All notes in the notebook.
A postmortem of NightLoop: why per-task scoped context degrades frontier coding agents, and why governance should verify outcomes instead of constraining what the model sees.
The US government ordered Anthropic to cut off Fable 5 and Mythos 5 access for all foreign nationals on national-security grounds. Strip the outrage and a real precedent remains: frontier models are being treated like export-controlled weapons — and we've run this experiment before.
A reported campaign hides payloads from AI-powered security scanners by injecting weapons-related text that trips the scanner's own safety refusals. Single-source so far — but the technique class is worth understanding now.
A developer-and-LLM benchmark of AMD's new Ryzen AI mini PC against Apple's M4 Pro Mac Mini lands at a more interesting place than 'who wins' — it's about price-per-capability.