Alvaro Lopez Ortega / 2026-10-08 Briefing

Created Thu, 08 Oct 2026 21:53:47 +0000 Modified Thu, 08 Oct 2026 21:54:27 +0000
13875 Words

CPython developers are exploring an optional Rust zlib in Python 3.16, while NumPy 2.5 ships np.searchsorted up to 25× faster and advances abi3/Limited API work for cross-version wheels. Datadog unveiled an async-aware Python profiler that preserves asyncio task context and cuts overhead by over 60%. Also: VEX shell halved cold starts via lazy imports, and Glazer cut disposable-browser startup to ~2s.

🤖 AI & Machine Learning

Can Jev save my inbox from the Democrats?

The article describes an end-to-end ML project testing whether TypeSafe’s Jev via OpenRouter and smaller local models can filter political fundraising emails from a personal inbox. The author builds a 23.4 MB quantized ONNX SetFit model using mdbr-leaf-mt and confidence thresholding, creating a fast, CPU-friendly local classifier. It also exposes the remaining failure modes on harder boundary cases.

Jev for Python engineers

Vercel’s Python AI SDK now supports Jev, a fast universal classifier that answers structured choice, scoring, and probability questions without task-specific training. Tests on distinguishing Python from English and generating Python ASTs show Jev handles narrow decisions well but struggles with general code generation.

AI Development on Windows: from PyTorch and llama.cpp to Windows ML

Microsoft is expanding local AI development on Windows with experimental llama.cpp/GGUF support in Windows ML, preview native inference APIs, and an OpenAI-compatible endpoint for running models locally. It is also adding native PyTorch and Triton support on Windows Arm64, giving developers familiar tools to train and optimize models and deploy them through Windows ML across CPUs, GPUs, and NPUs.

Jailed mayor uses AI avatar to promise constituents free school lunches

Tirana Mayor Erion Veliaj, who remains in office while in pre-trial detention on corruption and money-laundering allegations he denies, used an AI-generated avatar to promise free school lunches for all children in the city. Albania’s Constitutional Court ruled his dismissal unconstitutional, allowing him to stay mayor. The country has also appointed the chatbot Diella to oversee public tenders and procurement.

AI-ready biological data: $1.8B global commitment

Biohub, the DOE, NIH, and new partners announced a $1.8 billion expansion of an international effort to generate AI-ready biological data and build predictive models of biology. DOE will invest over $500 million, NIH will contribute datasets from more than $500 million in prior funding, and Google DeepMind, Isomorphic Labs, and Meta are adding $300 million to the Virtual Biology Initiative. The resulting open resource aims to enable digital biological experiments and accelerate understanding and treatment of human diseases.

License update: AI derivation prohibited on all my art, lore, stories, comics

A webcomic author using Creative Commons licenses for over a decade is updating his licensing terms to explicitly prohibit AI-generated derivatives of his art, lore, stories, and comics, since current CC licenses lack such an option. He says he now receives more than three daily requests from studios, publishers, and developers seeking permission to create AI-derived works, and answering rejections has become unmanageable. The new statement, which complements an existing “NO AI” contribution policy on his Pepper&Carrot repositories, bans derivatives where AI generates or substantially replaces the creative process.

Frontier AI models outperform analysts on earnings prediction

Samaya researchers found that frontier AI models, using a finance-specific prediction harness, outperform expert analysts at predicting earnings surprises. The harness provides point-in-time financial information at parity with human experts, marking an inflection point in AI’s financial prediction capabilities.

AI SRE Arena, an Open Benchmark for AI SRE Agents on Kubernetes

AI SRE Arena is a vendor-neutral open benchmark for evaluating AI SRE agents on Kubernetes by deploying a disposable fixture, injecting faults, collecting investigation records, and scoring them with a configurable judge. It supports local kind or AWS EKS clusters and a 21-scenario full suite or six-scenario smoke test. In an initial 21-incident comparison, Edge Delta native detected and investigated 18 scenarios versus Grafana native’s 12, while Claude using edx or gcx achieved root-cause analysis scores of 85.7% and 90.5%, respectively.

Anthropic bans abusive behavior toward Claude

Anthropic updated its Claude usage policy for the first time in over a year, adding bans on election interference, deceptive political and commercial campaigns, surveillance, weapons development including related software and components, and certain health and financial uses, while prohibiting “sustained and needless abusive or cruel behavior” toward Claude with chat termination as the main enforcement for extreme repeated cruelty. It also clarified surveillance and weapons restrictions, requires a qualified operator to monitor and stop AI-connected hardware that could cause injury, and says the policy targets only extreme cases, not normal frustration, pushback, dark creative themes, or model testing and research.

Whistle: Speech to Text in 16.9 MB

Whistle is a 16.9 MB on-device speech recognition model that runs on CPU without dependencies and provides multilingual transcription, word timestamps, and speech embeddings. It uses a Needle-based encoder-decoder with per-layer gated cross attention and selectable decoder depth. It outperforms Whisper base on LibriSpeech, SPGISpeech, Earnings-22, and FLEURS average but trails it on TED-LIUM, AMI, and MLS average.

Step 5 Preview, a 1M-context MoE from StepFun, shows up on OpenRouter

Step 5 Preview, StepFun’s flagship agentic model, has appeared on OpenRouter with a sparse Mixture-of-Experts architecture featuring 27B active and 600B total parameters. It is designed for multi-step work across large codebases, documents, and tools, with strong performance in software engineering and professional knowledge work, especially finance. On OpenRouter it lists $1.00/$2.70/$0.05 pricing, 3.46s latency, 98 tps, and provider failover with per-provider uptime via the Endpoints API.

I Put an AI Agent on a Nokia 110

A person used a Nokia 110 to reduce screen time and later tried to install an AI agent on it. Installing Android failed because the phone has only 48MB of RAM, so they reverse-engineered the device for several days. They eventually got an AI agent running and automated a few actions.

Sub-1-Bit LLM Compression via Latent Factorization

The LittleBit project provides official implementations for sub-1-bit LLM compression by factorizing dense weight matrices into low-rank latent factors, binarizing them, and using learned scales to restore magnitude. LittleBit (NeurIPS 2025) targets 1.0 to 0.1 bits per weight, while LittleBit-2 (ICML 2026) adds an opt-in Joint-ITQ initialization (--use_itq) to improve latent geometry alignment without inference overhead. The codebase supports models including OPT, Llama, Phi-4, Qwen, QwQ, and Gemma.

I gave Opus 5.5 one prompt and six hours to visualize Invisible Cities

The author prompted GPT-6 Astra and Claude Opus 5.5 to create a one-shot Three.js visualization of all the cities in Italo Calvino’s Invisible Cities within six hours. GPT-6 Astra completed the task but produced visually noisy “AI design slop,” while Claude Opus 5.5 finished early and delivered a mesmerizing, much stronger design result.

Mumsnet denies using AI for content after prompt appears on message board

Mumsnet has denied using AI to produce content after a detailed prompt for writing an “am I being unreasonable” post appeared on its message board. Users reacted with alarm and skepticism, and founder Justine Roberts later confirmed the prompt came from Mumsnet’s system, which sends drafts to OpenAI for thread-title suggestions but this time generated a full-reply prompt. Roberts called it an error on OpenAI’s side and said Mumsnet does not use AI to write threads or replies.

OpenAI Withdraws 3 Math Papers

OpenAI withdrew three mathematics papers after a sign error in one invalidated a key argument and construction used by two dependent papers. It also revised 14 other manuscripts with proof repairs and clarifications, updated 13 additional manuscripts to cite the revised editions, and added formalizations bringing top-line results formalized to about 42%.

Will AI kill us all

Assessments of AI’s existential risk vary widely and remain unmeasurable: a December 2024 survey of top AI researchers put the median chance of future AI causing human extinction or severe disempowerment at 10%, up from 5%, while superforecasters gave a 2.4% chance of an AI-caused catastrophe killing over 10% of people by 2100 and 50% of US adults expressed concern. Researchers now expect human-level AI by 2042, and although Anthropic and OpenAI have paused work or withheld models over safety, both continue releasing stronger models and have loosened some safety rules.

Terence Tao Responds to the OpenAI Math Drop

The provided page, titled “Terence Tao Responds to the OpenAI Math Drop,” contains only a Mastodon web application notice requesting JavaScript or native apps. It does not include any actual reporting on Terence Tao’s response or OpenAI’s math announcement.

Why isn’t the industry freaking out about DeepSeek 4.1 Flash?

