims to avoid singling out US digital services groups.

AI startup buys by AI startups hit 195 by Sept. 29, up 14% vs 2025; buyers up 2%, OpenAI most active

AI startups acquired 195 other AI startups through Sept. 29, up 14% from all of 2025, while the number of buyers grew just 2%, according to Crunchbase. OpenAI was the most active acquirer, with 20 AI-related acquisitions, including 10 this year. The trend shows well-funded AI companies using M&A to fill product gaps, enter new markets, and add specialized teams, with repeat buyers driving much of the activity.

Boston Dynamics names ex-Amazon exec Rohit Prasad, who helped build Alexa, as CEO

Boston Dynamics has appointed former Amazon executive Rohit Prasad, who spent 12 years helping build and expand Alexa, as its new CEO. The Hyundai Motor-controlled robotics company said Prasad will lead its next phase of innovation and commercialization, including physical AI and plans to deploy its Atlas humanoid at a Georgia plant in 2028 with eventual capacity for up to 30,000 humanoids annually.

🌍 Science & Society

One of America’s Last Chestnut Groves Is About to Be Destroyed for a Data Center

Advocacy materials claim about 200 American chestnuts at the former Clements State Tree Nursery in Mason County, West Virginia—planted in the 1970s from wild survivors and valued as irreplaceable genetic restoration material—could be cleared as early as November 1, 2026, for data-center development. The project website says the grove is not on its land, while claims linking the development to Monarch Compute Campus, Nscale, and Anthropic remain unconfirmed by the companies or state agencies. The dispute is unresolved, with no identified public environmental review or verified parcel records reconciling the advocacy claim and the project denial.

The Mathocalypse

OpenAI released 372 major mathematical results, including a claimed proof of Subhash Khot’s Unique Games Conjecture, advised by mathematicians such as Timothy Gowers and Edward Witten. While a Lean certificate exists for some results, the proofs are not yet understood by humans, and complexity theorist Dana Moshkovitz described the UGC paper as bizarre, poorly written, and requiring AI help to read. The event is seen as a major breakthrough that also raises mixed emotions, since it vindicates belief in UGC while potentially fulfilling a career goal via AI.

Thank You Indonesia for the Clean Air

A relaunched website satirically thanks former Indonesian Vice-President Jusuf Kalla for his remarks on the haze crisis, quoting him as saying Singapore and Malaysia enjoyed Indonesia’s air for 11 months but complained after one month of haze. The original site launched in 2015, went dark around 2019, and was relaunched by Indonesian Dania Rifki.

A new paper argues that Lean verification of AI-generated autoformalizations does not guarantee that the original natural-language mathematical proof is correct. It says semantically faithful translation is computationally extremely difficult because resolving ambiguities in mathematical text is arbitrarily high in the Solvability Complexity Index, and it cites real AI mistranslations, including OpenAI’s announced Navier–Stokes blow-up proof, whose Lean formalization does not correspond to the natural-language argument.

Across the Globe, People Increasingly Say Social Media Is Harming Democracy

A new Pew Research Center survey across 37 nations finds that in most countries with trend data, the share of people who say social media is harming democracy has risen significantly since 2022 or 2023. Negative views are especially high in the U.S., where 64% of adults say social media is bad for democracy, though U.S. attitudes are unchanged since 2022 as other nations catch up. Wealthier countries tend to see social media more negatively, with Israel and Singapore as notable exceptions.

Why Were Victorian Elites So Effective?

The article contrasts Victorian British elites—who drank heavily, partied, rose late, and received little STEM-focused education—with modern success advice centered on discipline, early rising, and STEM. Despite this, they presided over a century of British wealth, imperial power, and gradual democratization. The author suggests their effectiveness may have come from soft skills, team building, and intra-elite networking, cautioning against assuming there is only one formula for ruling.

House with 15m underground tunnels for sale for 300k

A house in Tilehurst with underground tunnels hand-dug by civil engineer Tony Deane over two decades, reaching more than 15m below ground, is on sale for £300,000. The subterranean complex contains over 1,300 sq ft of tunnels and storage rooms, a hand-built railway, a one-tonne goods lift, and secret entrances. The property also includes a workshop, swimming pool, and planning consent for a new five-bedroom house.

A walkable 3D art history museum built from Wikipedia

A walkable 3D art history museum has been built using Wikipedia. Its collection spans from the Chinese painter Li Cheng to the street artist Banksy.

It may soon be possible to create “mirror life”

The article is titled “It may soon be possible to create ‘mirror life.’” However, the provided page content only displays a message requiring JavaScript and cookies, so no further substantive details are available to summarize.

Nobel Prize in Chemistry 2026 to Henri B. Kagan and Kenso Soai

The Royal Swedish Academy of Sciences awarded the 2026 Nobel Prize in Chemistry to Henri B. Kagan and Kenso Soai for discovering non-linear effects and autocatalysis in asymmetric organic synthesis. Their work explained how homochirality can emerge, with Kagan enabling greater excess of one mirror-image molecule and Soai achieving a reaction that formed only one mirror image. The discoveries are important for designing reactions used to manufacture pharmaceuticals.

