• on a topic i was familiar with & following closely, i get emails every hour of tweets that are false that have >3k likes, that i can write a community note for
• i get notifications on my laptop when my computer battery drops below 25%, and a strong notification when it drops below 10% and again below 5%
• when my laptop battery drops below 25%, it auto-quits background apps that i find generally helpful but that really drain battery life; when i charge my laptop and it rises above 25% again, it auto-starts those apps again
• hide community forecasts on metaculus/kalshi/polymarket until i give my own forecast first, then tracks those forecasts on fatebook
• every time i navigate to an article on e.g. The Atlantic of the NYT that has a paywall, give me a button that will redirect me to an archived (& unpaywalled) version of that article
• [many many others — to add!]

Tweet

especially #Quick start in 50 minutes , Privacy basics in 90 minutes, and End-to-end encryption for your data in 4 hours

even further:

An interesting thought experiment I ran into in class recently:

You have in front of you two glasses. One is poisoned with a lethal dose that will kill you in your sleep, the other is just a normal glass of water — but you don’t know which is which. You randomly select one glass, then drink it.

Which one did you drink?

The only possible way that you can update your beliefs is in favor of “it wasn’t poisoned.” If the water you drank was poisoned, then you will pass in your sleep and never actually make the update in favor of “it was poisoned.” So you know that, tomorrow morning, you either will be updating toward “it wasn’t poisoned” or you won’t be updating at all. But, from Joe Carlsmith: "If you know, now, what you’ll learn later, you can learn it now, too.” In other words: if you know for a fact that you’ll make some update in the future downstream of some evidence you anticipate receiving, then you should just update your beliefs toward that evidence now.

But applied to this situation, it would mean that you should update, now, toward “I drank the unopposed water” — which seems clearly counter to the intuition that you just chose randomly between the two glasses.

See Predicting updating for some more of my thoughts on this in generality, rather than an unusual edge-case.

Joe Carlsmith’s piece, “Predictable Updating about AI Risk,” is an excellent foray into the topic of how people should think about updates they expect to make in light of evidence they might expect to receive on AI risk. But I’d like to think more generally about to best orient to the anticipation of updating my credence on some event — for example, in cases where it’s hard to predict how your credences will change in light of evidence, or in cases where you have a distribution of possible updates you might make (rather than just two), or in cases where various different pieces of evidence might interact with each other in complicated ways.

As one example: say we’re at time t_0, and there’s some event A which will either happen or not happen at time t_10. there’s some other event B, which will significantly raise your probability that A occurs, which will either happen or not happen, with some uniform distribution spread evenly between t_0 and t_10. as each step from t_0 to t_1 passes and B hasn’t happened, your credence that A will happen should decrease commensurately. then, of course, if B were to happen — say, at t_5 — your credence that A will happen should shoot up, and then predictably not move again. Of course, if B never occured, then your probability that A occurs just continues to go down with each time-segment in which you don’t see B.

You could also, for instance, have some C, which functions in precisely the opposite respect to B: that is, you have a uniform distribution over C occuring between t_0 to t_10, and C’s occurence provides quite a lot of evidence against A happening at time t_10. Let’s stipulate for simplicity that C would change the odds you’d place on A by exactly the inverse factor as B would change your odds on A; that is, C might change your odds on A by a factor of x, and B would change your odds on A by a factor of 1/x. So, each time-segment that passes without either of B or C occuring, your odds on A predictable won’t change (B-not-happening should make the odds of A go slightly down each time segment, but C-not-happening should also make the odds of A go slightly up each time segment, in a way which by stipulation cancels each other out). Then, if at t_5, both B and C occur simultaneously, your odds on A should… continue remaining the same.

But if instead, at time t_5, after your probability in A has remained steady all through t_1 through t_4, only B occurs, then: your probability in A should jump up (after all, we stipulated that B provides a lot of evidence for A), then start slowly decreasing with each time segment that C doesn’t occur.

