Improvement curve for good people

From Issawiki
Revision as of 00:51, 17 July 2021 by Issa (talk | contribs)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to: navigation, search

Given that someone eventually becomes good, how steep is their improvement curve? Is it possible to distinguish such people from noise early on in their trajectory? This topic matters when thinking about "beating the market" on finding good people. It's also relevant when thinking about how often to check back on people you have previously dismissed.

Here are some reasons for pessimism at beating the market:

  • An interesting case is when someone suddenly and magically appears online in very good condition, without displaying any bad traits from back when they were still learning how to be good. In other words, they have zero output prior to good output. I think Carl Shulman is a good example of this phenomenon. He suddenly appears around 2007 in pretty much his present-day condition.
  • In other cases, people have a bunch of output that is indistinguishable from the output of bad people (or at least not especially good), which makes it hard to identify them early on. Jonathan Blow seems like an example here. Could one have spotted promise in The Inner Product or his early games/demos? Or did it take until Braid (which made him super famous anyway, and would mean it wasn't possible to beat the market by identifying him early)?

Given the above two cases, if you identify someone faster than when others identify them, it'll be because you can see the value of the good thing when others don't (e.g. EAs vs me when it comes to Wei Dai),[1] or because you expose yourself to more people so you catch the good ones sooner.

I feel like some people have a really abrupt transition from "indistinguishable from bad person" to "output of good person" without going through "output of bad but promising person who will become good in a few years". Why is this?

I should more systematically look through examples to see the output/improvement curves for various people. Some interesting examples to look at: Wei Dai, Eliezer, Buck, Carl, Abram, Jessica Taylor, Grognor.

References

What links here