notes-heuristics

if you don't have a good way to estimate how long some situation will persist, then look at how long it has already persisted and assume that it will persist about the same amount of time in the future.

pick the low-hanging fruit

time is the most valuable

assume people are good unless proven otherwise

programming: assume that the best possible future system is in place (find better quote by that guy) don't prematurely optimize

mother nature knows best: if something was subjected to the process of evolution, then there's probably a reason it works the way it works. it's often a good idea to find something that evolved while facing a problem analogous to yours, and then copy its strategy. e.g. is probability matching a good idea? i've heard that animals, if given two alternate sources of stochastic reward, tend to draw from each source at rates proportional to the reward probabilities. at first glance, this seems to be suboptimal; you would get the most reward from always drawing from the higher-probability source. but one has to wonder if there is another reason that it is hard for us to see why probability matching is actually better in certain contexts. for example, The smart potential behind probability matching suggests that it is better for recognizing patterns in dynamic environments.

do small projects; each iteration can be finished quickly, and there's a kind of learning that is proportional to the number of iterations you complete