Against Learning From Dramatic Events

Argues you should mostly not update beliefs on single dramatic events -- 9/11, mass shootings, lab-leak pandemics, harassment scandals, FTX, the OpenAI board -- but instead predict distributions in advance (power-law style) and make only the small Bayesian update one data point warrants; the yo-yo of maximal opposite-direction overreactions (FTX lessons vs OpenAI-board lessons) is the failure mode. The load-bearing move is the closing distinction: dramatic events are epistemically near-worthless but coordination-critical -- they act as Schelling points / common-knowledge generators / hyperstitional cascades (#MeToo, the War on Terror) -- so it is fine to exploit them to coordinate as long as you are not yourself epistemically driven by them. Wide-ranging, vivid, genuinely generative.

Quality 77 · Excellent. Excellent floor. A memorable, portable principle ('predict distributions, don't update on the dramatic instance') joined to a real synthetic payoff -- the epistemic-worthless / coordination-critical split that explains the entire dramatic-event news economy. Crosses into Excellent on generativeness and reach; kept at the floor because the Bayesian core is standard and the structure is somewhat enumerative.

Claude’s paradigm shift 58 · Moderate. Notable. The updating discipline and common-knowledge machinery are pre-existing (Scott has used them before), but reframing dramatic-event news cycles as a coordination/Schelling-point phenomenon distinct from their epistemic content is a fresh, non-obvious synthesis for informed readers.

Real-world impact 3 · Moderate. Moderate. An influential, frequently-cited essay within rationalist/EA discourse whose frame ('don't learn from dramatic events') recurs as a reference; influence stays inside that subculture.