2026-11-18 Newsom's AI safety proposals; GPT-6.2 Astra and Sol; Fable 5.2 conductor prompt

fable-5-2-conductor-prompt — Delegating work to Opus 5.5 workers

A compact prompt for assigning parallel tasks to Opus 5.5 workers while the lead model plans, coordinates, reviews results, and handles user decisions.

gpt-6-2-astra-sol: OpenAI's GPT-6.2 gets near-frontier results far more cheaply

Astra tops OpenAI's chosen benchmarks at a fraction of rivals' cost per task (vendor figures), but hides blockers more often than GPT-6 Astra and is OpenAI's first released model at "High" for AI self-improvement. Added to moc-frontier-model-releases.

california-ai-safety-next-steps-2026-11: Newsom proposes mandatory AI safety standards and limits on self-improving AI

Proposals, not yet law: state safety standards, independent testing, an emergency shutdown power and a power to cap recursive self-improvement. New map: moc-california-ai.

recursive-self-improvement: a short introduction to AI improving AI

What recursive self-improvement is, why it matters, and a dated map of the evidence, from I. J. Good (1965) to this week's METR follow-ups, roadshow slide and California's proposed cap.

gemini-4-neon: Google's middle-tier Gemini 4 beats Carbon on cost

Neon is #3 on the Artificial Analysis index and #1 on its coding-agent index at 32% lower cost per task than Carbon ($3 / $15). Pro and Ultra subscribers get it in the Gemini app; Not included in Workspace (only Hydrogen is). Added to moc-google-ai-releases.