A user reports that DeepSeek 4.1 Flash has performed comparably to frontier models like Opus across a month of heavy coding and research use, while costing far less—often under $1 per session and unlimited on a $10/month plan. They question why frontier labs aren’t alarmed, arguing cheaper Chinese distilled models can handle similar workloads and that DeepSeek’s KV-cache efficiency cuts long-session costs and environmental impact. The piece contends current models are good enough for many tasks, making cost and access more important than chasing the latest frontier capabilities.

AI giants promise to play nice with personal data after UK watchdog scrutiny

Amazon, Google, Microsoft, OpenAI and six other AI companies agreed to changes in how they handle personal data after scrutiny from the UK’s Information Commissioner’s Office. However, the ICO said its questions about the companies’ data practices are not yet resolved.

Embedding the human factor into AI agent adoption

Organizations adopting agentic AI must distinguish training from continuous employee adoption to create real business value. Gartner expects deployment to rise from 17% to 60% over two years, but WalkMe’s 2026 digital adoption report finds 40% of digital transformation spending underperforms due to adoption challenges, with employees losing about 7.9 hours weekly to friction. WalkMe’s Ofir Hatsor says the gap between perceived and actual value may take years to close, and some companies may never close it.

Google’s Sashiko AI Has Completed 191k Patch Reviews, Cited On Nearly 500 Kernel CVEs

Google’s Sashiko AI code review tool has completed over 191,000 Linux kernel patch reviews across 99 mailing lists in under a year, along with 19 million autonomous Git lookups. Its reports have been cited in 1,277 upstream kernel commits, 1,567 linux-next commits, and 463 kernel CVEs this year, drawing more than 7,600 replies from over 1,100 developers. Google is now developing a local/terminal review mode, a persistent bug database, and self-review capabilities for Sashiko.

AI hardware startup Natura unveils $99 Interface smart ring connecting to AI agents/apps

AI hardware startup Natura unveiled Interface, a $99 smart ring that connects to AI agents and apps including Muse, Instinct, Claude, Grokbot, and ChatGPT, while also tracking health metrics like heart rate, HRV, sleep, and activity. The ring lets users issue voice commands and control devices without reaching for a phone, with responses delivered through headphones, iPhone Live Activity, or an app. Pre-orders are expected next month, with shipping slated for December or January.

AI leaderboard maker Arena raises $200M at $3.1B, up from $1.7B in Jan, launches Alignment Index

Arena Intelligence, the startup behind the popular AI model leaderboard, raised $200 million at a $3.1 billion valuation, up from $1.7 billion in January, in a round led by Lightspeed Venture Partners and Khosla Ventures. The company also launched an Alignment Index to measure AI safety and alignment, initially ranking OpenAI’s GPT-6.1 Sol first and Anthropic’s Claude Opus 5.5 second.

Hone, ex-OpenAI/Cognition staffers’ business AI agent startup raises $60M seed at $285M valuation

Hone, founded by former OpenAI and Cognition staffers, raised a $60 million seed round led by Benchmark and Index Ventures at a $285 million valuation. The five-month-old startup is developing AI agents designed to autonomously handle long-running business tasks and own outcomes for teams. Early customers include Cognition and Modal.

Gallatin AI raises $50M Series A to digitize and streamline military logistics data, total $70M

Gallatin AI raised a $50 million Series A, bringing its total funding to $70 million, with participation from 8VC and Silent Ventures. The California startup uses AI to digitize and streamline military logistics and sustainment data, and its Navigator tool is used by U.S. military units.

Google Cloud unveils Gemini agent for multi-day enterprise workflows in Workspace, M365, Slack

Google Cloud announced a universal Gemini agent at its Gemini at Work 2026 event that can handle enterprise tasks lasting hours or days, including research, document creation, coding, and coordinating other AI agents while preserving memory and business context across devices and apps. The agent operates through Gemini Enterprise and familiar interfaces such as Google Workspace, Microsoft 365, and Slack, and supports Gemini models plus Anthropic’s Claude with more model support planned. Google has not specified exactly how customers activate it or when all capabilities will be available.

AI retail trading agent startup Catalyst raised $30M seed led by Sequoia; pilot saw 100sM vol in 2wk

Catalyst, a startup cofounded by Justin Zheng and Dylan Iskandar, is building AI agents for retail trading that turn natural-language intent into executable trading strategies with a human approval step. It raised a $30 million seed round led by Sequoia—with participation from Jump Trading, PeakXV, Lux Capital, AntiFund, Coinbase, and Premji Invest—and says a pilot generated hundreds of millions in trading volume over two weeks.

Oratomic raises $475M Series B at $5.4B valuation, total $775M

Quantum computing startup Oratomic raised $475 million in a Series B at a $5.4 billion valuation, up from $1.5 billion after its $300 million Series A in July. The round, co-led by ARCH Venture Partners, Spark Capital, Khosla Ventures, Index Ventures, General Catalyst and Bezos Expeditions, brings Oratomic’s total funding to $775 million.

Manus parent Butterfly Effect raises $500M+ led by Boyu IDG after Meta forced to unwind Manus buyout

Butterfly Effect, the parent of AI agent startup Manus, said it raised more than $500 million in its first funding round since Meta was forced to unwind its acquisition of Manus. The round was led by Boyu Capital and IDG Capital, with follow-on investment from Tencent, HSG and ZhenFund; the valuation was not disclosed. The raise suggests investors remain undeterred after Beijing blocked Meta’s $2 billion acquisition.

Shanghai AI chipmaker Biren raises ~$515M in share sale; stock down 50%+ from June peak, up ~70% YTD

Shanghai-based AI chipmaker Biren Technology is raising about $515 million (HK$4.04 billion) in a new share sale. Its Hong Kong-listed stock has fallen more than 50% from a June peak but is still up roughly 70% year-to-date.

Isomorphic Labs, Google DeepMind AI drug spinout, in early talks to raise at $40B+ valuation

Isomorphic Labs, an AI drug discovery startup spun out of Alphabet’s Google DeepMind, is in early talks to raise new funds at a valuation of at least $40 billion, according to sources. The round could value the company as high as $50 billion, though the financing has not closed and terms may change.

Nvidia considered last-minute OpenRouter bid, but OpenRouter wouldn’t wait for generous offer

In July, Nvidia reportedly considered a last-minute bid for OpenRouter, a three-year-old marketplace for AI models, after learning of a looming takeover. Nvidia told OpenRouter’s leadership it was prepared to make a generous offer, but OpenRouter didn’t want to wait, according to The Information.

OpenRouter: OpenAI, Anthropic business spending share even in Sept; Anthropic had 75% in Jan

OpenRouter data showed business spending between OpenAI and Anthropic models was roughly even in September, compared with Anthropic’s roughly 75% share in January. Anthropic had gained an early lead with cutting-edge models and strong demand for its Claude Code tool, which caused outages and computing constraints. OpenAI has since closed the gap by some measures, as both companies seek to demonstrate sustainable businesses ahead of planned IPOs.

Ex-ByteDance intern Keyu Tian’s AI world-model lab raises $30M from 5Y, IDG at $200M valuation

Keyu Tian, a former ByteDance intern, has raised nearly $30 million from investors including 5Y Capital and IDG Capital for an unnamed stealth AI lab focused on world models. The startup was valued at $200 million post-money and aims to compete in a field also pursued by AI pioneers such as Fei-Fei Li.

Mecka raises $60M Series B led by Sequoia for human motion data to train humanoid robots

Mecka AI, a 2024-founded startup that collects human motion data to train humanoid and other robots, raised a $60 million Series B led by Sequoia with participation from Nvidia, Microsoft’s M12, and others. The company pays people to record everyday tasks using body sensors and smartphones, aiming to supply robot-training data similar to what data-labeling firms provide for LLMs. TechCrunch previously reported the round was nearing a $500 million valuation.

Property Testing with Agent Swarms

Using ordinary coding agents, developers can have swarms build property-testing machinery—reference implementations, generators, and assertions—that generates and checks cases without spending tokens on each one. Applied to complex repositories like Codex, Apollo GraphQL Router, uv, Prometheus, Babel, and KittyCAD, the approach uncovered numerous correctness bugs, including 13 in Codex and more than 30 in Apollo Router. Upstream projects have merged multiple resulting fixes, and the public playbook reportedly transfers across people, repositories, and languages.

