3 Comments
Jul 15, 2022Liked by Étienne Fortier-Dubois

Instead of optimizing for a single variable, maybe aesthetics is a euphemism for Multi-Objective Optimization? Colin has a whole Substack about it. Also he noted on incomplete data as an issue along side bad modeling. https://desystemize.substack.com/p/desystemize-5

Expand full comment
author
Jul 15, 2022·edited Jul 15, 2022Author

Yeah that seems like a good way to put it! Although I'd say that optimizing for multiple things can also be a trap. Your list of things to optimize is never going to completely match the list of things you care about. So perhaps what I'm saying is, optimize to some extent, but then stop before you reach the zone where you might inadvertently sacrifice things that are in the second list but not the first. Follow the 80/20 rule and don't try to reach 100.

Expand full comment
Dec 15, 2022Liked by Étienne Fortier-Dubois

This reminds me of the whole Maximizers vs Satisficers debate, where it is hard to replicate "the 85% rule" for machines. https://www.nature.com/articles/s41467-019-12552-4 https://www.scotthyoung.com/blog/2022/07/05/85-percent-rule

The problem with this method is that there is no paired indicators that says which of the 85% should I try to aim for (assuming it is not something like RPEs and "training max" at the gym). https://www.businessinsider.com/use-85-rule-to-be-more-productive-successful-at-work-2020-11 https://coffeeandjunk.com/85-percent-rule/

Question: how would you draft a paired indicator for creative writing or learning about humanities/arts/philosophy/sociology?

Expand full comment