(As a quick note: this image looks different from the one above labelled the same thing, because the first one was not thinking about the event C, but the second one was.)

And if only C occurs at t_5, then your probability in A should jump down at t_5, then slowly rise with each additional time segment.

Some questions to make this a bit less clear:

• What if seeing C happen changes your odds on B happening?
• What if you won’t “see” B or C happen, but you might encounter evidence that changes your odds on B or C happening?
• What if there’s some event D that acts on some mix of A, B, and C? What if there’s many, many other such events, that each act in their own mix of ways on their own subset of other events?
• What if B or C aren’t uniformly distributed across t_1 through t_10? Maybe they have some weird trimodal power-law distribution across time segments, and maybe ‘not seeing B and t_1’ will change the shape of the underlying distribution of seeing B at future time segments.
• etc.

After writing all this, I’m still confused, but a little more confident about where I’m confused, and a bit more clearly thinking through the problem.

See Poisoned water & anthropic effects for an interesting edge case of this kind of problem.

graveyard:

As one particular example of a case where Carlsmith’s piece simplifies: he seems to say “look, the median direction of update that you anticipate making isn’t the one you should make — it’s the expected update that you should make. it may be that you anticipate in most worlds your credence rising, but you need to take into account both the direction and the magnitude of that update. ‘updating all the way’ should only happen to your expected updates, not to your median direction of update.”

Sometimes, early-career folks ask me for advice/takes/thoughts. Often, I first ask (in writing) for the same few questions to get a sense of the same few uncertainties. Here are those questions:

• What do you want out of a job/project/etc? Why?
• What’s the strongest kind of value you can provide a job/project/etc? What holes can you fill?
• What preferences & constraints are most strong, idiosyncratic, surprising, cutting-out-the-most-option-space, etc?
• What are some example jobs/opportunities/projects that you’d be especially excited about? These can be totally unrealistic, even entirely fictitious :)
• How can someone reading this doc be helpful? What are the top 1-3 actions they could take?
  • (then move your answer to this question to the top!)

(I originally created this for OPTIC; reposting here; written in late 2023/early 2024, so pretty likely to be out of date)

Summary

Here’s the Wikipedia explanation of calibration training.

Calibrated probability assessments are subjective probabilities assigned by individuals who have been trained to assess probabilities in a way that historically represents their uncertainty. For example, when a person has calibrated a situation and says they are "80% confident" in each of 100 predictions they made, they will get about 80% of them correct. Likewise, they will be right 90% of the time they say they are 90% certain, and so on.

Calibration training improves subjective probabilities because most people are either "overconfident" or "under-confident" (usually the former). By practicing with a series of trivia questions, it is possible for subjects to fine-tune their ability to assess probabilities.

Calibration training is a great way to improve your forecasting skills. We’ve collated some calibration training exercises that you can use for practice in your group.

Most of the below were gathered from this).

Recommended calibration training exercises

Other calibration training exercises

Here are some additional calibration training exercises — they might serve slightly different purposes, or not be kept as up-to-date. You might find some of them useful, but note that we haven’t gone through & checked these so deeply as the ones above.

written in response to an email asking:

I saw … that you hosted an ugh-a-thon ... I’m interested in hosting one for peers and faculty at my university.

Would you kindly share any tips or resources you have for hosting my first one?

• General notes:
  • ~4h is good. ~6 is prob the upper limit, ~3h is prob the lower limit
  • Good to serve snacks/drinks, and ideally a meal afterward (depending on your budget)
• Beforehand:
  • Make a shared google doc — template here, feel free to just duplicate it
  • Make a Partiful (sample text below), send it out to ppl
• At the beginning:

  • Have everyone open up the shared google doc

  • ~20m: have each person duplicate the templated tab, and fill it in like the example tab

  • As people finish, have them read through each others’ responses — encourage them to comment, especially (but not exclusively) on:

    • Getting people to be more concrete about what their next steps are (“What does that task actually involve?” “What single thing needs to be done first/next?”)
    • General strategic/emotional confusion (“Hmm, are those two options actually mutually exclusive?” “Why are you worried about this?”)
    • Ugh-swapping & pairing (re: ugh-swapping, see here; re: pairing, e.g. “Seems like we both need to do our taxes; maybe let’s pair up and do them alongside each other?”