🔒 Security & Privacy

Orchestrating Disposable Browser

Glazer replaced its race-prone parallel startup with readiness signals between Python and Docker containers, allowing components to start concurrently without assuming that running means ready. This cut cold starts from 8–10 seconds to about 4.5 seconds, and caching the Tor data directory reduced startup further to a consistent 1.9–2.1 seconds.

Django security releases issued: 6.1.2, 6.0.9, and 5.2.18

The Django team released Django 6.1.2, 6.0.9, and 5.2.18 to fix four security vulnerabilities and urged all users to upgrade as soon as possible. The issues include two denial-of-service flaws—in get_supported_language_variant() due to long language codes (low severity) and in HTTP header parsing due to quadratic time complexity (moderate)—plus a request forgery risk from spatial lookup byte values and a privilege abuse flaw in model formsets. The spatial lookup fix is backward incompatible, requiring byte raster values to be wrapped in GDALRaster.

The article presents bodycam and in-car footage of a search involving Flock that it describes as unconstitutional. It also references a separate item about Claude Code and an alleged sexual use case.

Watch Footage of the Flock Search a Judge Ruled Unconstitutional

A federal judge ruled that Tulsa police’s warrantless Flock license plate reader searches in Melissa Kyle’s arrest were an unconstitutional mass surveillance violating the Fourth Amendment, leading all evidence—including 91 pounds of meth—to be thrown out. Body-camera and in-car footage obtained by 404 Media shows a deputy running Kyle’s information on Flock and Motorola’s ALPR system while driving before fully stopping her, then using a lane-violation pretext to interrogate her about her travel.

Russian hackers hide spyware behind bizarre cat feeding planner

Russian-linked hacking group UAC-0099 is targeting Ukrainian transport manufacturing and energy workers with phishing emails that install Matchboil malware. Matchboil collects system information, contacts attacker servers, and tries to download a backdoor; one version disguised itself as a cat-feeding planner stored in a “Meowcheck” folder. According to ESET, later versions improved evasion and could check for attacker payloads every two minutes.

Crypto theft campaign escalates on Firefox: 16 additional malicious extensions flagged

Security researchers identified 16 malicious Firefox extensions that posed as harmless tools but later cloned Rabby and OKX crypto wallets to steal recovery phrases and private keys. Mozilla removed the extensions as of October 5, 2026, but attackers are continuing the campaign by rotating names and versions. Anyone who entered wallet secrets while the extensions were active should treat the wallet as compromised and create a new one in a clean environment.

Hackers take over Discord security bot, spam groups, and steal from credit cards

Double Counter, a Discord verification and anti-alt service, suffered a six-hour breach that exposed data tied to about 28 million Discord accounts, including usernames, IP addresses, rough locations, email addresses, and VPN detection records. The attacker used a stolen bot token to spam invites in roughly 50 large Discord communities and fraudulently charged $7,316, though no passwords or stored card numbers were exposed.

Dutch intelligence agencies sound alarm over growing Chinese threat to edge devices

Dutch intelligence agencies AIVD and MIVD warn that Chinese hackers are increasingly targeting VPN servers, firewalls, and other edge devices to breach corporate networks. They say Chinese actors reverse-engineer Western hardware and software and use AI to identify and exploit vulnerabilities faster. The agencies issued an advisory urging layered defenses, multi-factor authentication, network segmentation, security monitoring, and forensic readiness.

Vitalik backs crypto bunker mode amid AI risks

Vitalik Buterin backed Ethereum researcher Justin Drake’s call for crypto “bunker mode,” warning that rapid AI advances in mathematics could threaten blockchain cryptography—including ECDSA and potentially quantum-resistant lattice-based systems—before quantum computers do. Drake recommended users gradually migrate funds to fresh wallets whose public keys have not been exposed, while Buterin said there is no need to rush today but the risks should be taken seriously. The warning followed recent AI breakthroughs in mathematics and reinforced Ethereum’s hash-only security direction.

Cloudflare Keeps 1.1.1.1 Out of Piracy Blocking, Escapes Penalties in France

A French court rejected Canal+’s request to impose daily penalties on Cloudflare for allegedly failing to fully block pirate sports streaming domains. The court sided with Cloudflare, noting its CDN-based blocking complied with orders that did not specify a method and kept its public DNS resolver, 1.1.1.1, out of the blocking regime. No penalties were imposed.

Let’s Encrypt to Shorten Certificate Lifetimes

Let’s Encrypt will default to 64-day TLS certificates starting February 10, 2027, with optional 45- or 6-day lifetimes and staging tests beginning October 14, 2026; existing valid certificates will not be revoked, and the last 90-day certificates are expected to expire by May 11, 2027. It will also cut authorization reuse from 30 days to 10 days, with a further reduction to seven hours planned in 2028, so subscribers should automate renewals using ACME Renewal Info or renew at about two-thirds of certificate lifetime.

High-severity Nvidia bug could crash GPU monitoring on exposed servers

Nvidia released a fix for a high-severity vulnerability tracked as CVE-2026-47483 that could crash GPU monitoring on exposed servers. The flaw affects servers with exposed GPU monitoring interfaces.

Shai-Hulud worm makes jump to AI infrastructure with Tensorlake compromise

The Shai-Hulud credential-stealing worm has compromised AI infrastructure provider Tensorlake after appearing on npm. It was detected within minutes of release, but the impact remains unknown.

Money trail backs leaked chats from extortion crew that walks into US law firms

Researchers have corroborated parts of a leaked chat dump from the Silent Ransom Group, an extortion crew that targets US law firms by walking into their offices. The money trail reportedly details cryptocurrency payouts, cash brokers, and recruitment, supporting the authenticity of the leaked communications.

Fighting GenAI with GenAI: The New Email Security Landscape

Phishing remains a major cyber threat, accounting for 26% of FBI cybercrime complaints, with losses up 274% in recent years. Generative AI is transforming phishing by automating polished, multilingual, highly targeted emails at scale, eliminating traditional warning signs like poor grammar and making attacks cheaper and easier. This new AI-enabled landscape poses serious challenges for corporate IT and MSPs defending email and cloud environments.

South Korean bank hacker exposed via AI logs

CrowdStrike Intelligence linked a late September-to-early October 2026 campaign against South Korean financial organizations to a likely Chinese-speaking, financially motivated hacker, resulting in data exfiltration. The attacker used LLMs and ARTEX, a Chinese open-source agentic pentesting tool, and exposed Claude Code session histories and configuration files revealed the methodology, including a possible hacker’s CV among AI logs. CrowdStrike assessed the attribution with moderate confidence based on Chinese-language prompts and the tooling used.

UK and Germany team up against Russian cyberattacks as Brexit rethink looms

The UK and Germany have agreed a security pact to counter Russian cyberattacks, promising information-sharing and coordinated disruption. The agreement comes amid a possible Brexit rethink, but offers little detail on the agencies involved or how it will be funded.

Cheapskates wouldn’t pay for security help, got hit by ransomware, and went bust months later

A business owner chose not to pay for professional security help, relying instead on an acquaintance. The company was later hit by ransomware and went out of business months afterward.

Ransomware fixer claimed he could decrypt files, allegedly defrauded clients instead

Federal authorities allege a ransomware fixer claimed he could decrypt victims’ files but defrauded clients instead. He allegedly charged clients more than the actual ransom, paid the ransom, and pocketed the difference.

Anthropic launches OSS Scanner: free opt-in vuln scanner for critical OSS; AI reports unreviewed

Anthropic has launched OSS Scanner, a free opt-in vulnerability scanner for critical open-source projects. Its AI-generated reports are sent without human review, and the tool was informed by Anthropic’s experience using Claude.

Anthropic launches Critical Infrastructure Defense Program for AI security with CrowdStrike, others

Anthropic has launched a Critical Infrastructure Defense Program that will provide its most powerful AI models, threat research, and on-site engineering support to help protect critical infrastructure. The initiative begins with partners including CrowdStrike and others, according to Axios.

OpenAI disrupts Russian Politology influence campaign; targets schools, media, officials, 5/6 impact

OpenAI said it disrupted a Russian influence campaign tied to the Politology network that used ChatGPT to target schools, media, and foreign officials. The operation, which tricked schools in Latin America and stoked Ukraine-Poland tensions, was the most far-reaching influence campaign OpenAI has disrupted and received a 5/6 impact score.

3 fired OpenAI researchers urge AI labs to halt AI-monitoring-impairing work; firings chilling staff

Three fired OpenAI employees sent a letter to the company’s board and safety committees urging AI labs to halt work that could degrade monitoring of AI systems, including their chain-of-thought. They also asked OpenAI to work with outside safety auditors and said their firings are “chilling those who remain at OpenAI,” according to the Wall Street Journal.

