Octomind launched as an open-source AI coding-agent runtime, claiming 24 of 25 held-out tasks solved at lower cost. Users report OpenAI silently re-enabled model training on Codex chats despite opt-outs; disable via privacy controls. Anthropic’s IPO filing warns future models may be imminently close to recursive self-improvement, though none have crossed it. itch.io expanded generative-AI disclosure filters.
🤖 AI & Automation
Octomind – open-source AI harness that saves your tokens, static rust
Octomind is a CLI-first open-source AI coding agent runtime that gives LLMs repo access via MCP tools for file operations, shell commands, code search, and sub-agents, supporting interactive, piped, daemon, WebSocket, and ACP workflows. It is built for autonomous use with TOML-based policy guardrails, flexible model/provider selection, session compression, and packaged specialist agents from taps. Its README claims it solved 24 of 25 held-out tasks at lower cost than the closest competitor and includes installation and quick-start instructions.
AI agents hallucinate prices: LLM grounding with live data
The article explains that AI agents often report wrong product prices from stale memory, bad web pages, or tools that falsely report success. It tests models that matched live prices for only 2 of 15 products from memory, and even with web search still produced wrong prices despite source links. It recommends grounded agents that use live scraped data, detect failed checks clearly, and let code—not the model—insert each price into the answer.
- AI agents hallucinate prices: LLM grounding with live data — blog.apify.com
Open Models Slash Agentic AI Costs
Gartner reports that open models can perform about 90% of agentic work at roughly one-twentieth the cost of a frontier LLM. It argues that models are becoming commodities, while the surrounding harness is the valuable part.
- Open models can do 90 percent of agentic work at a twentieth of the cost — theregister.com
Nvidia – open-source auto-ontology
NVIDIA Auto Ontology is an open-source library that builds a reliable context layer from enterprise metadata, reconciles inconsistent definitions across systems, and exposes that context to agent harnesses through MCP, CLI, and API. It helps agents discover, ground, validate, and learn business meaning for metrics, entities, and relationships, complementing customers’ existing semantic models.
- Nvidia – open-source auto-ontology — github.com
The Nordic Token Tax
A study of 3,000 identical sentences across 12 recent AI models from nine labs found that every model charged more for Swedish, Norwegian, Danish, and Finnish than for English because non-English text is split into more tokens. Swedish cost 32% to 68% more, while Finnish cost up to twice as much. The extra tokens also mean higher billing and slower responses for the same content.
- The Nordic Token Tax — research.klang.ai
Own Your HackerTube,an AI Curator Inspired by Karpathy’s HackerTube
Own Your HackerTube is an AI curation tool inspired by Karpathy’s HackerTube that lets users follow trusted YouTube channels and define their own standards. Each day, the AI checks new videos, scores them against those criteria, and publishes worthwhile picks with reasons and creator chapter-based key moments on a personal YouTube-style site. It learns from user feedback to refine future selections and is free to start with one channel and 30 AI checks per month.
- Own Your HackerTube,an AI Curator Inspired by Karpathy’s HackerTube — ownhackertube.cc
World of Warcraft: Forever Is Already Overrun with AI Slop
World of Warcraft: Forever, launching Wednesday, is already seeing subreddits overrun with AI-generated addons that rip off longtime developers’ work. Warcraft Logs is changing its API to limit commercial projects and scrapers that surface basic data, citing an explosion of development and at least one commercial request per day.
Forget retraining - AI can now tweak another AI’s weights
A technique called “Weight-Space Programming” allows one AI to directly tweak another AI’s weights instead of retraining it. The method is reported to boost benchmark scores by adjusting internal parameters.
- Forget retraining - AI can now tweak another AI’s weights — theregister.com
Sources: Anthropic’s 15-person chip team uses Claude for custom chip design work
Anthropic has told investors that its 15-person chip design team is already using Claude for engineering and design work on its custom chips, according to The Information. The company is touting Claude’s chip-design capabilities to investors, while its AI models have also helped uncover serious cyber vulnerabilities and transform software coding work.
- Sources: Anthropic’s 15-person chip team uses Claude for custom chip design work — theinformation.com
Google expands Dreambeans to India, first market outside US, with AI visual stories in 13 languages
Google has launched its experimental AI app Dreambeans in India, its first market outside the U.S., with Android and iOS support for 13 local languages including Hindi, Punjabi, Tamil, Telugu, and Bengali. Dreambeans uses users’ connected Google services to generate personalized AI-illustrated stories without prompts, and Google cited strong Indian interest and high usage overlap across Gemini, Gmail, and Google Photos. The app currently limits users to one language at a time, which may constrain appeal in multilingual India, and its reliance on Google account data raises privacy and data-access questions.
- Google expands Dreambeans to India, first market outside US, with AI visual stories in 13 languages — techcrunch.com
🔒 Security & Privacy
OpenAI seems to be silently re-enabling model training on Codex chats
Users report that OpenAI appeared to silently re-enable model training on Codex chats by resetting the “Improve the model for everyone” data-sharing setting, despite prior opt-outs. The issue has reportedly been noted before, and users can disable data sharing through ChatGPT’s data controls or OpenAI’s privacy settings.