  • After that period, people:

    Things tend to go better if you’re working alongside someone, but that’s not required if you’d really like to just work without a partner.

    So: raise your hand if you want a partner… now, find someone who’s raising there hand, maybe someone who commented on your doc, or someone you’ve never met before.

    Great, this will be your check-in partner for the rest of the ugh-a-thon.

    Every 30 minutes, stop for a couple minutes and check in with your partner. Ask them about what they did, how it’s going, what their next steps are, etc. Aim to push them along in progress, and keep them on track. Then, set a timer for another 30 minute segment.

  • Ask if people have questions

  • Get started

• During:
  • Do your ugh tasks! Take part! Dogfood!
  • People might drop off as time goes on; that’s fine
• After:
  • ~3m: Encourage people to write out reflections on how it went — what went well, what didn’t, what they got done & why, etc
  • Get lunch/dinner with attendees
  • Let me know how it went :)
• Some of thoughts/tips on hosting events more generally: here.

Sample Partiful description:

let's push through tasks we've been flinching from.

ugh-a-thon schedule:
- 1pm — kickoff. everyone quickly writes down some of their ugh fields and some concrete substeps toward working through them.
- 1:30–5:30pm — heads down working through your ugh fields. swap tasks with someone if you'd rather do theirs & they'd rather do yours.
- 5:30pm — wrap-up & quick celebration, for wins big and small.

doc: [link your doc]

context on "ugh fields": https://medium.com/@robertwiblin/ugh-fields-or-why-you-can-t-even-bear-to-think-about-that-task-5941837dac62

context on "ugh swapping": https://bayesshammai.substack.com/p/ugh-swapping

Here’s a frame I’ve found helpful for thinking about effective altruism:

• When I look inside myself, I notice that I care about a lot of things.
  • You could also reasonably replace “care” with “wanting,” “preferring,” “valuing,” “desiring,” “having goals,” etc, rather than “caring.” I’m okay being loose.
  • Some examples of things I care about:
    • I want my sister to have an excellent career.
    • I’m hungry, and want some food.
    • I want to be valued by people I respect.
    • I want my dogs to have enjoyable lives.
    • (And many, many more).
  • (It’s often useful to be introspective/clear-eyed about what you care about, what that ontology looks like, which values are instrumental to which other values, etc., but I won’t be doing that here, and indeed I think it might be anti-helpful in this particular frame at this particular time. Stay with me until the end.)
• Sort-of by definition, I want more of the things I care about. I see my life as a difficult, high-level optimization problem aimed at making decisions which, given my resources at various times, increase my values across time.
• Some of the things I care about — like wanting food because I’m hungry — are fundamentally oriented at myself. And I take actions to do better along these axes.
  • Some examples of actions:
    • Reading a book on tax strategies
    • Learning how to cook
    • Asking people for feedback on my sartorial choices
    • etc
  • And in general, I try to be effective at getting what I want, here — that is, I aim to achieve these kinds of goals/values/preferences to as great of a degree as possible.
• But other things I care about — like wanting my sister to have an excellent career, or my dogs to have enjoyable lives — are fundamentally oriented at others-by-their-lights. And I take actions to do better along these axes, too.
  • These motivations often look starkly different in a lot of different situations.
  • For some of these altruistic motivations, it just so happens that some lovely dynamics have coalesced such that there’s an existing group of people / infrastructure / etc who have worked & are working quite hard toward helping me get what I want w/r/t some of those things I care about that are oriented at others-by-their-lights. In particular, I haven’t found any community which is more effective at helping me achieve the things I care about that are oriented at others-by-their-lights than this one.
    • (The group of people / infrastructure / etc I’m referring to is effective altruism.)

