Probabilities Without Models

[Not original to Scott] Defends assigning probabilities even without a well-defined model. The alien-saucer President dialogue shows that refusing to estimate ('we have no model') is incoherent — any real decision (alert the military, not the banana plantations) reveals an implicit probability judgment. The NSF-grant example (safe Proposal A vs 1-in-1000 desalinization Proposal B) shows you can't escape estimating ~1/1000 probabilities, and 'Pascal's Mugging' doesn't apply when the improbable option recurs over your lifespan. Framing choices as probabilities lets you check calibration, compare people, and — crucially — call out overconfidence (e.g. 'one in a million chance of AI risk'). Clarifying and durable.

Quality 69 · Strong. Strong (69): a genuinely clarifying epistemics point (you can't opt out of implicit probabilities; making them explicit enables calibration) with vivid examples; held FIRM below Excellent as Scott flags it 'not original to me'.

Claude’s paradigm shift 48 · Moderate. Moderate (48): a sharp reframe of a debate about model-free probability, acknowledged as borrowed.

Real-world impact 3 · Moderate. Moderate (3): circulates in rationalist calibration/forecasting discussion; niche footprint.