Skip to main content



I really like how smooth and clean this retention curve is - this is for my episode with Evan Hubinger, the height of the line is what fraction of viewers are still watching at any given time.


TIL that an experience that I've had ~once every month or so for my whole life, and assumed was near-universal, is actually relatively rare, and correlated with various bad things that I'm not aware of experiencing in relation to it (EBV infection, migraines, head trauma).

Basically, as I experience it (typically right as I'm falling asleep) everything visually starts to feel very small and far away, except that my tongue feels large and cumbersome in my mouth.

It's called Alice in Wonderland Syndrome; other people experience similar size distortions though the details vary a lot.

in reply to Ben Weinstein-Raun

I don't know if we have discussed this . . . but me, too. So maybe it was passed down.
in reply to Ben Weinstein-Raun

I have also experienced this!! Rarely, but enough for me to have noticed the pattern.


Mercer Girls


There are a bunch of places around Seattle named Mercer: Mercer Street, Mercer Island, Mercer Slough. We were walking in Mercer Slough today and I was trying to explain about Asa Mercer (one of the historical figures all this stuff is named after). I looked him up later. He looked like this.

One of his big claims to fame is that he brought the Mercer Girls to Seattle. Saith wikipedia:

The Mercer Girls or Mercer Maids were women who chose to move from the east coast of the United States to the Seattle area in the 1860s at the invitation of Asa Mercer. Mercer, an American who lived in Seattle, wanted to "import" women to the Pacific Northwest to balance the gender ratio. The women were drawn by the prospect of moving to a boomtown with a surplus of bachelors.


Now I'm imagining moving to Seattle in the 1860s and discovering that the men in Seattle did their hair like that. I am not entirely sure I would stay!



How much nesting can we do in English verb tenses, and what controls that? For an example of what I mean, I can say:
- I eat
- I will eat
- I will have been eating
- I will have been going to eat

But I don't think we can say "I will have been going to have eaten".

in reply to Daniel Filan

One possibility: basically it goes as far as it makes sense to add extra timing information. But this only works if you disagree about your last positive example, which I personally don't actually think I've ever heard used.

Like, imagine a timeline. "I eat" describes a period of time encompassing now. "I will eat" describes a period of time in the future. "I will have eaten" describes two times; one in the future and one in the past of that future. "I will have been going to eat" describes a time in the future, a time in the past of that future, and a time in the future of that past of the first future. But in some sense this collapses back to the semantic content of "I will eat", and so my guess is that it's basically never used.

in reply to Ben Weinstein-Raun

Or, maybe I think your last positive example is sometimes acceptable, but only if the "going to" is actually describing an intention rather than tense information.
in reply to Ben Weinstein-Raun

I guess I don't get why it makes sense to talk about two times but not three.
in reply to Daniel Filan

I think what I mean is that additional times around the loop aren't really adding any extra information, because they introduce new reference points along the timeline that typically don't connect to anything else.

Like, there's some implicit time T that I'm trying to locate with a given statement, and there's an additional time Now that I get from just being in the present.

It makes sense to be like "Some time between Now and [implicitly / contextually defined] T, X will happen", and this is ~ the two-level wrapping. But if you say "Some time between Now and [newly introduced / 'bound' / 'scoped-to-this-statement'] T1, it will be the case that X happened after [implicit / 'free' / contextual] T2", T1 is kind of irrelevant, since it's introduced and used only within the statement.

In principle I guess you could have extra context that disambiguates, but I think it's also kinda relevant that verbs tend to have a subject, a direct object, and up to one indirect object, and typically not more than that.

This entry was edited (6 months ago)
in reply to Ben Weinstein-Raun

idk, I'm not sure this actually makes sense; the real answer might just be "ultrafinite induction"
in reply to Ben Weinstein-Raun

Yeah I guess I'm stuck on "well why can't there be a bunch of relevant times".
in reply to Daniel Filan

Also FWIW I'm still stuck on the fact that however natural it is, I have a strong intuition that "I will have been going to eat" is grammatical in a way that "I will have been going to have eaten" is not.
in reply to Daniel Filan

my take is that arbitrary nesting is in some sense grammatical, but when interpreting things like this in the wild, I have to weigh up "they really mean the complicated thing" vs "they mean a simpler thing, but have said it incorrectly", and as the things become more complicated the latter explanation becomes more and more likely
This entry was edited (6 months ago)


Life Update, December 2024


in reply to Gretta Duleba

Overall sounds really exciting 🙂 - I hope the holidays are as unstressful as it's reasonable to hope for; sounds like a huge effort!

