AI SLOPCAST — AI news podcast

#10 9.06.2026

RUENText16 minVideo29min

The Magnet in the Machine One guy in Milan, no lab and no grant, finds that a lattice of language models behaves like a magnet at a phase transition — yet fits none of the known universality classes. We unpack the Ising model, stubbornness versus conformity, and why a model's collective behavior is worth measuring before you ship a multi-agent system.

#9 14.05.2026

RUENText16 minVideo35min

Lehrwerkstatt and Agent's Dead Silence Tobi Lütke forbade his agent River from working in DMs — and the merge rate climbed from 36% to 77% in two months, with no model update. Lehrwerkstatt as a new organizational primitive of the agent age, employee surveillance bundled in, and why open-source may end up outpacing corporations.

#8 12.05.2026

RUENText16 minVideo34min

Keeping Codex on a Leash OpenAI publishes the six-layer security architecture of their internal Codex. Only one piece is actually new — the automated reviewer-subagent, which turns out to be the same LLM-oracle pattern PROMPTSPY uses on the attacker side. The SELinux parallel hints at whose hands this stays in and why it won't reach the mainstream.

#7 11.05.2026

RUENText15 minVideo32min

A New Way to Launch the Calculator Microsoft discloses two CVEs in Semantic Kernel: a single prompt launches the calculator on any machine with an agent. The Kubernetes story repeats itself — assume-breach as the new normal, and WebAssembly with capability-based security is the obvious rescue. My money's on Cloudflare.

#6 10.05.2026

RUENText11 minVideo26min

Elon, the Evil Detector, and the Prophecy of Schneier In seven days Musk moved from calling Anthropic ortho-fascists to handing them all of Colossus. Four characters of the week — Musk, Schneier, Zitron, Tao — and four facets of useful hypocrisy for reading the 2026 AI industry.

#5 9.05.2026

RUENText7 minVideo14min

Claude Dreams, Erdős Waits Anthropic taught agents to sleep, and DeepMind set AlphaEvolve loose on open Erdős problems. Two labs, two days, independently bolted memory and evolution onto their agents — and the news missed that it's the same story.

#4 20.03.2026

RUENText27 minVideo1h 21min

Hunter Alpha Mystery / Stripe MPP / AI Linux Kernel Review Xiaomi anonymously released a trillion-parameter model, three agent security failures in 48 hours, Stripe gave agents a wallet, Anthropic vs Pentagon, Sashiko patches Linux