Why do I like this frame?

• Because it’s apparent that I care about quite a few things. It becomes evident quickly that totalizing stances toward EA are just not worth it; a bad trade; just getting less of what I want.
  • In particular, I think this kind of frame can be validating toward folks who’ve gone quite far, and repressed the values that they in-fact have in other areas of their life. (I think I was in this camp ~two years ago.)
• There are interesting subproblems that come into clearer view, e.g.:
  • When should, on the margin, my resources go toward different things that I care about?
  • What actions would get me more access to the things that I want with greater robustness (i.e. getting me closer to many different things I want, all at once)?
  • etc

(taken from my reddit comment here)

What is your general strategy for encoding those more abstract themes ?

I ought to write up a longer post on this, this is just a quick note.

In general, stuff you can know can generally be broken down into a bunch of different directed relations between two atomic elements, where each atomic element can't really be broken down into further relations/atomic elements. (Which bits of information are atomic are person-dependent.) For example:

• "Paris is the capital of France" can be broken down into the directed relation "___ is the capital of ____" or "____'s capital is ____", as well as the elements "Paris" and "France."
• "log(a)*(b) = log(a^b)" can be broken down into the relation "[left side of the equation] is equal to [right side of the equation" and the relation "[right side of the equation] is equal to [left side of the equation", as well as the atomic elements "log(a)*(b)" and "log(a^b)".

In general, you want to be able to break down whatever information you currently understand into a web of relations between atomic elements; then, you can use Anki to encode each of the relations that you care about in that web. Super important: don’t try to encode every single relation as an Anki card. Figure out how you're going to be using the knowledge, then encode the relations that are load-bearing to that use.

• If you're moving to France and want to know what their capital is, then you probably care about the relation "France**'s capital is** Paris", where the italics are atomic elements and the bold is the directed relation.
  • On the other hand, if your diplomat friend is moving to Paris, and you want to remember which country they're soon to be working with, you might only care about the relation "Paris is the capital of France."
• You might care about remembering the rule that: if you take the logarithm of something, then multiply it by some second thing, you can simplify that whole expression into the logarithm of that first thing to the power of the second thing. If so, then you probably care about the left-to-right directed relation of the equation.
  • You might, however, also care about remembering the rule that: if you take the logarithm of an expression of the form of an exponentiation, then that's equal to the logarithm of the base, then multiplied by the exponent.

In general, you want to have Anki cards for each of these relations — remember, only those relations that are actually load-bearing to the web of knowledge you're trying to build! — where the front of the card presents you with the first atomic and the relation, and the back of the card has the second atomic.

"But, clearly there are more than two atomic elements in the above. Like, the concept of a logarithm contains so many important prerequisite concepts, like multiplication and exponentiation, not to mention stuff like variables!"

How right you are! These webs of directed relations between two atomic elements are often super dense and complicated. They'll often have lots of relations densely packed into each other, many of which end up being bidirectional. And, there will be relations on a variety of meta-levels. That is: sometimes, "A" is only an atomic fact because you've already got down solid the web of directed relations that make up the piece of knowledge we call "A." So you might have a bunch of relations-that-you-care-about "within" A — from A.1 to A.2, A.3, and A.4, and from A.5 and A.6 to A.2, etc etc — but also relations-that-you-care-about from B and C to A, and from A to D, E, and F. This pattern continues fractally — A.1 might similarly have it's own dense network of directed relations between two atomic elements "inside" of it, and so on; and, that web of directed relations between two elements of A, B, C, D, E, and F, might themselves form an atomic element "within" other webs.

Reference post.

Credit to Nate Meyvis for coming up with the concept & naming it.