I have never heard of geezer gyms; maybe I should look into this?

in reply to Ben Weinstein-Raun

I don't think it's a real category except in my own mind! My gym is small, densely packed with equipment, clean, and full of senior citizens. I think they just know value when they see it. :)


The UHC CEO murder has made me feel like I'm surrounded by bad people. I get the sense that "the UHC CEO assassination was good" is the default leftist stance.

It seems so absurd to me. They agree that people are innocent til proven guilty, but they're happy for CEOs to get executed based on a really flimsy understanding of their behavior.

Someone who runs prominent events I enjoy posted something that was (IMO) kind of misguided and cruel. I used to think "maybe these people would hate me if they knew my views" -- now I'm thinking "maybe *I* can't accept *them*."

Hank Green expressed sympathy... for the murderer.

This is such hatred. Bigotry feels like an understatement.

I just ordered some colorful genderqueer clothing from an indie brand. Do they support the assassination? I want to pick up fresh local bread from Berkeley Bowl. What about them? Am I supporting bigots? I wish it were practical to get away from this.

in reply to kip

Oh thanks, that set of poll results is actually mildly relieving
in reply to Daniel Ziegler

oh good. to be clear I didn't mean to imply "look how bad things are!" -- more like "here's some actual data; yeah things aren't extremely bad"
in reply to kip

I posted about this on Facebook: facebook.com/share/p/Yg3bYqU18… (I think it would be more preaching to the choir here)


I've now rescheduled my entire life around getting a hernia consultation twice, only to have UCSF reschedule at the last minute.

kip doesn't like this.



"Vouching" / friend-of-friend interactions


in reply to Gina Stuessy

I just realized that a title isn't necessary here; I think the "Submit" button looks disabled even when it's not (FYI @Ben Weinstein-Raun --also, I got an error with the "like" button a bit ago, but I just tried it again on my own post and it seems to have worked, but now I can't remove it, so I look like a dufus)


MOAR AXRP


This time with Erik Jenner, on a paper he's presenting at NeurIPS tomorrow - check it out if you're there!

Lots of people in the AI safety space worry about models being able to make deliberate, multi-step plans. But can we already see this in existing neural nets? In this episode, I talk with Erik Jenner about his work looking at internal look-ahead within chess-playing neural networks.

Video
Transcript



Tried using a portable vertical #hamradio antenna in my back yard this evening, as a replacement for the one I took down from the tree. It worked okay. Nowhere near the coverage I was getting from the 107ft wire, but I did manage to make a couple ft8 QSOs a few states away (South Dakota being the furthest).


Llama 3.3-70b is quite good; I think it's clearly the best local model I've tried. Not quite as good as GPT-4 on things I've tried so far, but I think better than GPT-3.5.
in reply to [object Object]

I'm not sure; I have an MBP with 128GiB of unified memory, which is plenty.


A wind storm two nights ago took a big branch down from the tree my antenna was in, so I took the antenna down until we have a chance to get the tree looked at. Very sad to have to pause my FT8 fun, but even if this is the end for a while I've had a great time.


Pascal's Wager doesn't go far enough:

Granted, the Christian God offers infinite rewards, but as far as I can find this is always in terms of "eternal" life or "eternal" communion with him, and so we can be confident that he is offering rewards only as large as the cardinality of the continuum.

So come on down to Crazy Georg's Omega Plus First Church of G...d: If you can conceive of a God advertising any size of infinite reward, G...d will match it.

niplav reshared this.

in reply to Ben Weinstein-Raun

Granted, the Christian God offers infinite rewards, but as far as I can find this is always in terms of "eternal" life or "eternal" communion with him, and so we can be confident that he is offering rewards only as large as the cardinality of the continuum.


