I like the way you tie real-world advice to principles in ML and RL. In general I think there are a lot of risks to naively applying epistemic deference and worldview aggregation and you articulate some really nicely here.
Something I’ve noticed with a few of your posts is that they often contain a lot of nuggets of ideas! And for you they seem to cohere into maybe a single high-level thought, but I sometimes want to pull them into smaller chunks[1]. For example, I imagine you (or others) might want to refer individually to the core idea in the paragraph beginning
However, even if in practice we end up mostly evaluating worldviews based on their epistemic track record, I claim that it’s still valuable to consider the epistemic track record as a proxy for the quality of their advice, rather than using it directly to evaluate how much we trust each worldview...
Now, the rest of the post gives this core idea context and support, but I think it stands on its own as well.
One compromise :D between putting lots of ideas together and splitting them apart too atomically could be to add meaningful sub-headings. (This also incidentally makes it easy to link out to the specific part of the text from another place via # links.)
I like the way you tie real-world advice to principles in ML and RL. In general I think there are a lot of risks to naively applying epistemic deference and worldview aggregation and you articulate some really nicely here.
Something I’ve noticed with a few of your posts is that they often contain a lot of nuggets of ideas! And for you they seem to cohere into maybe a single high-level thought, but I sometimes want to pull them into smaller chunks[1]. For example, I imagine you (or others) might want to refer individually to the core idea in the paragraph beginning
Now, the rest of the post gives this core idea context and support, but I think it stands on its own as well.
One compromise :D between putting lots of ideas together and splitting them apart too atomically could be to add meaningful sub-headings. (This also incidentally makes it easy to link out to the specific part of the text from another place via
#links.)Maybe we differ in the number of effective working memory slots we have available (for what I mean see https://www.sciencedaily.com/releases/2008/04/080402212855.htm though see https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4159388/ which challenges this)
Just wanted to note that this comment was quite helpful for me, and has influenced other blog posts that I’m writing. Thanks!