Police Dept. Quits After Finding Out What It Feels Like to Be Tracked
Milner, Georgia Mayor Craig Timmons acknowledged placing a GPS tracker on Police Chief Lamar Bunn’s vehicle and arranged a surveillance camera toward City Hall, citing concerns about Bunn’s timesheets. Bunn, all five full-time officers, most reserve personnel, and City Manager Tausha Grose resigned, effective October 28, 2026, leaving the town without a full-time police force. The Lamar County Sheriff’s Office was reported to be providing coverage, though details were not officially confirmed.
- Police Dept. Quits After Finding Out What It Feels Like to Be Tracked — gadgetreview.com
Mysterious iPhone malware puts German politicians on alert
Several German lawmakers and public figures have received Apple alerts warning of possible mercenary spyware targeting on their iPhones. Germany’s BSI says it is unclear whether the devices were actually compromised or only targeted, though such spyware can infect phones without a click and access messages, calls, location, camera, and microphone. Affected parliament members have been advised to treat their devices as potentially compromised.
- Mysterious iPhone malware puts German politicians on alert — cybernews.com
Stop counting stolen passwords: Here’s what really brings organizations to their knees
Outpost24 found that 72% of 4,824 organizations in the US and Europe had credentials linked to infostealer malware logs over 12 months. Infostealers can steal session cookies, allowing attackers to access accounts even after passwords are changed; less than 3% of exposed credentials came from these logs, yet they affected nearly three-quarters of organizations. Outpost24 recommends investigating infected devices and revoking active sessions rather than only resetting passwords.
- Stop counting stolen passwords: Here’s what really brings organizations to their knees — cybernews.com
Harvard breach compromised records from 851,000 unique accounts
Have I Been Pwned confirmed that a Harvard data breach exposed 850,500 unique accounts. Harvard said attackers accessed Alumni Affairs and Development systems through phone-based phishing, and the leaked data includes names, emails, birth dates, phone numbers, addresses, donation details, and event attendance. Harvard said the affected systems did not store Social Security numbers, passwords, payment card data, or financial data.
- Harvard breach compromised records from 851,000 unique accounts — cybernews.com
Explainer: Could hackers get inside my fridge?
A cyberattack raised temperatures in supermarket fridges and freezers in several European countries, following similar incidents that spoiled food and caused fires. Any internet-connected device—from smart fridges and cameras to locks and thermostats—can potentially be hacked, though routers account for more than three-quarters of attacks on connected devices. Experts recommend strong passwords, two-factor authentication, regular updates, and removing devices that no longer receive security fixes.
- Explainer: Could hackers get inside my fridge? — cybernews.com
Hacker groups claim 23TB Rosstat breach exposes Russia’s hidden war loss data
Belarusian Cyber Partisans and Silent Crow claim they breached Russia’s Rosstat, stole about 23TB of data, and accessed records suggesting up to 50,000 Russian soldiers die in Ukraine each month. Rosstat confirmed targeted cyberattacks but said its databases were not compromised. The claim comes as Moscow stopped publishing cause-of-death data in 2024, limiting public insight into war losses.
GM settles claims over sharing drivers’ location and behavior data with insurers
General Motors agreed to pay $220 million to settle claims by 35 U.S. states that it unlawfully shared drivers’ location and driving-behavior data with data brokers, which passed it to insurers to market products and adjust premiums. GM must now obtain drivers’ consent before collecting or sharing such data.
The UK’s spy agency struggles to locate thousands of USB sticks
The UK’s GCHQ cannot immediately locate thousands of removable storage devices, including USB sticks and hard drives, some of which may contain sensitive data tied to the Investigatory Powers Act. The IPCO says the number is significant and will require GCHQ to report progress throughout 2026. GCHQ says its tracking was not robust enough but that it has addressed the regulator’s recommendations and improved media tracking.
- The UK’s spy agency struggles to locate thousands of USB sticks — cybernews.com
ShinyHunters hit 850k alumni, donors, students, and fam in last year’s Harvard University raid
ShinyHunters reportedly breached Harvard University last year, exposing personal and donation data belonging to about 850,000 alumni, donors, students, and family members. A data-sifting breach site disclosed the claim a year after the attack.
- ShinyHunters hit 850k alumni, donors, students, and fam in last year’s Harvard University raid — theregister.com
Britain on notice: AI-powered ransomware attacks to arrive by 2028
Britain’s cyber agency has warned that AI-powered ransomware attacks are expected to arrive by 2028, with critical infrastructure a likely target. Its chief urged bosses to stop doing “tech on the cheap” and improve cybersecurity preparedness.
- Britain on notice: AI-powered ransomware attacks to arrive by 2028 — theregister.com
Chainalysis, Elliptic: North Korean hackers may take months to launder Binance proceeds via DeFi
The provided text contains only navigation elements and promotional snippets, including podcast listen times and invitations to explore Financial Times events and exclusive communities. It does not include an article or substantive content to summarize.
- Chainalysis, Elliptic: North Korean hackers may take months to launder Binance proceeds via DeFi — ft.com
UK NCSC: AI transforming cyber; China/Russia to routinely use it by 2028
The UK’s National Cyber Security Centre says AI is “transforming cyber operations” and expects China and Russia to routinely use AI-enabled cyber capabilities to pursue geopolitical goals by 2028. The NCSC head is set to warn that security defences, particularly for critical infrastructure, are inadequate.
OpenAI partners with Cathedral to test security, revamps bug bounty, 4x critical rewards
OpenAI is partnering with Cathedral for its next internal security sprint, with Cathedral researchers using OpenAI models alongside OpenAI teams to test and strengthen its defenses. OpenAI is also revamping its Security Bug Bounty program, including more than quadrupling rewards for critical findings, as part of an ongoing effort to use AI to find and fix system weaknesses.