FWIW I think it's plausible that the Greek words used in the NT doesn't have this sort of connotation.

in reply to Daniel Filan

I would find this surprising, since I don't model the ancients as having concepts for infinity that could correspond to larger infinities than this
in reply to Ben Weinstein-Raun

I'm imagining that the words / concepts they used were vague enough to include those higher cardinals - e.g. my understanding is that a lot of the words that get translated as "everlasting" could also be translated as "of the ages".
in reply to Ben Weinstein-Raun

I found this relevant and interesting chapter from Unsong by thinking "hmm, but Omega is an ancient word in some sense, and it's been more recently used in the context of infinities... and Jesus also referred to 'alpha and omega' to represent something like infinitude. So I can probably make a joke about kabbalah. Oh, but Scott Alexander will have already done that."

unsongbook.com/interlude-%D7%9…

in reply to Ben Weinstein-Raun

This entry was edited (6 months ago)


What compression algorithm did God use to send Jesus to earth on Christmas?




On Friday I tried to show @Daniel Filan how FT8 works, but I was having a really hard time getting QSOs. I was worried something was wrong with my #hamradio antenna setup, since the internet claimed that band conditions should be good. But this afternoon and evening I had a great time and got 17 QSOs across 5 different bands! So I think Friday must have been something transient rather than my (very janky) setup degrading.

I now have confirmed QSOs in 40 states, and unconfirmed ones in all but 3! (North Dakota, Delaware, and Vermont. Almost managed to get one in Delaware today, but wasn't quite able to complete the protocol) Plus 28 "DX entities" (mostly countries, but includes e.g. Alaska and Hawaii separately) on 6 continents!

Map of listening stations that heard me this afternoon:

W0AMT Jon reshared this.



in reply to kip

"It is good when people are happy" is one of the things I am going to most prioritize trying to instill in my children.
in reply to kip

I think this is a good direction but encourage you not to completely lose empathy for the other reactions – people who still don't like being interrupted or mansplained to do have their reasons for that and I wouldn't want to respond to their negative reactions by saying "that reaction is unconditionally wrong"


Am now up to knowing five words for types of slave in Latin.


Ended up deactivating my facebook yesterday. I wish I could have emotionally handled whatever was going on, but the only way I know how to productively deal with expressions of anger at that depth, apparently doesn't scale past one or two people at a time.

Last night I felt really conflicted about it. Like, I had just been trying to get people to give me harsh feedback, hadn't I? Doesn't this undermine that, or feel like a petty table-flipping move?

I still have some of those worries, but today I'm feeling like it was obviously the right move. Like if I had a gangrenous limb or something and had cut it off: It's pretty awful that I lost a limb, but it's way better than losing my whole self. Plus in this case I can reattach it if I figure out how to get rid of the gangrene.

in reply to Ben Weinstein-Raun

what was going on on FB that made you want to deactivate it, if you want to share? the last thing I looked at of yours seemed to be positively received
in reply to Gina Stuessy

I think this was a different post; basically, I wrote a post about the United Healthcare CEO assassination (the gist was, "it's wrong to express glee about someone's death"). It got a decent number of mildly positive reactions, but also a small cascade of intense negative reactions, a couple of which were kinda vicious.

Daniel Filan doesn't like this.



It's so fucking shitty that the easy way to feel better when someone is angry at you is to totally dismiss them as crazy or evil.


Jeroen Henneman, The Long Way Home
From: https://x.com/opancaro/status/186529216161008481


Wilhelm Kranz
From: https://x.com/0zmnds/status/1865291905249980735

#art

#art



Just realized I've been paying for my old group house's internet for almost three years. I'm out $3000, since there's no way I'm asking them to pay me back all of that; it's my responsibility to look at my own bank statements. Honestly relieved my burn rate will drop, but also I really need to pay better attention to my account statements.
in reply to Ben Weinstein-Raun

if $3000 is too much to ask them to pay you back, surely they should at least pay you the maximal amount that isn't too much :P

minimally you could be like "I don't think it's fair for me to charge you the full amount but if you could think about what seems affordable / reasonable to you and pay me that I would be grateful"

although tbh I think it kind of is fair to at least suggest they might pay the full amount, even if you don't want to insist on it

