Which facts are basic itself is a hard one for me to grokk, particularly in the political and historical sphere. Also, there is a problem of, I'm not sure this is the best word for it but, mereology? When you study cultures from more than one or two centuries ago, how do you point to a specific thing and say "this area/population is that culture" without a theoretical lens that smooths out the heterogeneity a ton? Because the further back you go, the longer distances get in terms of communication, and the less centralized control the government has over th... (read more)
I agree with the overall premise of the blog post, but I don't feel like I have gained strong object-level direction from it? The problem is that we always view facts from a particular lens, in other words, all facts are theory-laden to some degree. I don't want to commit a surface level trivial error like the Fallacy of Gray, I'm not saying all facts are equally theory-laden and it is meaningless to make the distinction the article makes, but the actual examples used in the article all feel strongly theory-laden to me. (Aside from perhaps the abortion one?)
I do enjoy the lens of "which configurations take", and largely use it as my primary point of view. Which is to say, what's interesting isn't which configurations are the most natural, but which configurations conflict the least with our baseline, and are the most stable as a result. The two aren't necessarily the same thing; I can think of highly artificial configurations, which have nothing to do with what evolutionary pressures we have undergone or how we used to live in the far past, that have nevertheless integrated into daily life quite seamlessly. Arguably being a frequent "blog content consumer" is one.
I'm not proposing that each company gets to pick the sense in which they pause. I'm proposing something like:
What are your thoughts on a sort of in-between approach: instead of pausing incrementally or abruptly+extensively, we go for an extensive pause which us not abrupt? I.e. we agree to do a sweeping pause, but with an allowance of several months for every AI lab to comply.
It feels like this would allow each company greater ability to do their own pacing in a way that best suits their own needs. I'm unsure whether that is a big enough upside to be worth it on net, but maybe?
I will personally admit that prima facie I found the rapist example to be the most surprising one in the entire enumeration (in a "wait, this inference is probably wrong but more reasonable than the other ones" sense) and now I need to introspect on that.