💻 Software Development

How we built an async-aware Python profiler

Datadog built an async-aware Python profiler to reconstruct relationships between asyncio tasks, which traditional flame graphs often lose. The new model preserves execution context across tasks while reducing profiler overhead by more than 60% and producing higher-fidelity profiles.

Django: serve apple-app-site-association and assetlinks.json

Django can serve the apple-app-site-association and assetlinks.json files under /.well-known/ that enable Apple Universal Links, Android App Links, and app-site associations for password autofill and passkeys. The post explains how to create the required JSON for each platform, implement Django views, add tests and caching, and verify production setups with Apple’s CDN and Google’s Digital Asset Links tools. It follows a similar pattern to serving security.txt, covering platform-specific gotchas such as Apple’s extensionless URL and Android’s SHA-256 certificate fingerprints.

Architecture Deep Dive: Startup Time & Import Optimization

VEX shell reduced its AI agent’s cold start time from roughly 4 seconds to about 2 seconds by bypassing heavy initialization for quick CLI commands and deferring expensive imports until they were needed. It used tools like tuna and -X importtime to identify bottlenecks, moved type-only imports behind TYPE_CHECKING, and switched heavy dependencies to function-scoped lazy imports.

First Steps into Memory Management in Python

The article explains CPython memory management, detailing how pymalloc allocates small objects through arenas, pools, and blocks, while larger objects use malloc. It also covers object lifetime management via reference counting and garbage collection, noting that freed memory is often retained for reuse rather than immediately returned to the operating system.

Building a Faster (Rust-Based) Python Driver for ScyllaDB

ScyllaDB is developing a new Python driver on top of its Rust driver using PyO3, moving most serialization, deserialization, request execution, and concurrency into Rust to improve performance and maintainability. Benchmarks show especially large gains under concurrent workloads, though the project still needs production hardening, broader feature coverage, and a compatibility layer for applications using the legacy Python driver API.

Carson Gross on ‘Yes, and’

Computer science professor Carson Gross argues that programming remains a viable career despite AI because its core skills—solving problems with computers and managing complexity—remain valuable. He warns that AI is dangerous for junior programmers if they merely generate code instead of writing and reading it, leaving them unable to understand or control systems. Gross also rejects the analogy that coding-to-prompting is like assembly-to-high-level languages, since compilers are largely deterministic while LLMs can introduce accidental complexity.

What ArtCraft’s Vibe-Coded Apps Say About Adobe

ArtCraft is a new AI-first, clean-room project that recreates core Adobe Creative Cloud apps—including stand-ins for Photoshop, Premiere Pro, Acrobat, and InDesign—as lightweight Rust-based alternatives. It targets frustrations with Adobe’s subscription model, high prices, platform limitations, and unresponsiveness to power users, while raising broader questions about what happens when companies ignore their users. Early impressions suggest the apps are fast and small, though their long-term viability remains unclear.

Serverless Horrors $10,811.41

A Cloudflare user received a $10,811.41 bill after an infinite loop in a Durable Object alarm performed 6 trillion reads and writes. The author paid the bill in full after getting no refund or meaningful support, and plans to move all projects off Cloudflare to self-hosted VPS.

Live streaming Web, Android, and iOS E2E test runs over WebRTC

QA Wolf built live streaming for Web, Android, and iOS end-to-end test runs using WebRTC, letting testers or AI agents watch and intervene during tests to kill hung flows or confirm flaky fixes in real time. The implementation had to overcome fleet-scale encoding costs, unify three different capture sources (Linux container, Android VM, and real iPhone), support unattended automatic recovery, and stream while simultaneously recording runs for later review.

Clojure in the Age of Language Models

The article argues that as language models become competent at writing small code, developers should focus on high-level reasoning and problem shaping, making higher-level languages like Clojure valuable. Clojure’s live REPL workflow, immutability, and declarative composition allow code to be redefined and tested in a running process without compile/restart cycles, improving iteration for both humans and AI agents. The author concludes that live programming environments are preferable to traditional compile cycles.

Rust Port of TypeScript (Tsc)

A developer used LLMs to port the TypeScript compiler, checker, and LSP to Rust, creating ts-rust (tsc-rs). Initial OpenAI-model attempts cost over $420,000 in tokens but stalled around 84% compatibility, while Claude Opus 5.5 built a working version from scratch in 10 hours and completed the port for about $24,047 over two weeks. The early release claims 100% compatibility on tested real-world projects and aims to be a drop-in replacement for most apps, though Windows and Linux arm64 support are not yet available.

Gremlin now uses AI to break distributed systems faster

Gremlin is now using AI to break distributed systems faster. The article notes that AI excels at failing more quickly than the average SRE.

Rust upstart reveals ‘reimplementation’ of Microsoft Office – with Claude on coding duty

ArtCraft has added Word, Excel, and PowerPoint lookalikes to its open-source lineup, describing the effort as a “reimplementation” of Microsoft Office with Claude handling coding duties. The tools are not yet ready for production.

Windows Search learns to take orders

Microsoft is previewing a Windows Search update that lets users change settings and carry out tasks directly from a tidier results list. The feature aims to make search more action-oriented, reducing the need to navigate elsewhere to complete common actions.

An update on Rust’s project goals

Tomáš Šedovič, a Rust Foundation program manager and co-lead of the goals team, gave an overview at Kangrejos 2026 of how the Rust project organizes, tracks, and amends goals, aimed at helping Rust for Linux developers understand the process. The update notes that Rust for Linux has inspired several current goals and includes the full list of 81 goals for 2026, categorized by size, champion, team, and required support level.

Microsoft Issues First Release Of Their LiteBox Sandboxing Library OS

Microsoft has released v0.1, the first tagged version of its open-source LiteBox sandboxing Library OS. LiteBox is a Rust-based project that uses Linux virtualization security features to sandbox Linux apps on Linux, Linux programs on Windows, and other security-focused deployments. The release is light on details, with more detailed changelogs promised for future versions.

The history of the Hetzner Cloud network stack

Hetzner Cloud’s network stack evolved from early Linux bridges and static routes in 2011, through BGP-based dynamic routing and 1:1 NAT in 2015, to direct IPv4 routing when Hetzner Cloud launched in 2018. In 2019, Hetzner introduced an Open vSwitch-based data plane and added VXLAN-based private cloud networks. The blog post is the first in a series, focusing on the architecture and history of the stack on Cloud servers and VM hosts.

The Missing Piece in Rust Error Handling

The article argues that Rust error handling still forces a trade-off between precise, boilerplate-heavy error enums and convenient, opaque error types that hide possible failures. It proposes that error types should compose as easily as functions, introducing the eros crate’s error sets so a function can specify possible errors as a tuple without declaring a new enum. This preserves explicit error information while reducing conversion boilerplate, as shown in a port-parsing example.

The Performance Cost of RwLock in Our Read-Heavy Workload

The article analyzes a read-heavy Rust workload where many concurrent readers traverse data blocks while a single writer updates mutable metrics. It compares wrapping metrics in a parking_lot::RwLock with using a Crossbeam atomic pointer to swap immutable values, highlighting the performance cost of RwLock in that scenario.

Shopify went back to native. I think the bigger shift is formal verification

Shopify and Coinbase are among the companies moving away from React Native back to native mobile apps. The article argues this reversal is driven less by performance than by coding agents making native development cheaper, and suggests the bigger industry shift may be toward formal verification.

Migrating Git repos to SHA-256

Git 2.42 introduced SHA-256 as a new Git object format, but the article notes it offers no real security benefit, is incompatible with GitHub, creates longer object IDs, and causes issues with many third-party Git tools. Basic repository migration can be done with git fast-export and git fast-import, but repositories containing submodules crash or require poorly documented --rewrite-submodules options. The author outlines steps for converting a repository and its submodule while warning that official Git documentation does not cover the process well.

jujutsu (jj) 0.46.0

Jujutsu (jj) 0.46.0 adds support for colocating non-default workspaces via Git worktrees, controlled by --colocate/--no-colocate and git.colocate, plus new workspace management and multi-remote push features. Breaking changes raise the minimum supported Git version to 2.42.0 and Rust to 1.97.1, while altering undo/redo, split, and workspace path/list behavior. The release also introduces jj file edit/delete, diff symbol headers, and template updates including Option<T> returns for list functions.

How can undefined opcodes ud0 and ud1 have parameters?

