Confidence levels inside and outside an argument

Distinguishes a model's INTERNAL confidence from your EXTERNAL confidence in its conclusion. If a model says the incumbent wins with probability 999,999,999/billion, your actual confidence should be far lower, because the background chance that the model/argument/coder/typist is simply flawed vastly exceeds one in a billion — and any flaw drags you back toward the 50% prior, not symmetrically. Distinguishes genuine high-confidence cases (lotteries, where flaws don't favor any player) from argument-based ones. Vivid examples: the LHC cosmic-ray safety argument (1/10^25 internal, but the argument has a real flaw re black-hole momentum), and the creationist 'probability of a cell forming by chance is 1 in 10^4,478,296' — a textbook confusion of internal with external probability, and 'possibly the most wrong anyone has ever been.' Extends to never assigning 100% even to proved theorems (chance you erred).

Quality 80 · Excellent. Low-Excellent. A foundational calibration lesson — cap your confidence in any conclusion by the probability the argument itself is unsound — argued rigorously with unforgettable examples. Widely cited; co-tier with the LW concept-coiners. 80.

Claude’s paradigm shift 60 · Notable shift. Notable shift. The inside/outside-an-argument framing of confidence was a sharp, novel 2010 formulation (building on 'infinite certainty' / 0-and-1-are-not-probabilities) that became a standard calibration concept. 60.

Real-world impact 3 · Moderate. A foundational calibration lesson (cap your confidence in any conclusion by the chance the argument itself is unsound), widely cited within rationalist discourse. Broad conceptual uptake, but no material or population-level change — modest RWI.