niplav @niplav@schelling.pt

Dafür, daß sie Jainisten heißen, sind sie aber ganz schön resolut

licerology: the study of betting and auctions, auction theory

short story about a cancer researcher who starts murdering patients in the control group because he expects there to be a null result, but he always puts on a mask that prevents him from directly seeing the person he kills—double blind

doing pickup has definitely wambified me a bit

not a lot, but it's noticeable

impromptu explanation of coq by introduction in a deep codebase

i missed hackerspace

favourite latin word of the day: "biduum", n

A period of two days

damn dude you look disembodied as fuck

hot em

most tulpas turn out as very slight variations of the original person

why would you even bother, you're both the same depressed anime-picture person who is also slightly quirky

the only relief i currently have is that the number of not-completely-autistic hot women in the bayrat/ea communities is very low, so we can't be that deep into social bullshit yet

“We are trying to accomplish an extremely difficult thing that requires everything to run as smoothly and efficiently as possible, but luckily our status incentives are so optimal that the high status men getting to sleep with a lot women is just the best!”

that is to say, both might be relevant for personal effectiveness, and both are tangentially related to big problems (how do we bring about good states of consciousness & how does society organise sexuality so that everything works ~well), but not much beyond that.

i think posting about personal meditation practice on the EA forum is *bad*, because personal meditation practice is as relevant to EA as advice on seducing people.

cut your losses. slash their bodies wide open

“What, of course I want to spend my entire life on this very specific spot on the happiness/productivity Pareto frontier, dummy.”

This should be called "pathologizing a graph".

While this wouldn't tell us *how* we should make preferences consistent, it could give us a way of determining which ways of making preferences consistent exist and are more complex (maybe penalizing deleting edges differently from merging or adding edges?).

I should try to find out whether this has been done before, and if it hasn't, I really want to look into this more.

Example: If we have a cycle graph ({a,b,c},{(a,b),(b,c),(c,a)}), then we can turn it into a path graph by removing any of the three edges (three ways of pathizing the graph).

If we have a DAG ({a,b,c,d},{(a,b),(a,c),(b,d),(c,d)}) where b and c are deemed incomparable, we can turn that into a path graph in two different ways: either putting c before b or b before c (two ways of pathizing the graph).

There's probably some recursive thing going on here, if we have more complex graphs.

I wonder whether there is mathematics about the combinatorics of turning directed graphs into DAGs and DAGs into path graphs (maybe with the intermediate step of trees?).

This feels kind of relevant to AI alignment: If we have the right level of abstraction (i.e. we don't just overfit to human behavior and biases), one way we can represent inconsistent preferences is via directed graphs (fully connected?). But we would like to turn these inconsistent preferences into consistent ones.

@nyx in general, if you don't know this site yet, my model of you predicts you would enjoy it: https://traditionsofconflict.com/

"Until you have used a sieve to find the finger bones of a newborn baby in ploughsoil it is hard to explain how small they are"
– Timothy Taylor, “The Buried Soul — How Humans Invented Death”, 2002.

@nyx had to think of u