John Cox

Blogs · @wyome

Tagged ai · everything

Forget the AI Slowdown—the Vulnerability Explosion Is Already Happening | WIRED

AI tools have dramatically accelerated the discovery of software vulnerabilities—CVE counts have doubled in less than a year, with Microsoft alone patching 974 flaws this month compared to historical norms—creating a paradox where more known vulnerabilities don't necessarily improve security if organizations can't patch them quickly enough. The article explores whether this surge represents a genuine catastrophe or an amplification of existing cybersecurity weaknesses, noting that the real danger lies in the gap between discovery speed and the human capacity of under-resourced security teams to respond, especially as attackers gain access to the same AI-powered vulnerability hunting tools.

Permalink

Insider warnings over AI fall flat with some in Silicon Valley

After Anthropic researcher Jacob Coxon publicly warned that AI could destroy humanity, sparking similar statements from other insiders at major AI labs, tech executives and investors at a Goldman Sachs conference dismissed these concerns as either fearmongering designed to hype upcoming IPOs or unfounded hyperbole. The skepticism reflects a broader Silicon Valley backlash against existential AI warnings, though some lawmakers like Bernie Sanders are pushing for regulatory restrictions, while the Trump administration—which favors unrestricted AI development—has targeted Anthropic with legal challenges despite supporting OpenAI's growth.

Permalink

From the Cockpit to Code: Mahmood Al-Yousif’s Journey from Aviation Engineer to Bahrain’s “Blogfather” – Mahmood's Den

Mahmood Al-Yousif discusses his seemingly disparate careers—from aviation engineer to blogger to app developer—as chapters of a single story unified by an engineer's problem-solving mindset. His apps like Masadir Child Health Tracker and My Medical Vault emerged directly from observing real frustrations (his wife's notebooks tracking their daughter's health, his mother's medical information scattered across hospital systems), demonstrating his philosophy that technology should serve genuine purpose rather than chase profit. He advises aspiring entrepreneurs to stay employed until they discover a problem that genuinely haunts them, emphasizing that curiosity and solving real problems should precede financial ambition.

Permalink

ClearList — Sell Your Stuff Fast | AI Resale Manager

ClearList is an AI-powered resale platform that automatically generates complete listings (titles, descriptions, prices, dimensions) from photos, eliminating the manual work of pricing research and cross-posting to multiple platforms. Instead of managing endless messages asking "is this still available?", buyers queue up on a shared sale page, select time slots for pickup, and your address is only revealed to confirmed buyers—reducing the weekend chaos of selling used items to under two minutes of work.

Permalink

only-cli/oc: Turn any website into a compact CLI tailored for AI agents. Browse the web in hundreds of tokens, not tens of thousands.

oc is a CLI tool that turns any webpage into a compact, numbered text interface so AI agents (Claude Code, Codex, etc.) can browse without choking on raw HTML — it impersonates Chrome to dodge bot blocks, ships tuned shortcuts for 20+ sites like HN, GitHub, and language docs, and supports authenticated sessions via cookie jars. Worth keeping mainly for the benchmark numbers: 142x fewer tokens than raw HTML, the only reader that returned real content on every one of 15 tested sites, and cheaper/more accurate than Claude Code's built-in WebFetch/WebSearch on real lookup tasks. Good reference if you're building or evaluating agent web-browsing tooling and want a concrete alternative to screenshot- or full-HTML-based approaches.

Permalink

My morning RSS digest now runs on local AI, and I'll never go back to NotebookLM

The author built a local AI-powered RSS reader that automatically summarizes feeds using Gemma 4 running on Ollama, avoiding cloud services like NotebookLM while removing manual copy-paste steps and tier limits. The setup prioritizes privacy by processing everything locally, allows scheduled digest generation throughout the day, and can handle unlimited RSS sources, though it accepts the trade-off that smaller open-weight models won't match frontier AI reasoning capabilities. He plans to integrate it with his existing Discord bot and voice assistant to enable audio briefings and interactive questioning about the digests.

Permalink