- OpenAI partners with Cathedral to test security, revamps bug bounty, 4x critical rewards — openai.com
Firefox Security & Privacy Newsletter 2026 Q3
Mozilla’s Q3 2026 Firefox Security & Privacy Newsletter highlights a faster release cycle, more security fixes aided by frontier LLMs, and a shift to separate advisories for each internally identified memory-safety bug. It also covers upcoming Firefox 157 warnings for potentially non-private TLS configurations, stricter WebPKI requirements under Root Store Policy v3.1, and new protections that treat mailto: links as popups and require explicit extension permission for file:// access.
- Firefox Security & Privacy Newsletter 2026 Q3 — attackanddefense.dev
💻 Software & Open Source
Bema: Work messaging built for humans
Ben, founder of Bema, has launched a work messaging platform designed to encourage asynchronous communication while offering an urgent “break glass” option when needed. The tool aims to address common complaints about existing work messaging software, including too many channels, lost information, and distractions from actual work.
- Bema: Work messaging built for humans — getbema.app
uBlock Origin filter to block autoplay videos in TechCrunch
The article explains that TechCrunch articles autoplay JW Player videos and load Connatix’s floating ad player, which continue running even when hidden. It provides uBlock Origin filter rules for Firefox’s full MV2 version to block the scripts and hide the empty player box, noting that new network rules may be needed if TechCrunch changes video vendors.
- uBlock Origin filter to block autoplay videos in TechCrunch — cornfieldlabs.github.io
Open a 300k-row XLSX in the browser in ~300ms, no server
@p10node/office-engine is an npm library for viewing and editing DOCX, XLSX, and PPTX files directly in the browser, with progressive loading, virtualized rendering, and save-back support. It can open a 300,000-row XLSX in about 300ms without a server, using HTTP Range requests and Web Worker parsing, and offers a one-line React embed. It also supports real-time collaboration and a Google Docs-style toolbar for document, spreadsheet, and slide editing.
I built a little browser game to practice Vim commands
The developer created a browser game called hjkl to help themselves and others practice and remember Vim commands. More commands are planned, but the game is already available for casual Vim practice.
- I built a little browser game to practice Vim commands — hjkl.duy.run
Rustable is an open source database client built in Rust
Rustable is an open-source database client written in Rust and compiled to WASM for browser use, with an egui-based workspace for browsing, querying, editing, importing, and exporting data. It includes real SQLite and in-memory Redis demos, while saved queries persist in the browser and demo reloads reset sample data. Version 0.4.8 adds native macOS, Windows, and Linux builds and supports PostgreSQL, SQLite, MySQL 8.4, MariaDB 11.4, and standalone Redis 7.4, with server connections limited to the desktop app.
- Rustable is an open source database client built in Rust — rustable.app
Hacker Times, a New York Times-Inspired Version Hacker News
Hacker Times is a New York Times-inspired version of Hacker News designed for browsing the site like a newspaper. It updates once an hour.
- Hacker Times, a New York Times-Inspired Version Hacker News — hackertimes.ca
DiscImageStudio: burn visible artwork onto CDs, DVDs and Blu-ray discs
DiscImageStudio is a tool that converts images into disc data so CD, DVD, and single-layer Blu-ray surfaces show visible artwork after burning. It offers a Windows desktop app for design and burning, plus an experimental Linux/macOS command-line version for offline image generation. It can export ISO, RAW, and WAV+CUE formats and includes experimental live streaming burn, with a warning to use only blank or disposable test media.
Vibe-Coded AO3 Clone Down After 2,000 Fanfic Takedown Notices in One Day
ao3.life, a vibe-coded site that scraped AO3 fanfiction and sold AI-generated audiobook versions without authors’ permission, went down Monday evening after widespread backlash. Its creator said the site received more than 2,000 takedown requests from authors in the single day it was online.
pfSense merges paid and community editions: all features now free except one
Netgate will merge pfSense CE and Plus into a single release starting with version 26.10, giving free users all formerly paid features plus security and bug fixes on supported hardware. However, free users will lose online major-version upgrades after the final CE 2.9 to 26.10 upgrade and must reinstall manually, while online upgrades will require a subscription or Netgate hardware. The change has divided the community, with some welcoming the full feature set and others threatening to switch to OPNsense over added maintenance and plugin concerns.
Microsoft is testing a new PC recovery option
Microsoft is testing Cloud rebuild, a Windows 11 recovery feature that reinstalls the OS from the cloud even if a PC cannot boot, without requiring USB media. It can also erase files or securely wipe a drive before reinstalling Windows. The feature is rolling out gradually with Windows 11 versions 25H2 and 26H2.
- Microsoft is testing a new PC recovery option — cybernews.com
Pokémon bakeoff reveals best JavaScript frameworks for AI
A Pokémon-themed bakeoff tested JavaScript frameworks for AI, with Claude favoring lean, HTML-first tools over bloated single-page applications. React enthusiasts may be disappointed, as the results do not favor React.
- Pokémon bakeoff reveals best JavaScript frameworks for AI — theregister.com
Everyone has the file. Nobody has the right version.
Distributed teams need to collaborate on shared information, but duplicated files, delays, and competing versions keep getting in the way. The article highlights that everyone may have a copy of a file, yet nobody has the right version.
