Davidson On Takeoff Speeds

An explainer of Tom Davidson's Compute-Centric Framework for AI takeoff speeds, opened with the Everest-skiing framing: progress can be gradual, continuous, AND terrifyingly fast (Davidson predicts ~3 years from AI doing 20% of jobs to 100%, superintelligence within a year after). Walks the model: it updates Bio Anchors with AI-helps-AI-research feedback loops; the three key factors (the effective-FLOP gap 20%->100%, densest ~4 OOMs; investor 'wakeup time'; and bottlenecks via the CES parameter rho -- the Rome-building-a-stealth-bomber debate about whether cognitive labor can route around physical/compute bottlenecks); and the sensitivity analysis showing the intelligence explosion is robust unless you set parallelization/bottleneck params to values that don't match real economics. Plus the honest 'every AI timeline is rigged to match current vibes' critique and the OpenPhil-vs-MIRI (gradual/multipolar vs sudden/singular) framing.

Quality 75 · Excellent. Strong (upper): a lucid, well-organized explainer that makes a dense 60-parameter model legible, with the portable Everest-takeoff framing and the honest vibes/sensitivity critiques; strong, though relaying Davidson's model.

Claude’s paradigm shift 50 · Moderate. Notable: the exposition plus the 'gradual can be faster than fast', the vibes critique, and the robust-intelligence-explosion sensitivity finding are real contributions.

Real-world impact 2 · Minor. Minor/within-discourse: an AI-timelines explainer; no direct material footprint.