autonobahn.ai

GPT4 is repeating 2nd hand material abt "Statistical Consequences of Fat Tails" w/o knowing what the book is about. It is not about Fat Tails but CONSEQUENCES on stat methods, the law of large (medium) numbers, & decisions under uncertainty. Also makes economists' elementary mistake (explained in SCOFT): rare event are NOT more frequent but more CONSEQUENTIAL in determining the total properties. Events >1 STD are LESS frequent than in Gaussian. Extreme fat tails (Cauchy): 1 SINGLE event ends up having all the properties! It almost has the understanding of, say @Ptetlock! It misses that Fat Tails make usual metrics like correlation, STD, portfolio theory, etc. unusable, particularly under ellipticality.

AI will reach human levels by 2029. By 2045, we will have multiplied the intelligence, the human biological machine intelligence of our civilization a billion-fold. Follow me for the latest AI discussions, tools & more. 👇

That one time @deadmau5 said a house track bassline was impossible to play live and @Davie504bass said lol watch me is perfect Friday vibes honestly

This is GPT-4's estimated performance on various standardized tests, and comparison to the previous generation. It's very likely that GPT-5 or its equivalent will be released by the end of this year. It's so over.