This entry was edited (7 months ago)
in reply to Ben Weinstein-Raun

also thank you for this story because I do read and categorise my bank statement line items and sometimes I say this to people and they're like "why do you do this" and it is sometimes hard to have a good answer :P




Have you ever successfully caused yourself to love something that you didn't naturally love? How did you do that? I'm especially interested in cases where you made a conscious decision that you were going to learn to love the thing, and then succeeded via strategy.
in reply to Ben Weinstein-Raun

My not-quite-the-answer is that I think I've encouraged myself in loving things. There's a feeling of getting obsessed with something that I occasionally notice myself having. I can try to cultivate that intentionally. The most notable example where this felt like a real successful guidance was in getting really into Rust.


it feels to me like in public writing there is an axis of ownership, personal writing vs. collaborative writing, and there's also an axis of completion, from publishing things that are done and no longer changed, to publishing things that are updated and amended and try to reflect your latest thoughts on something.

personal blogs are a thing
collaborative wikis are a thing
collaborative blogs are... somewhat a thing?
personal wikis are... much more rarely a thing?

but I think I want a personal wiki

in reply to Ben Millwood

I use LogSeq partially like a personal wiki, though don't publish it. I've observed some people using Obsidian (including publishing their graphs) basically like this as well.


Okay, I think a worthier target of "whoa, this programming language is cool" than V, is Koka.
in reply to Ben Weinstein-Raun

The other downside:

the delimited continuations in OCaml must be used linearly – every captured continuation must be resumed either with a continue or discontinue exactly once. Attempting to use a continuation more than once raises a Continuation_already_resumed exception.
It is left to the user to ensure that the captured continuations are resumed at least once. Not resuming continuations will leak the memory allocated for the fibers as well as any resources that the suspended computation may hold.


whereas I think Koka ensures things don't leak and also lets you resume multiple times, so you can do things like this: github.com/koka-lang/koka/blob…

in reply to Ben Weinstein-Raun

Apparently I was wrong about the standard library not having a sort function - it's just undocumented AFAICT! github.com/TimWhiting/advent-2…

Oh, I see, it's in the community std library: github.com/koka-community/std/…

This entry was edited (7 months ago)


the trouble with it being December is that I can no longer say "but it's not allowed to be this cold, it's only [current month]" because December is an Authorised Cold Month


:o

However, I should mention that I may be hallucinating these specific book titles and dates since I don't have access to a current book database.


  • Claude Sonnet 3.6
in reply to Ben Millwood

Kinda interesting that they chose that word to describe LLM falsehoods, like it gives the impression they are constantly having extremely boring trips XD


Gustave Doré
From: https://x.com/0zmnds/status/1863475184344174739
#art
#art

Lambda reshared this.



Misc notes on Latin learning


in reply to Daniel Filan

also it's kinda wild that in chapter 20 of the companion book the teacher is complaining to his slave how much his right arm hurts from beating his students. his solution to the pain? day drinking.
in reply to Daniel Filan

then he has a conversation about how he sucks at teaching and should just give up and live on enough money to buy bread and books

in reply to Daniel Filan

Also, if you have a "national parks passport", bring it! You can get it stamped at the end, which is Land's End, part of the Golden Gate National Recreational Area.


so I've long had an issue with my laptop battery, and e.g. in Windows had to figure out how to configure it from the command line to really never emergency shutdown when plugged in, which inexplicably you can't do from the UI

on Linux I knew it didn't last long, although mercifully I didn't have that issue specifically

Today (after buying a replacement battery at no small expense, but not having fitted it yet) I tried an experiment to try to set a baseline to see if the new battery would really be better, so I started up my laptop not plugged in. As usual, the battery charge reported immediately like 3%, and shortly thereafter 0%, where it stayed for two hours of normal usage before the emergency shutdown kicked in.

Obviously two hours is not a great battery life either, but I was expecting like five minutes and probably wouldn't have bothered buying a replacement if I knew it was capable of that. Oh well.



