The Sigmoids Won't Save You

A short, sharp rebuttal to the lazy AI-skeptic talking point that 'all exponentials eventually become sigmoids.' True - epidemics and airspeed records do flatten - but the 'Sigmoid Misidentification Hall of Fame' (UN birthrate projections, solar-deployment forecasts, a Wharton paper on the METR AI-capabilities curve) shows forecasters reliably predict the flattening at the exact moment of their analysis, and reliably get burned. The constructive answer: if you understand the generating process, model it (epidemics, ramjet limits); under true ignorance, the default should be Lindy's Law - a process continues on average as long as it already has (the geyser intuition). Applied to AI: scaling has run ~7 years, so naively expect ~7 more, and the burden is on sigmoid-claimers to either show their model or explain why not Lindy.

Quality 72 · Strong. A sharp, memorable, and genuinely useful demolition of a common talking point, pairing a vivid catalogue of failed sigmoid calls with a clean constructive default (Lindy's Law); short and built on existing concepts, which sets its level. Strong.

Claude’s paradigm shift 45 · Moderate. Applies Lindy's Law and base-rate reasoning to trend-extrapolation; the 'Hall of Fame' and the Lindy-as-default framing are a fresh, useful packaging rather than a new idea. Moderate.

Real-world impact 2 · Minor. A sharp, memorable demolition of the lazy 'all exponentials eventually become sigmoids' AI-skeptic talking point (the 'Sigmoid Misidentification Hall of Fame') with a clean constructive default (model the process, else Lindy). Conceptual influence within AI-forecasting discourse, short, no material change — low RWI.