Oct 18, 2025
There is a thing people do where they care a lot about getting a better estimate of a prior, even though the prior doesn’t actually matter that much. The archetypical example of this is something like fitness bros combing through the research to figure out which curl variation grows your biceps the most. Sure, on average variant A is probably a bit better than variant B, and this is useful to know if you’re going to make a recommendation on which exercise to choose to somebody you don’t know. But this average difference is probably not huge, and individual variance is big enough that after you’ve tried both variance and gotten a sense for how good they feel and what kind of pump they give you and how sore they leave you, you know which one’s better for you regardless of which one the studies said is 10% better.
Another area where I see this is in the gender wars, or whatever you want to call this absolutely radioactive thread of online discourse. I see a million trillion tweets about how men are more x and about how the true nature of woman is y, and somehow this feels important enough to generate lots of Takes, but again, the Bayesian impact of whether men are on average more x on y is, in the matters actually relevant to your life, dominated completely the particulars. If the men you talk with in your life are catholic farmers from the south of Poland, or French math students from your university in Marseilles, or Silicon Valley enlightenment junkies with rationalist tendencies, you have a prior so superior in richness and detail that the abstract broad and generalizing descriptions people argue so much about in the discourse are irrelevant. And even more so after you’ve talked to a person and gotten to know them at least a little!
All this is not to say that these group differences don’t exist or must necessarily be small, not at all. But reality is fractally complex and very information dense and general high level priors must be weak to accommodate this, so don’t worry about them so much.