- Everyone has the file. Nobody has the right version. — theregister.com
Architect creates open-source replacement for VM migration tool VMware has hidden
An architect has created OpenVDDK, an open-source replacement for a VMware VM migration tool that VMware has hidden. The project is reportedly already being used in production at scale.
Fedora Linux 45 Released
Fedora Linux 45 has been released across Workstation, KDE Plasma, Server, Cloud, CoreOS, IoT, Atomic Desktops, and Spins after two release delays. It adds Stratis storage installation support, KMSCON, improved Qualcomm Snapdragon X laptop support, ptrace restrictions and signature enforcement, reproducible package builds, and a DRM Panic front-end, while updating to GNOME 51, Go 1.27, Python 3.15, GCC 16.2, and Linux 7.2 and retiring unmaintained NetworkManager VPN plugins.
- Fedora Linux 45 released — fedoraproject.org
- Fedora 45 Released With Many Leading-Edge Linux Innovations — phoronix.com
The Ubuntu 26.04 Rust sudo is not compatible with standard sudo
The author is blocking browsers with old or suspicious User-Agent strings to reduce load from high-volume crawlers, especially those apparently collecting data for LLM training. He notes that Inoreader, Feedly, Vivaldi users, and archive.* services may encounter the block due to their fetching or masking behavior, and suggests contacting him if using a current browser or using archive.org instead.
- The Ubuntu 26.04 Rust sudo is not compatible with standard sudo — utcc.utoronto.ca
Death comes for XSLT
The article criticizes Google and other browser vendors for deprecating and removing XSLT from Chrome/Chromium, arguing that this breaks the web’s backwards-compatibility social contract. The author created XSLT.RIP, a site built with XSLT that will stop working when removal takes effect—reportedly on 17 November 2026—and notes that some Chromium-based browsers already inject warning banners into XSLT sites. It calls Google’s approach arrogant and intrusive for blaming website owners.
- Death comes for XSLT — dbushell.com
Ideas for social networks
The author reflects on their motivation to build social software in the age of AI-assisted coding, concluding they are driven by creating tools people use. They recall their earlier social network Sweet, a cute, queer, ad-free community with chronological feeds and no like metrics, which closed due to moderation and maintenance burdens. They argue for better communal online spaces with safety, privacy, data sovereignty, and interoperability, and list mostly bad social-network ideas, including an allotment-style network with 128 user plots.
- Ideas for social networks — raphaelkabo.com
openSUSE Leap 16.1 Released
openSUSE Leap 16.1 has been released as the first minor release in the Leap 16 series, built from the same sources as SUSE Linux Enterprise Server 16.1. It integrates Leap Micro into Leap through a new Immutable Mode for immutable server deployments, and includes KDE Plasma 6.6, GNOME 48, and Linux 6.12 LTS as the default kernel.
- openSUSE Leap 16.1 Released With Immutable Sever Integration — phoronix.com
- openSUSE Leap 16.1 — en.opensuse.org
git-loom 1.0
git-loom 1.0 has been released. It weaves multiple feature branches into a single integration branch, letting developers commit, amend, split, and reorder across them without switching branches. Inspired by jujutsu and GitButler, it offers command-line and interactive TUI workflows.
- git-loom 1.0 — narnaud.github.io
RadeonSI & RADV Now Enabling D16 Support For GFX8 Era GPUs
Valve’s Georg Lehmann has enabled D16 support for AMD GFX8-era GPUs, covering GCN Volcanic Islands through Polaris, in both RadeonSI OpenGL and RADV Vulkan. The change arrives via 17 patches in Mesa 27.0-devel and extends D16—previously limited to GFX9 and newer—to help reduce memory bandwidth pressure and improve register efficiency on aging AMD hardware.
- RadeonSI & RADV Now Enabling D16 Support For GFX8 Era GPUs — phoronix.com
Linux Preparing Faster FAT32 Performance In 2027 With Up To 4.06x Improvement
Linux is preparing further FAT32 performance improvements, likely for Linux 7.5, with benchmarks showing up to a 4.06x gain on synchronous mounts. The patch by Matheus Alves de Almeida batches FAT mirror writes and improves submission, addressing serialized I/O in the current driver. It is currently in Andrew Morton’s mm-everything branch and could land in the new year.
The Linux Kernel Preparing To Remove The Last Of Its MMX Optimized Code
The Linux kernel is preparing to remove its last MMX-specific optimization, a patch that drops MMX-optimized code from the software RAID xor path. The change, likely for Linux 7.5 in 2027, affects outdated CPUs such as Intel Pentium MMX and Pentium II, which will fall back to scalar code. This RAID code was the kernel’s only remaining MMX optimization.
Intel Compute Runtime Continues Prepping For GPU D Series, OpenCL LEO for Nova Lake P
Intel released Compute Runtime 26.40.40196.8, continuing work on its GPU D Series (Crescent Island) AI accelerator with unified shared memory and deferred backing enabled by default. The update also enables standby mode for Battlemage and Crescent Island with the Xe kernel driver, adds performance optimizations, and enables OpenCL LEO by default for Nova Lake P.
- Intel Compute Runtime Continues Prepping For GPU D Series, OpenCL LEO for Nova Lake P — phoronix.com
GCC 17 Compiler Merges Support For AVX10V1AUX
GCC 17 has merged initial support for AVX10V1AUX, an auxiliary update to AVX10.1 from the AMD-Intel ACE initiative. The support, led by AMD engineers and exposed via the -mavx10v1aux switch, adds VNNI INT8/INT16 and other incremental capabilities as a bridge to AVX10.2, though it is not yet tied to specific processor generations.
