my best guess is some people find it really hard to back down and are also smart enough to argue for anything, and so they generate any claim that fits the context
leogao
Karma: 13,154
there are some people i talk to where any disagreement rapidly leads to establishing a common ground and then converging to some crux. there are other people where the more questions i ask, the more it feels like i’m unraveling a huge tangled ball of yarn where every claim i dig into uncovers even more bizarre claims
fyodor pavlovich from the brothers karamazov reminds me of the modern postironic internet troll
my hot take is circuit sparsity is still the best way to get natural features. by making it expensive to read from and write to many residual channels, you encourage frequently used concepts to be stored in only a few residual channels.
sometimes, there exists a good reason for doing something that seems silly from first principles, but the people doing it are incapable of articulating that reason; they might not even be aware that this is the reason. for example, some people might be conservative because they feel a strong disgust for things they don’t understand, and a fear of change. this is a completely irrational reason, but there also exists a more reasonable argument for conservatism, which is someform of chesterton’s fence / wariness of unforeseen consequence. because of natural selection or cultural selection, people can be selected to have traits that are good for their survival, even if their reasons for exhibiting those traits are totally irrational. so it’s worth being cautious about overly eager dismissal of things just because the articulated reasoning seems very irrational.
of course, on the flip side some things are just bad and maladaptive to the current world. some seemingly irrational things truly are just irrational.
a non exhaustive list of ideas i’d love to see microgrant applications for:
making the one true glorious interpretability (or alignment) metric so that when we have the number go up machine we can use it to solve interp (or alignment) (best so far is the ARC low probability estimation metric; but can we do better?)
tech that enables verification of the glorious international treaty; policy advocacy for it
fully interpreting the algzoo models, possibly using circuit sparsity techniques
improving democratic decision making via sortition
studying generalization in humans. when do humans generalize from some task to some other task, and how much?
better rationality techniques/frames. eg emotionally intelligent frames of rationality.
making prediction markets better, especially for AI questions
making really good ai safety educational resources on youtube/tiktok (being the next rob miles)
biosafety—eg proving safety of far UVC or making it cheaper, wastewater pathogen monitoring, etc
inoculating society against superpersuasion
improving mental health of alignment researchers
crazy that waluigi is a pun on the japanese word warui (evil).
i would love to fund a microgrant for someone to actually run an experiment like this
the learning curve for contact lenses is crazy. the first time i put them in and took them out, it took literally hours. on the 10th time, it took maybe 10 minutes. on the 100th time, closer to 30 seconds. i’m nearing 1000 times now and on a good day it can take only 5 seconds.
has someone run the experiment after the assertion?
this argument only justifies a partial order over hypotheses, where some hypotheses are strictly dominated by others.
a famous resulf in psychology is that if you tell a class of photography students to produce one extremely high quality photo, they actually produce worse photos than another class told only to produce a lot of decent photos, because in making lots of decent photos you learn way more about photography through trial and error, whereas the high quality photo group tends to spend irrationally much time making each try as perfect as possible.
does this effect actually replicate, especially across domains?
already did
on many forums, this is actively strongly discouraged (“necroing”)
i don’t have a go-to article, but some examples include: the train station ticket machines, hotel booking websites, the old “smart” mobile phones which were both extremely ahead of their time and very strange, many of the touchscreen vending machines have horrible UI, etc.
it has no rats
i spent a huge amount of time scrolling the internet (random curiosity directed consump of internet content) in the 2010s. in retrospect, i don’t even regret a lot of it, since it played an important role in bringing me into internet culture. but at some point the value of scrolling declined substantially; it’s hard to pinpoint when it changed. i wonder what happened.
actually, we played god many times in the past with positive results. for example:
eradicating smallpox
synthetic fertilizer and high yield crops
vaccines
water sanitation
antibiotics
surgery
genetic engineering
of course, sometimes it goes poorly too. but when it goes well it goes really well; I’m exceedingly grateful that i’m unlikely to ever die from smallpox or cholera or tetanus or appendicitis or starvation. let’s work on playing god well.
the train station fare gates are especially retrofuture. for some journeys, you may need multiple tickets simultaneously, so you feed a stack of tickets into the fare gate, and it automatically separates them, reads them (the tickets are printed on special magnetic stripe paper), stamps them, and returns them to you in a neat stack on the other side, all in under a second. also the ticket machines have some of the best coin funnels i’ve ever seen on any coin operated device (because clearly you’re going to use your bag of 500 yen coins to buy a 100,000 yen shinkansen ticket, why else?)
the ticket machines have the most confusing interface imaginable. thankfully there is a help button that video-calls in a customer service person who remote-desktops in to help you navigate the interface or even click on buttons for you. there’s even a little tray with a camera looking at it, for you to show your existing tickets to the support person, rather than having to hold them up next to your face and praying that they can read it.
it’s all so well thought out and completely outdated at the same time.
another hallmark of people who rapidly converge to some crux is they are much more willing to grant simplifying assumptions that don’t change the core disagreement, whereas people who don’t converge like to keep their options open