2025-08-10
decadent stress chamber
Via Marginal Revolution, David Sacks is calling GPT-5 as proof that AI progress is sigmoidal. This is also my prediction, but I feel like it’s premature to declare victory so early. Depending on to what extent we can figure something out for online or self-learning, there are three major categories of outcomes:
Outside a few domains with verifiable outputs, at the level of the best human.
Superhuman such that humans are rendered economically unviable.
Exponential self-improvement leads to the AI god.
I wonder to what extent this realization is leading more people to start considering how AI alignment is likely to actually be a mutual effort, which includes humans changing their behavior as opposed to unilaterally constraining the behavior of AIs. Nick Bostrom has an interesting paper which generalizes this idea to cosmic universal citizenship.
Nix comments on the FT article that has been making the rounds, showing increasing neuroticism and declining rates of both conscientiousness and extroversion across generations. It’s interesting that extroversion is declining so much, since it seems clear to me that the primary factor currently for inclusive fitness is one’s level of extroversion. This is related to my theory on what humans becoming economically irrelevant means: it implies that there is a superior non-human alternative to any task you might want another human to do for you. That we are making less effort to become palatable to other humans implies that this is already happening to some extent; it’s because of technological progress that we rely less on each other for survival, and can ignore coalition building and status chasing in lieu of self-expression and exploration of personal interests, even if that might cause offense or result in atomization and isolation.
Human Progress linkthread. Includes an interesting graph on rising levels of sport participation across all ages.
Nintil linkthread