- GCC 17 Compiler Merges Support For AVX10V1AUX — phoronix.com
Intel Brings RPL-M/RPL-P SoundWire To Upstream Linux For Raptor Lake Powered Automobiles
Intel has posted patches to add SoundWire support for Raptor Lake P and M platforms to the upstream Linux Intel AVS driver. The support is for automotive audio streaming, including Intel’s Raptor Lake-based “Malibou Lake” software-defined vehicle platform, and marks the first time this work is being brought to the public upstream kernel. The patch series is now out for review.
- Intel Brings RPL-M/RPL-P SoundWire To Upstream Linux For Raptor Lake Powered Automobiles — phoronix.com
AerynOS 2026.11 Released With New Installer Option, Other Improvements
AerynOS 2026.11 has been released, adding the lichen-installer as a new alternative installer on the ISO for testing while retaining the current default. The release also brings Fwupd to the repository for firmware updates and offers Budgie as an install-time desktop option. Package updates include GNOME 51.1, KDE Plasma 6.8.2, COSMIC 1.11, Mesa 26.2.6, Wine 11.21, and Linux 7.2.14.
💼 Business & Economy
The AI Bubble Monitor
The provided article, titled “The AI Bubble Monitor,” contains no accessible body text beyond a notice requiring JavaScript and cookies. Therefore, no substantive news details or claims from the article can be summarized.
- The AI Bubble Monitor — cepr.net
Anthropic’s IPO Filing Warns It May Be ‘Imminently’ Close to RSI
Anthropic’s revised IPO filing warns that future models may imminently meet its internal definition of recursive self-improvement, while stating no current model has crossed that threshold. The company’s own published data shows the loop remains open: Claude fully autonomously performs zero AI research tasks, and humans still choose problems and judge results.
Charles Tandy, the mastermind behind Radio Shack
Charles David Tandy (1918–1978) was chairman, president, and CEO of the Tandy Corporation. He helped create or expand three retail chains—Tandy Leather, Radio Shack, and Pier 1 Imports—after growing his family’s leathercraft business into an international corporation. He also took the Tandy Leather Company public through a merger with American Hide and Leather Company.
- Charles Tandy, the mastermind behind Radio Shack — dfarq.homeip.net
How AI is creating a lopsided US economy
AI-related companies now dominate the US stock market, with tech stocks making up about 44% of the S&P 500’s value and Nvidia the world’s most valuable company at roughly $7.5 trillion. Experts warn this outsized influence is masking weaknesses elsewhere in the US economy and creating new market risks of its own. The concentration was highlighted at a White House AI Summit attended by tech billionaires whose combined net worth is estimated above $2.5 trillion.
- How AI is creating a lopsided US economy — abc.net.au
Net worth for people aged 30–33 in the U.S.
Americans aged 30–33 have a median net worth of $68,000, with the upper quarter starting at about $235,000 and the top 10% at $440,000. The lower quarter falls below $7,100 and the bottom 10% has negative net worth, while the average is $88,400, pulled up by the top end.
- Net worth for people aged 30–33 in the U.S. — keepingup.com
WA is losing more from Canada than shoppers
A Canadian software company founder says his firm is abandoning planned U.S. growth, including withdrawing from a Silicon Valley accelerator, after new U.S. tariffs and procurement restrictions made cross-border expansion too uncertain. He is redirecting growth to Canada, Europe and Taiwan, warning Washington that it will see falling border traffic but not the business partnerships and jobs that never materialize.
- WA is losing more from Canada than shoppers — seattletimes.com
Quitting jobs: The two-week notice is dying. Maybe it should
The article argues that the traditional two-week notice is declining as employees respond to dehumanizing, efficiency-obsessed workplaces, mass layoffs, and faceless hiring practices by quitting abruptly or “revenge quitting.” Experts say workplace etiquette around notice has broken down, with employees mirroring the uncaring treatment they receive from employers. Still, the piece notes that while employee ruthlessness is new, corporate heartlessness and asymmetric firing-versus-quitting dynamics have long existed.
Intel stock collapse worsens as $3.5T chip rout spreads
Chip stocks extended a sharp selloff Monday even as most S&P 500 stocks rose, with the semiconductor group losing more than $400 billion in market value in one session and about $3.7 trillion since Oct. 6. Intel led the decline, sliding about 9% after its worst day since 1974 and ending Monday worth less than half its end-September value. Memory and storage chipmakers also fell sharply, while Nvidia held up better than most but remained down about 16% since Oct. 6.
- Intel stock collapse worsens as $3.5T chip rout spreads — finance.yahoo.com
Virgin Media O2 outlines plans to lay off thousands as revenue falls
Virgin Media O2 plans to lay off thousands of staff as revenue falls, aiming for £800 million in cost savings by 2029. About £300 million of those savings will come from staff cuts.
- Virgin Media O2 outlines plans to lay off thousands as revenue falls — theregister.com
Apple’s Mac business is booming, but the memory market has other plans
Apple’s Mac business is booming, with new CEO John Ternus celebrating record sales. However, soaring RAM costs threaten to constrain supplies and squeeze margins.