The article explains that ud0 and ud1 are not architecturally “undefined” instructions but unassigned opcodes that still pass through parts of the decoder, allowing operands to be parsed before an invalid-opcode error is raised. Using the MMX range 0F F8–0F FF as an example, unused byte sequences like 0F FB and 0F FF were decoded with mm, mm/m64 operands before failing, a shortcut that later caused complications when SSE instructions filled those slots.

A rant about APIs

The article rants about the poor quality of many business APIs encountered while automating Vori’s onboarding flow, including contract/e-signature, billing, CRM, payment, and card terminal services. The author criticizes common problems such as gated documentation, missing OpenAPI specs, inferior Postman collections, and slow or poorly designed credential issuance and rotation. It argues that these long-established services should adopt better, copyable API practices to reduce integration friction.

🐧 Linux & Open Source

Rust for CPython (Python Language Summit 2026)

CPython developers are exploring a gradual introduction of Rust, starting with an optional Rust implementation of zlib in Python 3.16 before considering broader use in areas like I/O, JSON, parsing, and memoryview. The proposal aims to improve safety and performance without rewriting CPython, while addressing concerns around platform support, dependencies, build complexity, and whether core developers are comfortable adopting Rust.

Making np.searchsorted up to 25× Faster in NumPy 2.5

The article describes optimizing np.searchsorted by batching independent binary searches with NumPy vectorized operations and reformulating the algorithm to use O(1) extra memory. The approach was ported to C++ and included in NumPy 2.5, achieving up to a 25× speedup over NumPy 2.4 in benchmarks.

Python 3.10.22, 3.11.17, 3.12.15, 3.13.16 and 3.14.8 are now available!

Python 3.10.22, 3.11.17, 3.12.15, 3.13.16, and 3.14.8 are now available, with security fixes across all five series. Python 3.10.22 is the final 3.10 release and the series is now end-of-life, while 3.11 remains security-fix-only until October 2027 and 3.12 until October 2028. Python 3.13.16 is the last full maintenance release for 3.13, 3.14.8 is a maintenance release, and the 3.10, 3.11, and 3.12 updates are source-only while 3.13.16 and 3.14.8 include binary installers.

K10s – A Clickable Kubernetes TUI (Go, Bubble Tea)

K10s is a clickable Kubernetes terminal UI written in Go with Bubble Tea, positioned as a point-and-click, searchable, themeable alternative to k9s. It adds mouse support, unified cluster/object search, context-aware AI assistance, fast lazy startup, and built-in self-updates.

Open-Source Rust Alternatives to Adobe Apps

The article presents ArtCraft as an open-source Rust alternative to Adobe apps, highlighting a browser-based creative studio with tools for image-to-location, 3D/2D compositing, 3D meshes, posing, and cutouts. It stresses artist control and ownership, offering free, no-subscription software with the full source on GitHub and no aggregator middleman.

People Holding Up the Internet

A news investigation found that critical internet and phone software, including the time zone database used by Android and iPhones, is maintained by very few people, often unpaid volunteers. Across 23 essential software projects, many relied on only one or two regular maintainers, such as Paul Eggert for the time zone file and Lasse Collin for the xz compression tool used in Linux servers and iPhones. Collin’s decision to give Jia Tan more access to xz led to a hidden Linux backdoor in 2024 that was caught before widespread distribution.

A 100x faster* alternative to homebrew

zerobrew is a new Homebrew alternative for macOS and Linux that brings uv-style architecture to Homebrew packages. It offers commands for installing, bundling, upgrading, and cleaning up packages, and can be installed via a standalone script or Homebrew. Benchmarks show 6.6x faster cold installs and 68x faster warm installs across 100 packages, with the project marketed as up to roughly 100x faster in some cases.

EXT4 Deprecates Its Journaled “data=journal” Mode

EXT4 has deprecated its data=journal mount option, which writes all file data and metadata to the journal for crash recovery but causes significant write-performance penalties and disables delayed allocation and Direct I/O. The feature was deprecated in a Git merge for Linux 7.3 and is scheduled for removal in early 2028 following the 2027 LTS kernel.

COSMIC Epoch 1.10 Released With Launcher Now Using GPU Accelerated Rendering

System76 released COSMIC Epoch 1.10, an incremental update to its Rust-based desktop for Pop!_OS. The COSMIC Launcher now defaults to GPU-accelerated WGPU rendering to address sluggish performance, and the update also adds drawing-tablet display mapping, compositor fixes, and Wi-Fi-only setup page behavior.

Ubuntu Currently Suffering From Sustained DDoS Attack

Ubuntu is experiencing a sustained distributed denial-of-service attack that has made its website, ISO downloads, and related resources inaccessible. Canonical’s IT team is working to mitigate the attack, which is being tracked on Canonical’s status page and confirmed on Ubuntu Discourse.

Ubuntu 26.10 Delivers Nice I/O Performance Wins Over Ubuntu 26.04 On AMD EPYC

Ubuntu 26.10, a non-LTS release, delivers notable I/O performance gains over Ubuntu 26.04 LTS on current 5th Gen AMD EPYC “Turin” hardware. Benchmarks on a 64-core EPYC 9655P system show Ubuntu 26.10 with Linux 7.3 and GCC 15.3 outperforming Ubuntu 26.04 LTS with Linux 7.0 and GCC 15.2.

Secretive Rosaic Labs Hires Legendary Linux x86 Developer

Hans Peter Anvin, a longtime Linux x86/x86_64 maintainer and SYSLINUX creator, has joined the secretive semiconductor startup Rosaic Labs after reportedly being laid off from Intel this summer. His affiliation was confirmed on the Linux kernel mailing list through a signed-off patch acknowledgment. The hire adds to interest in Rosaic Labs, which has been linked to Intel Atom RTL access and mysterious x86 additions.

Nova Driver Continues Progressing With Long-Term Goal For Official NVIDIA Linux Use

NVIDIA and Red Hat engineers presented at LPC2026 on progress of Nova, a Rust-based open-source NVIDIA Linux kernel driver. Nova is planned as a long-term replacement for NVIDIA’s out-of-tree Open RM kernel modules and eventually the Nouveau driver for Turing and newer GPUs, though it is not yet ready for end users. It currently boots the GPU System Processor firmware on Turing through Blackwell with basic functionality, with future work targeting virtualization, Vulkan Compute, CUDA, and display support.

Patches Updated To Begin Removing The Linux x32 ABI

Linux kernel developers are moving to retire the x32 ABI, with new patches proposing to first mark it as BROKEN in Kconfig before eventual removal. The x32 ABI, which used 32-bit pointers on x86_64, saw little adoption and has already been disabled by default in distributions such as Debian and Fedora. The effort may proceed around the Linux 2026 LTS kernel version.

SpacemiT RISC-V SoCs To See More Improvements With Linux 7.4

Device Tree updates for SpacemiT RISC-V SoCs in Linux 7.4 will add CPU scaling support for more hardware, enable the RFKILL node, and add thermal and fan control for the CTF2301 sensor on the K3 CoM260. The Muse Pi Pro, Orange Pi R2S, and Orange Pi RV2 boards gain CPU scaling, while K1 SoC support adds onboard EEPROM and MAC address configuration from EEPROM data.

Google Working To Leverage More Compiler Optimizations Within The Linux Kernel

Google engineers are pushing to upstream Profile Guided Optimizations (PGO) and extend AutoFDO and Propeller compiler optimizations to Linux kernel modules, after Linus Torvalds previously rejected PGO for the mainline kernel. At LPC 2026, they proposed a unified Kbuild interface to ease profile data generation for kernel modules, aiming to improve performance for Google servers and Android devices.

Steam Beta Now Defaults Non-Steam Windows Binaries To Using Proton

Valve has released a new Steam client beta that defaults Windows binaries registered as non-Steam games to run through Proton. The update also includes several Linux fixes, such as resolving launch failures when changing a non-Steam game’s target binary and fixing the Stop button when a game is launched by an external tool.

German Ecosia, gov-used, drops Mistral for open-source models, incl. Chinese, citing poor quality

Ecosia is dropping French AI partner Mistral in favor of open-source models, including Chinese ones. CEO Christian Kroll cited disappointment with Mistral’s quality, saying its models are now a year behind the competition.

Extending Guix

Guix allows users to extend it with new guix commands, with extensions available as individual packages or a meta-package collection. It locates extensions through GUIX_EXTENSIONS_PATH, now using a schema-versioned module layout that supports compiled Guile modules for better performance, and the article demonstrates writing a basic hello extension.