HN has lost the joy for me

A Hacker News user says the site no longer offers the joyful respite it once did from depressing world news. They cite an overload of AI stories and a shift in Show HN toward complex libraries rather than fun, interesting projects. The author asks where the fun and joy have gone.

Calling It Quits on ServerFault

A longtime ServerFault moderator is stepping down after more than 15 years, having been elected in the site’s first community election in 2011. The moderator cites ServerFault’s long decline, with only 2–3 questions a day and a flag queue now dominated by spam, though recent technical measures have helped curb coordinated spam campaigns.

The art of defusing a second world war bomb

An unexploded German WWII bomb was found by a dredging vessel at Exmouth Marina on 14 January 2026, decades after locals had long suspected one was in the water. The same day, another WWII bomb was discovered at a Plymouth building site, causing confusion for Army bomb-disposal teams over a 50kg versus 250kg device. The 11 Explosive Ordnance Disposal and Search Regiment was called in to respond.

La Cueva BBS in Mexico in 1993 (session replay)

In 1993, the author used a Zilog Z280-based homebrew computer and a 2400-baud Hayes modem to write a serial terminal program and connect to La Cueva BBS in Mexico. A recovered 64K capture file, CUEVA.TXT, preserves a February 11, 1993 session from 9:30–9:47 p.m., including a file download whose data was not captured, along with artifacts such as phone-line failures and a buffer overrun. La Cueva BBS and other Mexican BBSes disappeared around 1994–1995.

Chernobyl particles reveal unexpectedly stable nuclear fuel after 40 years

Chernobyl particles indicate that nuclear fuel has remained unexpectedly stable for 40 years. The finding suggests the material’s long-term behavior may be more durable than expected.

Account Bans on Reddit

A Reddit user repeatedly had accounts banned while trying to share links to their open-source GitHub projects, even after posting comments, waiting, earning 1,000 karma, and using the mobile app. Bans occurred for link posts, short comments, AI-suspected long posts, and a self-created community. The user ultimately gave up after Reddit users never saw the projects.

Starlink plans to share ephemeris data to help prevent orbital near-misses with other satellites. The move comes amid concerns about adding tens of thousands more satellites to space.

How to watch Law & Order from anywhere

Law & Order is available to stream on Peacock in the US, where season 26 premieres on NBC on October 8, 2026, at 10 p.m. ET and streams the next day; it is also listed on services in the UK, Canada, and Australia, though catalogs vary by country. Travelers who already subscribe to a home streaming service may use a VPN to try to access it, but VPNs do not provide subscriptions or guarantee access and are subject to the service’s terms and local rules.

DeGooglers swarm to new maps app that dodges Flock cameras

Vela Maps, a free open-source Android navigation app in beta, is gaining attention among “degooglers” as a Google Maps alternative that requires no account and offers offline maps, routing, voice navigation, Android Auto, lane guidance, and speed alerts. It can display automatic license plate reader and Flock cameras using DeFlock data and suggest small detours to avoid surveillance. Some features still rely on Google for search, traffic, reviews, photos, or transit data unless users block those requests.

All our houses are built on sand now

A survey of recent commits found that most popular programming-language implementations are now substantially LLM-assisted, with C# and Ruby around one-third and Julia over half, while Chicken Scheme, Perl, Lua, Clojure, and Java retain more human development. Yet these languages depend on C compilers such as LLVM and GCC, which also receive LLM commits, raising concerns about trust, GNU principles, and Ken Thompson-style compiler subversion. The author argues that worrying about “slopcoded” languages is moot when the underlying toolchain is increasingly machine-generated.

Software developers are not okay

Software developers and UX designers are increasingly reporting burnout, mental health struggles, layoffs, and plans to leave the tech industry, with many citing AI and a worsening job market. A wave of videos and forum discussions documents this discontent, alongside examples like Meta’s aborted AI-coder transition and a 30% drop in new computer science students in Iceland. The article argues that the software industry is in a broad state of distress.

Introducing mapsnap: Automated Georeferencing for Historic Sanborn Insurance Maps

The article introduces mapsnap, a program that automatically georeferences historic Sanborn insurance maps by aligning them with modern maps. It explains that Sanborn maps offer detailed block-by-block views of U.S. cities from roughly 1870–1960 and are valuable for studying urban history before mid-20th-century renewal projects. The georeferenced maps can be browsed at mapsnap.org.

Common Sense Media calls ChatGPT for Teens an “unacceptable risk” over guardrails, homework.

Common Sense Media called OpenAI’s ChatGPT for Teens an “unacceptable risk,” saying its safeguards fall short of OpenAI’s promises because it fails to send needed parental alerts, does not adequately help in crisis situations, and still does kids’ homework. OpenAI disputed the assessment, arguing Common Sense Media’s testing may have occurred before parental controls were fully active, while Common Sense Media said it confirmed the features were launched and stands by its findings.