- Apple’s Mac business is booming, but the memory market has other plans — theregister.com
Broadcom denied names of cloud providers backing VMware merger challenge
An EU court denied Broadcom’s request for the names of cloud providers backing CISPE’s challenge to the European Commission’s approval of the VMware merger. The confidentiality ruling clears a procedural hurdle but does not decide the merits of CISPE’s case.
- Broadcom denied names of cloud providers backing VMware merger challenge — theregister.com
October sponsors-only newsletter
The author sent out the October edition of their sponsors-only monthly newsletter. Topics include vibe game development, vulnerability news, LLM releases, API credits, current tools, and software releases. The September issue is available as a free preview, while sponsorship costs $10/month for early access.
- October sponsors-only newsletter — simonwillison.net
CrowdStrike to buy Fiddler AI for ~$400M, Dreadnode for ~$95M, mostly cash
CrowdStrike has agreed to acquire AI security startups Fiddler AI for about $400 million and Dreadnode for about $95 million, mostly in cash. The deals expand CrowdStrike’s push to monitor and contain AI agents as companies deploy the technology faster than some security teams can establish controls.
Corridian exits stealth with $38M seed led by Point72, plans first AUV in 2027
Corridian, a New England-based autonomous submarine startup, has emerged from stealth with a $38 million seed round led by Point72 Ventures. The company plans to unveil its first ultra-large autonomous undersea vehicle in 2027 and expects to double its workforce over the next year.
Shopify Q3 revenue up 32% to $3.74B; GMV +29%, net income +75%, adj EPS tops est.
Shopify reported third-quarter revenue of $3.74 billion, up 32% year over year but slightly below the $3.75 billion estimate. GMV rose 29% to $119.1 billion, net income climbed 75% to $462 million, and adjusted EPS came in at $0.49, beating the $0.47 estimate.
Ternus’ first call timid, scripted; Apple faces tough holiday quarter, 2027; AAPL down 4%+ premarket
Apple CEO John Ternus’s first earnings call was timid and heavily scripted, disappointing investors hoping for insight into his philosophy. Apple shares fell almost 3% during the call and were down more than 4% pre-market as the company faces a challenging holiday quarter and a difficult 2027.
- Ternus’ first call timid, scripted; Apple faces tough holiday quarter, 2027; AAPL down 4%+ premarket — bloomberg.com
Apple sees revenue growth despite memory costs
Apple’s fiscal Q4 inventory rose to $14.54 billion, up 31.1% quarter-over-quarter and 154.3% year-over-year, as it stockpiled components, especially memory; CEO John Ternus said memory costs are expected to rise in the December quarter and continue rising beyond it as Apple increases purchases to improve supply. For Q1, Apple guided revenue up 7%–10% year-over-year, above estimates, but gross margins of 45%–46%, below estimates, citing record or “unprecedented” memory chip costs. It predicted robust holiday hardware sales but a disappointing services outlook.
- Apple expects memory costs to rise into December and beyond; Q4 inventory hits $14.54B — techcrunch.com
- Apple sees Q1 revenue up 7%-10%, above est.; margins 45%-46%, below est. on record memory costs — ft.com
Larridin raises $30M Series A led by Pelion to track AI’s employee impact, guide adoption
Larridin, a San Francisco-based startup, raised a $30 million Series A led by Pelion Ventures to help companies track how AI affects employee productivity and identify areas for better adoption. Its measurement platform analyzes AI usage to assess ROI and training needs without storing employees’ prompt contents.
- Larridin raises $30M Series A led by Pelion to track AI’s employee impact, guide adoption — businessinsider.com
Salesforce plans to buy Cresta, AI contact-center software provider, for $2.7B
Salesforce announced plans on Monday to acquire Cresta, a nine-year-old provider of AI-powered contact center automation software. Salesforce agreed to pay $2.7 billion for Cresta, continuing its recent acquisition spree.
- Salesforce plans to buy Cresta, AI contact-center software provider, for $2.7B — theinformation.com
Match Group Q3 revenue down 3% to $891M, missing estimates; paying users fall 5%, Q4 EBITDA weak
Match Group reported Q3 revenue fell 3% year over year to $891 million, missing estimates, while total paying users dropped 5% to 13.8 million. It forecast Q4 adjusted EBITDA of $327.5 million to $332.5 million, below Wall Street projections, disappointing investors.
- Match Group Q3 revenue down 3% to $891M, missing estimates; paying users fall 5%, Q4 EBITDA weak — bloomberg.com
Trajectory raises $45M Sequoia-led Series A at $300M valuation for nontech AI post-training tools
Trajectory, a startup developing AI post-training tools that let nontechnical users improve models through feedback and continual learning, raised a $45 million Series A led by Sequoia at a $300 million valuation. The 18-person company is making its platform publicly available, with customers including Harvey, Rogo, Decagon, Mercor, and Nvidia.
- Trajectory raises $45M Sequoia-led Series A at $300M valuation for nontech AI post-training tools — techcrunch.com
The software industry: annealing, but wrong
Some software teams enforce small, single-purpose, fully tested pull requests to simplify review and reduce risk, though this can increase reviewer context switches. The article compares this to annealing: mature systems benefit from smaller changes, while AI coding tools readily produce large, interconnected changes that tend to be less coherent and more failure-prone. When AI-generated large changes yield good ideas, teams should still follow development processes and split them into small reviewable patches.