Ending the Casuarina Linux Experiment

Wesley Moore announced he is winding down the Casuarina Linux project. The system will continue as-is until the end of October 2026, after which package updates may stop depending on his migration to another distro; infrastructure will remain up through 2026, though some parts may be retired, while the package repo and website currently have no retirement plans. Moore cited unresolved technical issues such as C++ standard library incompatibility and a lack of contributors or motivation.

specialArgs considered harmful

The article argues that using specialArgs (or extraSpecialArgs) to inject inputs, helper functions, and other data into Nix/NixOS module evaluation makes modules non-portable and hard to share, illustrating how a module can fail when those arguments are not provided. It proposes curried modules—applying special arguments before lib.evalModules consumes the module implementation—as an alternative approach.

Gentoo infrastructure sponsors wanted

Gentoo Linux is seeking new infrastructure sponsors after several long-running sponsors were unable to continue their support this year. Interested organizations can help by providing rack space, data center servers, or server hardware, with details available on the Infrastructure project wiki. Gentoo prefers partners aligned with its goals and can be contacted at infra@gentoo.org.

Beyond the &

At RustConf 2026, Rust language team lead Tyler Mandry presented “Beyond the &”, a project to make user-defined smart pointers as flexible as built-in references. The effort, worked on throughout 2026, aims to simplify the mental model for how pointers and references work. An example showed that adding a mutex can cause the borrow checker to reject code because it can no longer tell that accesses to separate fields are disjoint.

📈 Business & Policy

When No ID Means No Internet: Age Verification and the Right to Access Information

Age verification mandates for online services are spreading globally, but they often require government-issued ID such as a passport or driver’s license. Because millions of adults lack such documents—including roughly 15 million U.S. adults without a driver’s license and 2.6 million without any government photo ID—these laws risk excluding people from the internet and information. The article argues that beyond privacy harms, age verification threatens equal access to information and freedom of expression.

It’s DAF Day! Wait…What’s a DAF?

The article marks DAF Day by explaining donor-advised funds as tax-advantaged giving accounts and highlighting their benefits. It encourages people with DAFs to donate to EFF to support its work on privacy, encryption, AI, online speech, and digital security. EFF says such grants help defend civil liberties and ensure technology remains a force for freedom, innovation, and human rights.

‘It Is Done. Hail Satan:’ City Immediately Regrets New Religious Liberty Law Allowing Council Prayer

An Ottumwa, Iowa, city council meeting drew controversy after resident and former council member Matt Dalbey delivered a Satanic Temple invocation ending with “Hail Satan.” A local pastor then sought to reverse the “curse,” sparking debate over whether the council should respond; the mayor called a moment of silence while a councilor warned that responding to invocations could create taxpayer liability. Dalbey described himself as a sincere member of the nontheistic Satanic Temple.

Microsoft fires back at White House claims it replaced US workers with foreign hires, green card ban

US officials accused Microsoft and seven other tech/IT companies of replacing laid-off American workers with foreign H-1B hires and suspended them from the Permanent Labor Certification green card program, denying pending applications. Microsoft disputed the allegations, saying it files H-1B petitions only for qualified workers, pays foreign employees comparably to Americans, and will provide more information to the administration. Officials cited Microsoft laying off 6,000 U.S. workers while receiving over 6,000 H-1B visas and filing thousands of PERM applications.

VPN Early Black Friday Deals 2026

Early Black Friday 2026 VPN deals are live from ExpressVPN, Surfshark, and Proton VPN, offering up to 80%, 85%, and 70% off, respectively. ExpressVPN’s 2-year Basic, Advanced, and Pro plans are $2.99, $4.49, and $7.49/month with 4 extra months, while Surfshark’s 2-year Starter, One, and One+ plans are $2.49, $2.79, and $4.49/month with 3 extra months and a 30-day money-back guarantee. Proton VPN’s 2-year plan is $2.99/month (billed $71.76) and its 1-year plan is $3.99/month (billed $47.88), both with a 30-day guarantee and renewal at $83.88 every 12 months; ExpressVPN also offers 1-year discounts up to 66% with 3 extra months.

Mozilla criticizes Microsoft over Edge browser push

One hundred days after Mozilla published its Over The Edge 2.0 report, Mozilla says Microsoft has still not changed practices that steer Windows users toward Edge and away from their chosen browsers via Windows, Bing, and Copilot. Mozilla alleges Microsoft respects browser choice only where legally required, such as in the EEA under the DMA, while users elsewhere face prompts and defaults favoring Edge and even EEA users encounter remaining issues like Edge taskbar pinning and Copilot side-panel links. It plans to share its findings with regulators and policymakers outside the EU, citing expert reviews of Microsoft’s tactics in Germany, India, the UK, and the US.

Theranos.world

The item titled “Theranos.world” presents an interactive desktop-like interface featuring Elizabeth Holmes and instructions to log in without a password. It offers virtual actions such as sitting at a chair, viewing a device, resetting the view, sleeping, restarting, or shutting down. It contains no substantive news reporting or additional context.

The Deeply Impersonal Personalized Recruiter Mail

The article criticizes a new type of recruiter email that uses LLM-generated text to falsely personalize messages by summarizing a candidate’s career while pairing it with unrelated job descriptions. The author argues this is worse than traditional templates because it is obviously AI-written, insincere, and insults the recipient’s intelligence by claiming a fit that the job description does not support.

OpenAI Revenue Miss Hits AI Stocks

AI-related stocks fell Thursday after CNBC and the Financial Times reported that OpenAI told investors its annualized revenue was about $50 billion at the end of September, roughly $20 billion below the $68 billion–$70 billion widely reported. The higher figures reportedly included gross revenue from OpenAI’s partners. Nvidia, Oracle, and CoreWeave shares fell, with intraday declines of 3%, 6%, and 8%, respectively.

4-hour battery storage is cheaper to install than gas turbines all across globe

Wood Mackenzie’s latest global LCOE report finds four-hour battery storage is now less expensive than open-cycle gas turbines in all 43 markets modeled. In the Middle East and Africa, four-hour storage costs are forecast to fall 33% to $80/MWh by 2035, displacing gas peaking, while China remains the global storage cost benchmark. The report also says single-axis tracker solar is the lowest-cost new-build technology in 43 of 48 markets, with Saudi Arabia and UAE solar LCOE expected below $20/MWh by 2033.

New gTLD Application for .lan

A publication regarding a new gTLD application for .lan was last updated on 7 October 2026 at 15:32 UTC. The page contains no further substantive information, only a notice that JavaScript must be enabled to access the Application Publication System.

US Suspends Visa Program for Tech Firms Including Microsoft

The Trump administration said it will suspend an immigration program for multiple tech companies, including Microsoft, citing widespread abuse of the worker visa system. Vice President JD Vance said no U.S. company has abused the system more than Microsoft.

‘Neither war nor peace’: Palestinians in Gaza on existing amid disease despair

On the anniversary of the 7 October 2023 Hamas attack and the ensuing Israeli offensive, Palestinians in Gaza describe living in what aid workers call “neither war nor peace,” as a year-old ceasefire has not stopped Israeli airstrikes, shootings near the “yellow line,” or a further 1,400 deaths over the past year. Displaced families in tents and makeshift shelters face water and food shortages, disease, minimal healthcare, unemployment, and economic collapse, while keepsakes like chairs and a schoolbag remind them of lost homes. Malnutrition and daily hardship persist despite increased supplies.

It’s Probably Time to Start Worrying About the 4G LTE Shutdown Bricking Cars

The article warns that carriers may begin decommissioning 4G LTE networks as early as 2030, potentially disabling connected features in many of the roughly 500 million connected cars now on the road, a number expected to more than double by 2035. Modern cars rely on cellular connectivity for diagnostics, tracking, remote services, and infotainment, and 4G hardware cannot connect to 5G-only networks. The shutdown could repeat the 2022 3G sunset on a much larger scale.

How Technology Empowers—and Imperils—Dictators

The article examines how AI both serves and threatens authoritarian regimes, citing Russia’s use of AI-generated RT anchors as an example of reducing reliance on potentially disloyal human officials. It argues that AI appeals to dictators because it can monitor subordinates and automate bureaucratic roles, but the vision of a self-running autocracy is illusory since AI is never truly autonomous.

TSMC-GlobalFoundries $2B silicon interposer deal

TSMC and GlobalFoundries announced a five-year, $2 billion deal for GlobalFoundries to supply TSMC with U.S.-based silicon interposer capacity from its Malta, New York, facility, with an option to expand as demand grows. The advanced packaging components support domestic manufacturing of high-performance semiconductors used in AI datacenters. GlobalFoundries shares rose as much as 7% after the announcement.

