Interesting article, thanks. > In decision-making GPT-3 performs nearly on par with humans. In searching specific information or causal reasoning, however, the

Interesting article, thanks. > In decision-making GPT-3 performs nearly on par with humans. In searching specific information or causal reasoning, however, the artificial intelligence clearly falls behind. Except where it doesn't! Clearly the value differential here is calculated by some formal method. The GPT-3 value in causal reasoning can be described in a lot of ways, IMO. The Linda problem / conjunction fallacy is quite an interesting one. To me it raises this question of rationalized-probability, which can be modeled in a number of ways: - Which option is more probable _regarding people?_ - Which is more probable _regarding people with the described background?_ - Which is more probable _regarding bank tellers with the described background?_ - Which is more probable _regarding Linda as she may be pictured by a random population, given a contextual description which seems at odds with the cultural perception of a bank teller?_ - Which is more probable _regarding the question as it may be posed to you by any given person, in your personal, social experience?_ - Which is more probable _regarding the question as it may be posed to you by researchers, in your personal, social experience?_ To me the differential probability would likely decrease as particular contexts are identified in that kind of order, but I'm curious to know more about it.