- The software industry: annealing, but wrong — apenwarr.ca
🌍 Policy & Society
Just a little AI resistance, as a treat
The author contacted Ocado about malformed calendar invites and received a customer service reply that seemed AI-generated, prompting them to complain in a satisfaction survey. They later received a personal response from a human apologizing and thanking them for the detailed diagnostic, which was passed to the technical team.
- Just a little AI resistance, as a treat — shkspr.mobi
Itch.io adding Generative AI disclosure filtering to few more places on the site
itch.io is expanding its Generative AI disclosure filtering to additional parts of the site, including its search results page, where it was previously missing. The platform says the change is intended to give users better control over what they see.
No Vehicles in the Park
The article introduces a 27-question game about language and rules, centered on a hypothetical park with the rule “No vehicles in the park.” Players decide whether each scenario violates the rule, can skip ahead after seven questions, and are told to disregard external legal or religious exceptions.
- No Vehicles in the Park — novehiclesinthepark.com
Too Expensive to Exist
A private Texas campground will discontinue individual camping and day-use activities starting October 2026, with its owner citing rising insurance costs she can no longer afford. The closure highlights the scarcity of publicly accessible open land in Texas, where a nearby state park was already booked for months.
- Too Expensive to Exist — mapwriting.substack.com
U.S. Strikes Iran’s Qeshm Island as Trump Talks Down War Risks
The U.S. attacked Iran’s Qeshm Island on Monday, the latest in a series of limited strikes Washington has secretly carried out against Iran in recent weeks. The strike came as President Trump touted the possibility of a diplomatic deal and played down the risk of returning to war with Iran after Tuesday’s midterm elections.
What Are the US Midterms
US midterm elections are being held Tuesday, with voters choosing all 435 House members and 35 Senate seats. The results will determine which party controls Congress and could significantly affect the rest of Donald Trump’s presidency. Key issues include the economy, inflation, immigration and concerns over AI.
- What Are the US Midterms — bbc.com
America Votes: ‘CEKEVPATIVG EDELDOA’
The article titled “America Votes: ‘CEKEVPATIVG EDELDOA’” is accessible only to paid members. Its summary says “Slopfest 2k26” continues with an election, a fanfic fiasco, and World of Slopcraft. The available text provides no further details.
- America Votes: ‘CEKEVPATIVG EDELDOA’ — 404media.co
How to Watch Barcelona vs Aston Villa for Free (November 3, 2026)
Barcelona will host Aston Villa in UEFA Champions League Matchday 4 at Spotify Camp Nou on November 3, 2026, at 9:00 PM local time. The article says fans can watch for free via Virgin Media Two in Ireland, JurnalTV in Moldova, or SBT in Brazil using a VPN, and notes Barcelona enter after a 6-1 La Liga win while Aston Villa have won their first three Champions League matches.
Democrats plan UFO disclosure committee as Trump trust wanes
Democrats have drafted a plan to establish a House UAP committee if they regain Congress, with Rep. Andre Carson accusing the current administration of delaying statutory disclosure requirements. UFO whistleblower David Grusch argued disclosure should not depend on one president, while Reddit users questioned whether either party would deliver faster transparency.
- Democrats plan UFO disclosure committee as Trump trust wanes — cybernews.com
NHS counts Palantir benefits at trusts that say they’re not using it
NHS figures credit Palantir with benefits at trusts that say they are not using its technology, with Medact flagging 29 disputed entries. The dispute comes as ministers weigh the future of a £330 million deal.
- NHS counts Palantir benefits at trusts that say they’re not using it — theregister.com
UK.gov lacks understanding of climate impact, says watchdog
The UK government lacks sufficient understanding of climate impacts, according to a watchdog. Government departments are poorly measuring the risks of a changing climate and the costs of transitioning to low-carbon energy.
- UK.gov lacks understanding of climate impact, says watchdog — theregister.com
OpenAI hires ex-Army Secretary Dan Driscoll for national security modernization, govt contracts
OpenAI has appointed former U.S. Army Secretary Dan Driscoll to oversee national-security modernization and advise on using AI to strengthen U.S. security capabilities and cyber defenses. The hire reflects OpenAI’s push to deepen ties with the defense sector, win government contracts, and shape policy on military and intelligence AI deployment.
- OpenAI hires ex-Army Secretary Dan Driscoll for national security modernization, govt contracts — wsj.com
China plans broad AI safety standards for domestic model makers after global incidents
The provided text contains only Bloomberg headlines and section labels, not a full article. Its listed topics include US voters’ referendum on Trump’s presidency, celebrity endorsements of prediction markets, Texas online schooling, business travel in Mumbai, and Wall Street investment in Indian hospitals. There is insufficient article content to provide a substantive summary.
- China plans broad AI safety standards for domestic model makers after global incidents — bloomberg.com
Alex Bores’ Who Decides seeks to make AI safety Democratic standard by 2028 amid tech-funded rivals
Alex Bores is launching Who Decides, an advocacy group with $10 million at launch and $20 million more planned, to push 2028 Democratic presidential candidates toward AI safety regulation and make it a standard party position. The effort follows his New York RAISE Act work and his primary loss, where a tech-funded super-PAC spent about $8 million to defeat him while rival AI-safety supporters spent $18 million.
- Alex Bores’ Who Decides seeks to make AI safety Democratic standard by 2028 amid tech-funded rivals — nymag.com
Nvidia tells cloud clients to sell 51%+ services to non-Chinese firms, curbing China AI-chip access
Nvidia has verbally told cloud-compute clients to sell at least 51% of their services to non-Chinese companies, according to Financial Times sources. The move aims to reduce exposure to Chinese customers and further restricts China’s access to AI chips.
