598 Comments
⭠ Return to thread

Another thing that can doom predictions is predicting the upside on things that take the stairs up and the elevator down. This is most notable when predicting market recessions - If anyone was dumb enough to predict there's no recession coming up, , they'd get buried in the news every time the market has a bad day (somehow people became aware of this, which is why people pretty much always predict a recession in the near future, at least since 2008).

This also applies to AI development: A month with nothing exciting happening should probably push your timelines back by about a month or two on average, but a single exciting event would make it jump down, which means that predicting "AI is coming faster than you think, no matter how fast you think" is great media - every time people adjust timelines down you get to be all smug, but adjusting timelines up is gradual and boring and no one talks about you when that happens.

(It feels like this is Eliezer's shtick this day, which is why people are right to be annoyed at him for refusing to make any concrete timeline predictions).

Expand full comment