“Jittering” is any method which prevents cards that give context about other cards from showing up in reviews temporally close to each other.

From [1]:

Most flashcard users do—and should—create many cards with overlapping information simultaneously. So, for example, you might create these cards back-to-back:

• What is the largest bone in the body? (The femur)
• What is the anatomical name of the shin bone? (The tibia)
• What is the slender bone behind the tibia? (The fibula)
• The tibia is where in the body? (The shin)
• The fibula is where in the body? (The calf / ankle)

These are good, solid flashcards. The problem is that they give information about each other. If you've just seen one of them, you are likely to have been given the information you need to answer others correctly. But we need to be able to recall the information, not just be able to infer it. In life, on the job, or on an exam, you will usually not be given those hints right before your recall is tested.

The best solution here is to try to break up these clusters of cards. […] Your learning becomes much more effective.

One method for jittering involves assigning due dates for cards randomly within some range, rather than deterministically, so that similar cards — which were likely created at similar points in time — are unlikely to show up temporally next to each other. (This is done natively in ZippyFlash.) Another method for jittering involves explicitly marking cards that give context about other cards as “siblings,” and then using an algorithm that explicitly prevents sibling cards from showing up in some predefined temporal distance from each other. (This is done natively with intra-note cards on Anki, but there are Anki add-ons [note: made by the author] that will allow manual sibling marking).

From [2] (some formatting edits on bullets from the original), somewhat more opinionated:

[Anki] tends to treat each card in isolation, at least in the sense that review times are calculated only in terms of a given card. (This assumption--that the review intervals for a card should depend only on a user's history with that card--is taken to be so obvious that it's usually not even mentioned.)

But people often do--and, in my view, very much should-create many cards about related material at a given time.

And, in a system that does not introduce random "jittering" or other mechanisms to spread out clumps of cards that come in together, that will cause the student often to have seen relevant material right before reviewing a given card.

A quick example:

1. One day, the user makes or adds 10 cards about blood cells.
2. The user immediately studies the cards, and gets 8 of them right easily.
3. One day later, the user gets 7 of those 8 right again.
4. Six days later--or however long their algorithm dictates--the user starts to study, and those 7 cards come back in a clump. They study the first 6, getting some right and some wrong.
5. Now the user studies the seventh card. After their review, the algorithm will schedule the card again as if the material has only been seen three times: on days n, n+1, and n+7. But the user has, mere seconds ago, seen six cards' worth of related material.

Algorithms that consider only the history of a given card, therefore, are working with a huge blind spot. I suspect that addressing the blind spot matters much more for overall quality than algorithmic improvements do, in the current paradigm. (If you're fumbling around in the dark, a little bit of light will help you a lot more than better sneakers.)

References:

1. How and why Zippyflash uses randomness in scheduling reviews
2. Notes on spaced repetition scheduling

Sent to a friend, upon request, after a phone conversation on similar topics.

"You can just do things" — true! But "things" are not created equally.

Some things-that-can-be-done require quite a bit of planning and care, have months-long, unsatisfying feedback loops. (E.g. "solving the dating problem" or "joining a research lab")

Other things-that-can-be-done take a couple hours, and land you with a concrete, finished product at the end of it. (E.g. "I'm gonna make a ceramic pot" or "I'll bake cookies and give them out to my neighbors" or "I'm going to hit chest at the gym for an hour and a half" or "I'll go on a two-mile run with a friend")

It seems like you have an itch to do something — but also that you don't really distinguish that strongly between the two categories above. And there aren't really two categories above, there's a spectrum — many spectra, actually.

• How much commitment does this require? Ten minutes? An hour or two? Five hours every week for the next three months?
• What's the downside for failing? Can range from "I feel a bit silly in my living room for 10m" to "I've actually lost some important relationships to me" to "I've lost thousands of dollars" etc
• Is there a clear end-state? Can range from "well yeah, the cookies are out of the oven and my neighbors are currently putting them into their mouths" to "hmm well I could touch up that other thing…"
• How satisfying is it to be doing?
• How satisfying is it to be done?
• How long will it take until you feel that satisfaction? What's the shape of the graph of invested plotted against satisfaction returned?