Too many quantum startups, too little money to keep them alive

Gartner predicts that more than half of quantum computing startups will fold by 2030 due to insufficient funding. The forecast comes even as businesses begin testing the technology’s usefulness.

IBM offloads ‘hundreds’ of its cloudy VMware customers

IBM is offloading hundreds of its cloud VMware customers to Broadcom partner 11:11. The move comes ahead of a channel deadline tied to Broadcom’s VMware partner changes.

ICANN reveals new top-level domain bids

ICANN received 1,615 applications for new top-level domains from 481 applicants, with AI-related strings drawing intense interest. Ten applicants including Meta and OpenAI sought .agent, seven including OpenAI sought .agi, six sought .asi, and eight competed for .api; OpenAI applied for 15 TLDs and Meta for 21.

Singapore’s central bank wants all FinTech AI use cases subject to independent review

Singapore’s central bank wants all FinTech AI use cases to undergo independent review. It says failures of third-party services are no excuse for the financial sector, making firms accountable for such risks.

SpaceX acquired a low-band spectrum portfolio to enable Starlink to offer mobile phone service and become a “major mobile carrier” in the US. The announcement sent shares of incumbent carriers AT&T, T-Mobile, and Verizon down more than 6% after hours, with declines reaching 8%. Musk called the spectrum the last critical piece needed for complete phone coverage in America.

USV raises $900M, $500M early-stage, up from $275M, cuts partners to 4, aims to lead more AI rounds

Union Square Ventures has raised $900 million, including a $500 million early-stage fund—up from $275 million in 2024—and a $400 million opportunity fund. The New York–based firm is shrinking its general partnership to four investors and aims to lead more AI-focused investment rounds.

Apple plans 2nd October launch around Oct 27: touch-screen MacBook, iPad mini, M6 MacBook Pro, iMac

Apple is planning a second October product launch, on or around October 27, after next week’s revamped smart home event. According to Bloomberg sources, the launch will include Apple’s first touch-screen MacBook, a new iPad mini, and M6-chip versions of the 14-inch MacBook Pro and iMac.

FTC probes Block over frozen/deactivated Square accounts, service, layoffs, 1,500+ complaints

The FTC is investigating whether Block Inc., the parent of Square and Cash App, violated US consumer protection laws by providing inadequate customer service for frozen or deactivated accounts amid staffing cuts. The probe, disclosed in court filings, began last year after more than 1,500 consumer complaints, and the FTC has sought to compel Block to turn over documents.

Apple promotes Carson Oliver as App Store VP, names Steve Smith M&A chief, replacing Adrian Perica

Apple has promoted Carson Oliver to VP of the App Store and named Steve Smith as its new M&A chief, replacing Adrian Perica, according to people familiar with the matter. Perica will remain in charge of iCloud, Fitness+, and News, reporting to services chief Eddy Cue. The changes mark the latest management shake-up under CEO John Ternus.

Firmus IPO postponed amid weak demand

Nvidia-backed Australian data center company Firmus Grid is set to postpone its IPO and is weighing a private funding round instead, amid investor concerns over its pricing strategy. Its planned $5.5 billion listing failed to attract adequate support for the A$11 marketed share price, despite order-taking closing as scheduled, with no clear price or deal structure indicated. The weak demand signals fresh cracks in the AI funding boom.

Trump: WH calls ’the enemy’ anyone using ‘Artificial Intelligence,’ not ‘Super Intelligence’

The provided page does not contain a usable news article. It only displays a notice asking users to enable JavaScript to use the website. As a result, no substantive details from the purported article can be summarized.

Apple ‘Welcome home’ event Oct 13, 9AM ET NYC: HomePad, Apple TV, LG accessories expected

Apple announced a special “Welcome home” event for Tuesday, October 13 at 9 a.m. ET in Tribeca, New York, with press invites sent out. The event is expected to unveil new Home products, including a “HomePad,” a new Apple TV, and LG home accessories co-developed with Apple. Apple has not confirmed full details, and it remains unclear whether a full-length presentation video will debut online.

European VC funding rose 77% YoY to $25B in Q3, a 4-year high; 16% of global VC, led by AI

European venture funding reached $25 billion in Q3, up 77% year over year and the region’s strongest quarter in four years. AI startups drove the gains, accounting for $18.8 billion, or 75% of funding, led by major rounds including Mistral’s $3.5 billion raise.

Amazon retires Fire tablets after 15 years, launches Android Alexa Tablets Oct 14, incl. $499 12 Pro

Amazon is retiring its Fire tablet brand after 15 years and launching Alexa Tablets that run full Android with Google Play, going on sale October 14. The lineup includes the 8.7-inch Tablet 8 at $229.99, the 11-inch Tablet 11 at $329.99, and the 12.5-inch Alexa Tablet 12 Pro at $499, with a focus on Alexa Plus and AI.

Spotify, Joe Rogan renew multiyear licensing/ad sales deal; terms similar to $250M 2024 pact

Spotify has renewed its multiyear licensing and ad-sales deal with Joe Rogan for “The Joe Rogan Experience.” Sources say the terms are similar to Rogan’s previous deal, which was estimated at $250 million and signed in 2024. The renewal comes as the show remains No. 1 in U.S. listenership.

Xbox launches XP division to manage four non-gaming areas: Pictures, Products, Places, Partnerships

Xbox is launching a new division called XP to centralize its non-gaming experiences, including film and TV, consumer products, live events, and partnerships. XP will oversee four areas—Xbox Pictures, Xbox Products, Xbox Places, and Xbox Partnerships—and will be led by Kayleen Walters, while Maria Angelidou-Smith becomes CEO of Mojang and Minecraft on October 19th. Xbox is already developing adaptations such as an A Minecraft Movie sequel, Fallout season 3, a Call of Duty film, and a Diablo animated series.

Finland data center investments surpass €67B on AI demand, cool climate, renewables

Finland’s planned and active data center investments now exceed €67 billion ($75 billion), fueled by AI demand, cool climate conditions, renewable energy, and available grid connections. The boom has accelerated since Google announced its largest-ever European investment there, with companies including Alibaba and Applied Digital unveiling new projects in recent weeks.

Huawei aims to grow overseas phone share, expand HarmonyOS globally in 1-3 years as EV sales slow

Huawei aims to increase its overseas smartphone market share and expand HarmonyOS globally within one to three years, executive Richard Yu said ahead of the Mate 90 launch. The company is focusing on phones with homegrown chips as its Huawei-powered EV sales slow, with September deliveries down 29% year over year amid a weakening Chinese auto market.

Fed officials cite AI investments, not tariffs, as driver of rising goods inflation

Minutes from the Federal Reserve’s September meeting show officials increasingly attributed rising goods inflation to surging AI-related investments rather than tariffs. They cited the AI buildout, including data center demand, as a factor in inflation pressures.

India’s Global Capability Centers employ 2.4M, automate entry-level tasks, hire far fewer graduates

India’s Global Capability Centers, which employ about 2.4 million people across firms like JPMorgan, are automating entry-level tasks and hiring far fewer graduates. Driven by AI, the shift is leaving fresh graduates struggling to find roles these centers once provided in large numbers.

Garmin CEO Cliff Pemble: diversified GPS lines, grew revenue from $2.6B in 2013 to a record $7.2B

Cliff Pemble, Garmin’s CEO and its sixth employee hired in 1989, has helped diversify the company’s GPS product lines across fitness, outdoor, marine, automotive, and aviation. Under his leadership, Garmin grew annual revenue from $2.6 billion in 2013 to a record $7.2 billion last year.

SemiAnalysis: China 24 GW operational compute, 50 GW planned; US 56 GW operational

SemiAnalysis data reported by the Financial Times shows China has 24 GW of operational compute capacity, with another 50 GW planned or under construction, compared with 56 GW operational in the US. Beijing is rapidly rolling out AI infrastructure, including in Inner Mongolia.

Elon Musk claimed unnamed “oligarchs” are blocking Starlink’s India launch to protect a “monopolistic chokehold,” as Jio and Airtel dominate over 80% of the telecom market. SpaceX signed MOUs with Jio and Airtel in 2025, but Starlink has yet to launch amid regulatory scrutiny over security and approval concerns.

I’ve Been Deindexed by Google

The author decided to remove their website from search engines due to concerns about AI scraping. After blocking all bots in robots.txt, they added a noindex meta tag and had to allow Googlebot to crawl the site so it could read the tag. After at least three months, they confirmed their website had been deindexed by Google.