- Nvidia tells cloud clients to sell 51%+ services to non-Chinese firms, curbing China AI-chip access — ft.com
🖥️ Hardware & Infrastructure
Could a modern chess engine on a Pentium have defeated Kasparov in 1996?
A custom port of the modern chess engine Stockfish running on a 200 MHz Intel Pentium Pro with 64MB RAM is estimated at 3300 Elo, far above Garry Kasparov’s roughly 2800. The author argues this suggests commodity 1996 hardware running a smarter chess engine could have beaten Kasparov, making IBM’s multimillion-dollar Deep Blue unnecessary.
- Could a modern chess engine on a Pentium have defeated Kasparov in 1996? — nicholas.carlini.com
Cloudflare experiencing multiple active incidents
Cloudflare’s system status page states that there are no known service issues and no recent incidents affecting its services. The article’s title claims Cloudflare is experiencing multiple active incidents, which conflicts with the status page content.
- Cloudflare experiencing multiple active incidents — cloudflarestatus.com
NASA’s Swift space telescope predicted to crash back to Earth tomorrow
NASA’s defunct Swift space telescope is expected to re-enter Earth’s atmosphere on November 4 (AEDT) after rescue attempts failed and it was shut down in late October. The uncontrolled 1.5-tonne observatory’s exact landing location and what will survive re-entry remain unknown.
Open-Source 5G Ran Platforms: A Dual Perspective on Performance and Capabilities
This paper presents a comparative evaluation of the open-source 5G RAN platforms OpenAirInterface (OAI) and srsRAN, addressing the lack of comprehensive comparisons across different SDRs, RAN configurations, user applications, and numbers of real UEs. It combines a qualitative assessment of supported features and deployment flexibility with quantitative analysis of radio resource control setup, compliance with theoretical data rates, and workloads including video-on-demand, live streaming, and cloud gaming.
Amazon cancels multiple Nvidia DGX Spark AI PC orders
Amazon canceled four orders for Nvidia DGX Spark AI PCs placed by AI educator Nick Saraev and closed his account after a verification process failed to resolve the issue. Epic Games CEO Tim Sweeney reported similar cancellations after trying to buy two Nvidia GPUs. Amazon has not explained the cancellations, though the moves may be aimed at curbing bulk buying and resale.
- Amazon cancels multiple Nvidia DGX Spark AI PC orders — cybernews.com
Six German states can now choose a sovereign cloud in schools
Schools in six German states—Berlin, Bremen, Hamburg, Hesse, North Rhine-Westphalia, and Rhineland-Palatinate—can now choose an IONOS-hosted sovereign cloud for file storage and collaboration tools under a Gov Digital contract tied to Germany’s Educational Cloud Infrastructure initiative. The platform keeps school data in German data centers to support digital sovereignty and reduce reliance on US cloud providers over privacy and lock-in concerns.
- Six German states can now choose a sovereign cloud in schools — cybernews.com
Drax plans £20B AI data center
Drax has signed an agreement with Brookfield and its Radiant platform to develop an AI data center on the Drax Power Station estate in North Yorkshire, initially targeting up to 100 MW by Q1 2028. The project plans an additional 500 MW between 2028 and 2031 and more than 1 GW eventually, with Brookfield estimating total investment, including semiconductors, could exceed £20 billion. The first phase still requires planning permission, government approval, a tenant, and final investment decisions.
- Drax, Brookfield, Radiant to build 100MW AI data center in North Yorkshire by Q1 2028 — cnbc.com
- Drax wants to feed Britain’s AI habit with £20B datacenter at biomass power station — theregister.com
Toyota shows off SoC that stores model weights and its OS on integrated memory
Toyota has unveiled an SoC that stores AI model weights and its operating system in integrated memory. The carmaker says this avoids the external-memory bottleneck that slows physical AI, offering an economical and speedy shortcut.
iPad mini (2026) review: great OLED, waterproof/dustproof, fast A20 Pro; pricey, no 120Hz/Face ID
The 2026 iPad Mini is the best Mini yet, adding a larger OLED display, waterproof and dustproof design, and a faster A20 Pro chip. However, it starts at $699, lacks a 120Hz display and Face ID, and can cost up to $1,249, making it hard to recommend despite its improvements.
- iPad mini (2026) review: great OLED, waterproof/dustproof, fast A20 Pro; pricey, no 120Hz/Face ID — theverge.com
Sources: Huawei’s flagship Ascend 960DT to largely use Samsung, SK Hynix HBM
Huawei has told some clients that a large share of its upcoming Ascend 960DT flagship will use HBM made by Samsung and SK Hynix. The disclosure highlights how U.S. restrictions on AI chip shipments to China can be circumvented, including through smuggling expensive memory chips from South Korea.
- Sources: Huawei’s flagship Ascend 960DT to largely use Samsung, SK Hynix HBM — theinformation.com
Designing a planned switchover for minimum disruption
Neki’s planned Postgres switchover minimizes disruption by separating the operation into preflight validation, convergence and promotion, and client recovery. Unlike unplanned failover, it does not require failure detection and can reject unsafe promotions during preflight checks. The article also explains how Neki establishes a final WAL position to safely promote a replica and what happens if the switchover cannot complete.
- Designing a planned switchover for minimum disruption — planetscale.com