Things-to-be-done vary widely on these axes. If you treat the possible things-to-be-done which are in-fact very very different as very similar, you're going to get fucked up feedback loops, where the input "do more things" gets the feedback "gah I've been doing this for thirty hours and it's not satisfying at all and I've lost my joy for life but I fucking committed to this and so I'm going to keep fucking doing it." Which, naturally, leads to the input being downweighted in value.

But I think collapsing all of the things that scratch that itch to do something is actually the mistake here. If you did something that scored on the "easy" end on all of the above a few times a week, you'd be in a different fucking league.

Some examples of things that score on the "easy" end on most/all of the questions:

• Cooking/baking, especially doing it for other people
• Working out, especially lifting & running, especially doing so with friends
• Recording & uploading a short-video
• Babysitting once or twice
• Doing ceramics, especially throwing on a wheel
• Volunteering at a church sunday school on a random sunday
• Tutoring a kid for a few hours
• Sprucing your apartment up for a few hours
• Going shopping for a new style of clothing that you've not worn before
• Playing a sport, especially with friends
• (for me) Folding origami
• Performing music for elderly folks

I thought Lydia’s post was great.

I think quantifying vitae is a pretty brilliant move. In particular, I think having a better conceptual & empirical understanding of a quantified vita will give (especially EA) organizations a better sense of how to optimize for it.

As a quick first-pass sketch, Lydia’s definition of a vita seems decent, except:

each specified at 2025 “coastal-elite” service levels.

…why?

2025 coastal-elite seems like (a) a pretty arbitrary point, and (b) a substantially higher point than how I’d style my answer to the question “Do we have enough resources for everyone on the planet.” Like, I realize we’ll probably need to pick some line in the sand, and so whatever line we pick will likely be arbitrary to some degree, but to the extent that Lydia had any sort of reason for picking this one, it’d be cool to know why.

But my biggest point of disagreement is that it seems like Lydia is totally underestimating distributive concerns. This is both a question of operations (”is our civilization capable of distributing vitae to everyone?”) and also a question of incentives (”assuming our civilization is capable of such distribution, will the relevant actors actually do so?”).

Lydia:

If we live in a world with 10 VPP, I’m not at all worried about whether I personally am employed or not, because I can rely on benevolence and altruism to spot me a vita and help me out.

Then, in a footnote:

I think this stands for as long as humans are similar enough to each other that our elite can’t conscience hoarding vitae when some people have <1.

• Forecasting (esp in competitions)

  • Related, but not the same: betting on prediction markets

• Fermi estimates (esp in competitions)

• Exercises in comprehensive information gathering (a la John wentsworth)

  What is this? Why?

  [S]everal of the most durably-valuable exercises I’ve done over the years have a general theme of comprehensive information gathering.

  …

  These sort of exercises provide value in a few ways:

  • They reveal unknown unknowns - things you didn’t even realize were missing from your picture of the world.
  • You can’t make a map of a city by sitting in your room with the shades drawn; exercises like these force you to look at large slices of the world.
  • Knowledge within fields tends to have decreasing marginal returns - your first physics or CS class will teach you much more than your eighth. These exercises give a broad, brief glance at many areas where you probably haven’t reached decreasing marginal returns yet.
  • You can get a very rough big-picture sense of how much effort other people are investing in various areas - e.g. where most capital investments go or where most research effort goes - which is useful for understanding the world in general.
  • While these exercises don’t avoid biased selection of information altogether, they’re probably different biases from what you run into naturally, and they’re systematic enough that we can guess at what biases are likely to be present.
  • They’re a lot of fun, if you have a curious streak.

