I completely agree that the two perspectives can be integrated (I even spend a paragraph discussing how, the one starting with “I also continue to think there is a place for micro-level skills...”). However, it is possible to micro-optimize successfully and still lose, but this is much less true of macro-optimization, so I actually do think it makes sense to present one as better than the other.
(There is a separate danger in macro-optimization, which is that it is easier to deceive yourself into thinking that a bad or neutral optimization is actually a good one. For classical examples, take going on diet X, or training yourself to sleep for only 5 hours each night.)
I think it’s about more than the scope of the decision problem. The techniques you use to optimize at the macro-level are fundamentally different and I really do mean in many cases “change the structure of the search space”. For many people, reading the Sequences had such an effect; or going to college; for me one of the most salient recent examples was working at Dropbox for the summer, which completely changed the way that I approached writing code. I didn’t decide to work there due to the output of any explicit optimization, though. It was “this has many characteristics in common with the sort of thing that will end up changing my life; so, even though I can’t give any specific mechanism as to how it will actually do that, I’m going to work there anyways”.
Re: Toyota; the difference between Toyota and a human is that Toyota makes millions of cars each year, so even saving 1 penny on each car manufactured is worth tens of thousands of dollars to the company.
The techniques you use to optimize at the macro-level are fundamentally different
It’s not clear to me that this is the case. The parts of the problem that are hardest change, but the problems are fundamentally similar in a deep way. I will agree that the training to improve macro-level optimizations and the training to improve micro-level optimizations have different focuses, and am working on a post that will be relevant to the former.
I completely agree that the two perspectives can be integrated (I even spend a paragraph discussing how, the one starting with “I also continue to think there is a place for micro-level skills...”). However, it is possible to micro-optimize successfully and still lose, but this is much less true of macro-optimization, so I actually do think it makes sense to present one as better than the other.
(There is a separate danger in macro-optimization, which is that it is easier to deceive yourself into thinking that a bad or neutral optimization is actually a good one. For classical examples, take going on diet X, or training yourself to sleep for only 5 hours each night.)
I think it’s about more than the scope of the decision problem. The techniques you use to optimize at the macro-level are fundamentally different and I really do mean in many cases “change the structure of the search space”. For many people, reading the Sequences had such an effect; or going to college; for me one of the most salient recent examples was working at Dropbox for the summer, which completely changed the way that I approached writing code. I didn’t decide to work there due to the output of any explicit optimization, though. It was “this has many characteristics in common with the sort of thing that will end up changing my life; so, even though I can’t give any specific mechanism as to how it will actually do that, I’m going to work there anyways”.
Re: Toyota; the difference between Toyota and a human is that Toyota makes millions of cars each year, so even saving 1 penny on each car manufactured is worth tens of thousands of dollars to the company.
I think it’s pretty possible to macro-optimize successfully and still lose. All you have to do is know what to do and not how to do it.
It’s not clear to me that this is the case. The parts of the problem that are hardest change, but the problems are fundamentally similar in a deep way. I will agree that the training to improve macro-level optimizations and the training to improve micro-level optimizations have different focuses, and am working on a post that will be relevant to the former.