Florian Cucurella

Daily reading — 2026-10-23

  • Quad9 plans to withdraw service from France on November 24 after a DNS-blocking ruling, arguing that jurisdictional filtering conflicts with its privacy model.

  • eMudhra announces its exit from global SSL/TLS issuance and promises certificate replacement and continued revocation services during the transition.

  • Sammy Hajhamid explains a tethered Switch 2 jailbreak and separates AI-assisted findings from the substantial manual work needed to complete it.

  • Pete Cooper apologizes for publishing advance PHP vulnerability information, saying he had been told there was no embargo.

  • Code and Cake uses a small container-runtime project to explain how namespaces, resource limits and process setup combine beneath Docker.

  • HackiOS describes running iOS 16.5 userspace over a modified XNU kernel on a POCO phone and QEMU, with major phone services still unsupported.

  • ModelCompare reports that its small prompt experiment elicited similar claims about human conformity from twelve text models, with several exceptions.

  • NVIDIA OpenShell combines sandbox enforcement, endpoint-bound credentials and formal checks on proposed policy changes for autonomous agents.

  • Anthropic and Theorem report machine-checked no-crash proofs for 239 fuzzing entry points across 123 libraries, while spelling out the limits of their specifications and RISC-V/Linux model.

  • TechSpot reports scrutiny of forest clearance for Google’s Finnish data-center project; Google says it complied with local law.

  • Local Wikipedia builds a small encyclopedia from nearby geocoded Wikipedia pages, with the chosen location kept in the browser.

  • GNU Taler plans a dedicated German operator for euro payments, with operations contingent on obtaining an e-money licence.

  • The Browser Company schedules Arc’s desktop shutdown for December 22 and describes export and migration options before accounts are deactivated.

  • Amazon Threat Intelligence reports PHP2Shell attempts in honeypot telemetry before public disclosure, while noting uncertainty about their success against real systems.

  • Phoronix reports that several unpatched Linux kernel vulnerability reports were sent to public mailing lists, apparently by mistake.

  • Evgen Bodunov walks through measured triangulation improvements in earstrip, retaining Delaunay fallbacks for difficult map geometry.

  • Project Hedra documents an open-world RPG and custom OpenGL engine that grew from its author’s high-school learning project.

  • Crystal 1.22 adds typed constants, native event loops across targets and improved DWARF parsing, and promotes aarch64 Linux to Tier 1.

  • Edsger Dijkstra’s 1978 essay argues that formal notation reduces ambiguity and that natural-language interfaces can increase the work of communication.

  • Matthias Endler advocates small, sustainable software businesses and deliberate execution over growth pursued at users’ expense.

  • The Last Museum offers semantic search across millions of artworks, with artist, culture and date exploration.

  • Matthew Cole’s Vanity Fair excerpt alleges secret foreign-funded accounts tied to Jared Kushner’s diplomacy; it includes denials from Kushner’s counsel and former ambassador David Friedman.

  • Aamna Mohdin reports that Margaret Atwood, Zadie Smith and Ian McEwan are offering character-naming rights in an auction supporting torture survivors, alongside readings by members of Freedom from Torture’s writing group. The Guardian

  • Adam Bonica and Jake Grumbach argue that Democratic campaign consultants’ fees reward spending rather than results; their analysis finds no relationship between firms’ electoral performance and subsequent revenue growth. The New Republic

  • Matthias Bastian reports that Demis Hassabis now estimates AGI within two to three years, using the ability to recreate Einstein’s 1905 work from earlier knowledge as a proposed test. The Decoder

  • The developer of Expense Tracker describes an offline Android app that captures Indian bank SMS and UPI notifications, categorizes transactions on-device and declares no internet permission. Repository

  • MiniMax describes M3.1-Flash as a multimodal mixture-of-experts model using sparse attention, quantization-aware training and speculative decoding, with a context window of up to one million tokens. Model card