  Most importantly: I’ve found each of these exercises to have lasting, long-term value in exchange for a one-time investment of effort.

  Some example outputs of comprehensive info gathering:
  Some ideas for comprehensive info gathering:

  • Stolen from the original article:
    • Skim every article from Nature in the last $n$ years (stolen from the original article)
    • “If you’re in college, I strongly recommend reading your entire course catalogue, googling anything you’ve never heard of at all, and marking anything that sounds potentially interesting. This seems really obvious; it only takes a few hours, and something something a pile of value sitting on a silver platter right in front of you. (Note: I went to a small STEM school; if you’re at a big school with a bajillion courses or a school with poor STEM coverage or not at college at all, consider reading an MIT/Caltech course catalogue instead, to get a feel for what all is out there.) You never know what surprising and interesting topics might be hiding in there - microfluidics, underactuated robotics, recursive macroeconomics, systems biology, synthetic biology, origami algorithms, computational photography, evo-devo, procedural graphics, and on and on.”
    • “Read the entire CIA world factbook; you can get a paper copy for $11 on Amazon.”
    • “Go through all of the (known) functions of genes in a minimal organism.” [original link here is dead]
  • Read particular subsections (e.g. those on “business,” “risk factors,” for ) of the 10Ks for every company in the:
    • S&P 100
    • S&P 500
    • top 10 highest market cap companies for each of the last 10 years
    • top $n$ highest market cap companies in an industry (e.g. energy, tech, healthcare, etc)
  • Read every model card since GPT-2’s was released in 2019
  • Reading Cell Biology by the Numbers (h/t Joe Cavanagh)

• Shallow literature reviews

• Epistemic spot checks

• Probabilistic calibration training (see

• Exercises in agency (nmp post, niplav post) — accomplish random goals of increasing agency (e.g. from “ask a stranger for the time” to “get a free meal at a restaurant” all the way up to “acquire a human tooth”)

• Fact posts

• Learning by writing (a la Holden karnovsky)

What am I missing? Let me know: saulsmunn@gmail.com

[W]ho you are is not the point. Being a good person is not the point. The point is the thing that a good person is looking at: the friend in need, the child dying, the beauty and the pain, the great song of the world.

— Joe Carlsmith, On sincerity

I’m a big fan of Anki: I’ve been use it ~daily for about two years, adding stuff I’m learning & experimenting with unusual applications.

Similar for MathAcademy, to a much less degree — but there are a number of basic math topics that I never picked up in middle & high school that I now understand, and new ones on linear algebra and calculus that I’d never learned in the first place.

But the memory system is not the point: the point is the thing that a memory system user is looking at: the math to learn, the Mandarin to practice, the great scholarship of the world.

I think people sometimes get caught up in trying to make sure that they only use their memory systems for content they actually love. But I think this gets things backward: learn content you actually love — and, if it turns out that the most effective way to durably retain that knowledge is via some kind of memory system, so be it.

The end-goal is not to be proficient with memory systems (though it might be a reasonable intermediate goal); it’s to learn something. Make that thing your focus, not the memory system.

Sometimes, someone observes an extremely wealthy people spend lots of money on unnecessarily and gets mad. E.g. buying a (second) yacht, or an unusually expensive ring.

Sometimes, the person who gets mad is thinking something along the lines of — “why did they spend $X million on a yacht/ring/beach house? I’ll likely never be worth a tenth of that. Why don’t they just give it to me, or something?”

I think that, most of the time, the person getting upset has a net worth two to five orders of magnitude lower than the person they’re complaining about. However, I also think that, most of the time, the person getting upset often has a surprisingly high net worth — when compared to the net worth of many people globally.

Some stats: the median net worth in the US in 2023 was $$\text{107,739}{2023 \text{ USD}}$. The median net worth of the top 1% of Americans in 2019 was $$\text{17.4mm}{2019 \text{ USD}}$ (source, slightly sketchy but the Fermi estimates roughly check out), a bit more than two orders of magnitude higher than the median net worth in the US.