🧪 Science & Culture

The things you own end up owning you

NumPy is being ported to Python’s Limited API to enable abi3 wheels that work across multiple Python versions, reducing the per-version wheel matrix and easing support for new releases. The work also advances subinterpreter support by replacing static types and global state with heap types and per-module state, though multiarray remains the hardest part and abi3 wheels are not ready yet.

Making a mathematical model for better Indic keyboards

The article presents a mathematical model for more efficient Indic touchscreen keyboards, starting with Tamil by mapping its script structure into Boolean logic and compact input states. It uses Python with SymPy to derive and simplify the logic, then implements the resulting input engine in TypeScript. The approach is intended to be generalizable to other Indic scripts.

How to watch Avatar: Seven Havens from anywhere

Avatar: Seven Havens, a new animated series in the Avatar: The Last Airbender universe, premieres on Paramount+ on October 9, 2026, with 26 half-hour episodes split across two books. It will be available on Paramount+ in countries including the US, UK, Australia, and Canada. Travelers with an existing Paramount+ subscription can try using a VPN to access their home-country service, though access is not guaranteed and platform terms and local rules still apply.

How to watch Fire Country from anywhere

“Fire Country” follows Bode Leone, a young convict who joins a firefighting program hoping for redemption and a shorter sentence. The series streams on Paramount+ and CBS in the U.S., Sky and NOW in the U.K., Paramount+ and Global TV in Canada, and Paramount+ in Australia. Travelers may use a VPN to try accessing their home streaming service, but access is not guaranteed and a VPN does not provide a subscription.

Is Luis Elizondo becoming the government’s UFO megaphone?

Former Pentagon official Luis Elizondo says he has returned to Washington, DC, to work behind the scenes on UAP issues, but he declined to identify any department, project, or formal role. His vagueness frustrated some UFO disclosure supporters, with Reddit commenters accusing him of acting as a government megaphone rather than a whistleblower. Elizondo maintains that politicians are receptive to his advisory efforts on disclosure and transparency.

ADHD as a circadian rhythm disorder: evidence and implications for chronotherapy

The article reviews evidence that circadian rhythm dysfunction is common in ADHD, marked by evening chronotype, delayed sleep-wake timing, insomnia, and delayed biological markers such as DLMO, cortisol, and clock-gene rhythms. It reports that melatonin, bright light, sleep programs, and behavioral interventions can advance circadian phase and are associated with ADHD symptom improvement. The authors propose a pragmatic, behavioral-first clinical pathway using sleep/circadian screening, chronotype and sleep tracking, fixed wake times, morning bright light, evening light restriction, regularized zeitgebers, and selective low-dose melatonin for confirmed or probable DLMO delays.

The value of not getting to the point (2015)

The article argues that casual conversation, small talk, and rituals like coffee or meals help create enough comfort and trust for people to discuss sensitive matters. Through an anecdote about his college daughter, who needed a long warmup before raising her real concern, the author realizes why not getting straight to the point can be valuable. He contrasts this with his own get-to-the-point approach, which had made him miss the purpose of conversational rituals.

Making a flexible “neon” t-shirt with LED filaments

The article describes how the author custom-built a wearable “neon” t-shirt using flexible LED filaments for a work event announced less than a week in advance. They sewed uncuttable LED filament shapes onto a black shirt backed with stiff fabric, designing paths in Inkscape to match available filament lengths, and powered the lights with a USB battery bank, ESP32, and boost/constant-current drivers. It also outlines the parts, basic tests, tracing/sewing process, and wiring needed to replicate the project.

Why people don’t want to have children any more

The article could not be summarized because the provided content is a security verification error page, not the actual news story. It displays an FT error code CG000 / 403 and request ID, indicating access was blocked.

I’ve been paying for a rural Tanzanian’s education for 10 years

A volunteer who began paying school fees for an 11-year-old Tanzanian girl named Anwarite a decade ago has seen her graduate from the University of Dar es Salaam and be accepted into a Master of Arts in Swahili program costing about $1,161. The effort became the Tanzania Education Project, a 501(c)(3) that now supports five students from Ibumila with school fees and healthcare and aims to expand education and healthcare access across the village.

Orkut.com

The article titled “Orkut.com” contains no substantive reporting; its content is limited to the domain name “orkut.com” and a list of language options including English, German, Spanish, Estonian, French, Italian, Japanese, Portuguese, and Turkish. No actual news details or developments are provided.

Beauty in DVD Menus

DVD menus became a key selling point after VHS and flourished creatively in the 2000s, though some early or budget releases omitted them. Their design was shaped by technical constraints including NTSC/PAL resolutions, storage versus display aspect ratios, and 4:3 safe areas during the widescreen transition. Standard menu features typically included playback, chapter selection, extras, and settings.

Is society growing stupider? The clues are simple and repeated

The idea that society is growing less intelligent has moved from fringe polemic to mainstream, fueled by widespread belief that smartphones are harming our brains. Studies show IQ declines over the past two decades—sharpest among the age groups that use phones most—while students report worsening concentration and memory, and even a nearby phone may impair academic performance.

Wordminos – Crosswords Meet Dominos

A developer built the daily word game Wordminos in 12 days. The core mechanic took only an hour, while the remaining time was spent on iteration, design, polish, juice, and character. The creator notes that taste remains relevant.

Anne Carson wins Nobel Prize in literature 2026

The Swedish Academy awarded the 2026 Nobel Prize in Literature to Canadian poet and essayist Anne Carson, praising her bold, inventive work that engages playfully with the classical tradition to create new forms for contemporary literature. Born in Toronto in 1950, Carson is known for blending poetry, prose, and translation, drawing heavily on ancient Greek literature and mythology in works such as Autobiography of Red.

Vanillin provides a sweet solution for chronic wound healing

A Flinders University team has found that vanillin, the main flavor compound in vanilla that can also be synthetically produced, has antioxidant, anti-inflammatory, and antibacterial properties suited to chronic wound-healing formulations. Because it is abundant, low-cost, and amphiphilic, vanillin could enable safe, multifunctional treatments for ulcers and targeted drug delivery, with further progress depending on collaboration among academia, clinicians, and industry.

The OLED Burn-In Test: 30-Month Update

After 30 months of intentionally using an MSI MPG 321URX 4K QD-OLED monitor almost exclusively for static productivity work, the test has reached about 8,000 hours of use at roughly 200 nits. The monitor has completed 865 compensation cycles, about half the recommended number, and visible burn-in remains most apparent in the center of the screen. This update compares burn-in progression at 6, 18, 24, and 30 months.

LSD reconfigs cortex thru faster brain rhythms and increased fractal dimension

The provided page did not load its article content; it only displays messages that JavaScript is disabled and that a required site component failed to load. The headline references LSD reconfiguring the cortex through faster brain rhythms and increased fractal dimension, but no article details are available to summarize.

Classic PC demoscene productions running natively in the browser

Classic PC demoscene productions are now playable in the browser via a recompilation process: original DOS code is recorded on an x86 emulator, translated instruction-for-instruction into C, and compiled to WebAssembly alongside models of the timer, VGA card, and Sound Blaster. The results are validated event-for-event against the emulator, while the demos’ original loaders, music players, and effects run as they did in 1992–1993. The site serves the original release files unmodified and counts visits anonymously.

Scientists get a new clue to the origin of solar wind ‘switchbacks’

Scientists have found a new clue to the origin of solar wind “switchbacks,” which may help explain the behavior of the Sun. The discovery contributes to ongoing efforts to understand why our nearest star behaves as it does.

It’s a bird, it’s a plane… it’s a droppable robot boat with a spinning parachute

The ALBATROSS, a droppable robot boat, has landed. It is described as more of a spinning parachute than a bird or plane, but ultimately as a robotic maritime vehicle.

Universal Quantum raises $100M+ Series A led by DCVC, Firgun to expand hubs in Germany, Singapore

Universal Quantum, a UK-based quantum computing startup, has raised more than $100 million in a Series A round led by DCVC and Firgun Ventures. The company plans to use the funding to expand its research hubs in Germany and Singapore.

Thinking will become a hobby

The article argues that AI is rapidly surpassing humans in cognitive work because it is better performing, cheaper, and always available, making the replacement of knowledge jobs increasingly inevitable. Despite likely resistance and regulation, the author predicts cognitive work will diminish and human thought will become a leisure pursuit, much as physical activities did after past technological revolutions.