Victo Ngai
From: https://x.com/opancaro/status/1863111407962599592
#art
#art


New AXRP! With Evan Hubinger!


This time I won't retract it, I swear!

The 'model organisms of misalignment' line of research creates AI models that exhibit various types of misalignment, and studies them to try to understand how the misalignment occurs and whether it can be somehow removed. In this episode, Evan Hubinger talks about two papers he's worked on at Anthropic under this agenda: "Sleeper Agents" and "Sycophancy to Subterfuge".

Video
Transcript

in reply to Daniel Filan

I like how it looks like the AXRP logo is the sun in this thumbnail.


I actually like it when YouTube waits a while to start processing the video I just uploaded. It strengthens my character.



Messed up that Latin became the language of the intelligentsia in the middle ages and therefore has more pedagogical materials available now, when Greek has classical authors you obviously should care more about. Like, it has the philosophers! Not to mention the New Testament (and the version of the Hebrew Bible that the authors of the New Testament were familiar with), the Iliad and the Odyssey, and Greek myths (let's be real nobody cares more about Roman myths than Greek myths). Yes, it's cool that Latin has De Rerum Natura, Apuleius, and Cato, and the tradition of scholarship is a nice bonus. But c'mon!
in reply to Daniel Filan

I further guess this is a cautionary tale about the tradeoff between writing and conquering.
in reply to Daniel Filan

Tbf I think the Romans owned this, they were like 'the Greeks theorize, we get shit done'
in reply to Amber Dawn

It's like the LessWrongers and the EAs. As a LWer myself, I know where my sympathies lie...
in reply to Daniel Filan

LW: democratic and philosophical but also factious and discourse-ridden
EA: run by 1 or 2 extremely powerful guys who sometimes turn out to be deranged and corrupt. A woman called Julia is also involved.
in reply to Daniel Filan

Counter-argument: the point of learning an ancient language is to read the poetry not the prose (since prose is easily translated) and Latin poetry is plausibly better than Greek poetry.



I think web-of-trust is underused for spam and abuse prevention.

e.g. there could be a pretty simple "endorse" button for each account, which is basically saying "I vouch that this person is a real human and not a troll/spammer". Webs/chains of endorsement could be used to prove that someone ought to be able to interact with you. And for any given active interaction attempt ("react", "friend request", "tag"), there could be an opportunity to mark it as "spam", and accounts with lots of spam could become untrusted, and accounts that endorse lots of spam accounts could become untrusted as well.

In principle you could even implement this in an entirely decentralized way with some public-key crypto, though it might be too expensive in practice.



Latin practice day 7


These aren't very inspired but:

I. Cūr quaeque littera Graeca pulchrior est quam quaeque littera Latīna?
II. Sī linguam Latīnam scīre vult, quotiēs quamque litteram Latīnam scrībere necesse est?
III. Vōlōne ā magistrō laudārī?
IV. In Capitulō XVI, quia Dominus Iēsus tempestātem facit apud navem Lydiae? Lydia ā Deō dīligiturne?
V. Num medicus labōrans vērē sanat hominēs aegrōs?
VI. Num parēntēs laudant magister discipulōs verberāntem?
VII. Suntne bēstiolae industriorēs quam apēs? Quid facit illae?
VIII. Quia dea est pulcherrima?
IX. Hōdiē, quae bonae rēs daminī ā deī?

#latinpractice

in reply to Daniel Filan

Hōdiē sum in domō parentum matris mea, in Arizonā. In hāc domō, saepe dormō in lectō parvō in cubiculō parvō, sed hōdiē habeō magnum cubiculum ac magnum lectum. Cēnābam cum parentibus matris meus, et cum amīcīs suīs. Aliī hominēs ēdēbant magnam avem, sed ego edēbam botulōs quī ex holeribus fīunt, nam Pythagoricus sum. Cōnspiciēbāmus pēs-pilam (harpastum? calcifollem? I guess Vicipaedia uses "Harpastum") - Leōnēs Detroitī, quī amantur ā parentēs matris meus, vincēbant contra Ursōs Sicāgoensis!

(I only know the imperfect past tense, forgive me)