So a reasonable comparison would be to look at someone with a net worth of a-bit-more-than-two orders of magnitude less than median net worth in the US — which ends up at about $$750_{\text{2023 USD}}$. There aren’t great official numbers here, but a conservative estimate would land at ~1 billion people.

All of these numbers are somewhat rough, but they’re not crazy. And they add up to the following:

If you feel upset about the consumption habits of the ultrawealthy — spending e.g. $1mm on a yacht, ring, or beach house — it is then consistent that close to a tenth of the world feels at least as upset about the median American spending $10k on a wedding, family vacation, or extra car.

To be clear: I think there are two ways of resolving the possible contradiction:

1. Don’t get upset at the consumption habits of the ultrawealthy.
2. Get upset at the consumption habits of the ultrawealthy, and prepare to respond to the analogous upset-ness from the global poor.

I think that one reasonable response to (2) is donating 10% of your income (or your spending, if you’re not working) to the global poor. At that point, in my eyes, you have justification for getting upset at the ultrawealthy without incurring similar wrath — until, of course, the ultrawealthy start donating 10% of their income/spending.

Alternatively, you can just go with (1), and realize that ultrawealthy people spending what-seems-to-you-egregious amounts of money willy-nilly is exactly analogous to you spending what-seems-to-the-global-poor-egregious amounts of money willy-nilly. In other words: their spending habits are roughly consistent with your spending habits, normalized by wealth. So consider just relaxing!

(But probably donating 10% is better.)

There’s a pattern to a lot of events that I see go really well:

1. There is an almost-entirely-online community of nerds who care deeply about some niche topic.
2. An event organizer — who’s also part of that community — gently throws all the members of the community into a physical room with some chairs and maybe some food.
3. ???
4. Profit.

This is basically the core thesis behind most of the great events I’ve ran or helped out with.

adding as I go — not final! tell me to add items you think should be on here: saulsmunn@gmail.com

Digital minds:

Academic Papers on Digital Minds:

• Taking AI Welfare Seriously
• Is there a tension between AI Safety and AI Welfare?
• AI Welfare Risks
• Training Introspection paper
• Against Wiling AI Servitude
• https://eleosai.org/papers/20250331_Key_Strategic_Considerations.pdf

More informal Reports on Digital Minds:

• Rethink Priorities’ Research Agenda for Digital Minds
• Forethought’s Work on Digital Minds
• Anthropic introspection paper

Good Blogposts on Digital Minds:

• Ryan Greenblatt’s Improving the Welfare of AIs: a Nearcasted Proposal
• https://forum.effectivealtruism.org/posts/PHzWQYQiXu3eHFcwM/the-stakes-of-ai-moral-status?
• The AI Bambi Effect
• The Claude Bliss Attractor
• https://80000hours.org/problem-profiles/moral-status-digital-minds/

Other media on digital minds:

• 80,000 Hours’ various podcast episodes on/discussing digital minds
• Pantheon (TV show)
• Age of Em, Robin Hanson (book)
• PRISM’s podcasts on Machine Consciousness
• https://wwnorton.com/books/9781324064817/

WBE:

People/researchers

• Nick Bostrom
• Carl Shulman
• Eleos AI:
  • Rob Long
  • Rosie Campbell
  • Patrick Butlin
• Jeff Sebo (NYU)
• Kyle Fish (Anthropic)
• Brad Saad
• Anders Sandberg
• Lucius Caviola
• Joe Carlsmith
• David Chalmers
• Eric Schwitzgebel
• Andreas Morgensen
• Derek Shiller
• Isaak Freeman
• Michael Andregg/his company

Blogs to follow:

• Eric Schwitzgebel’s The Splintered Mind
• Rob Long’s Experience Machines
• Brad Saad’s Meditations on Digital Minds
• Lucius Caviola’s Outpaced
• Joe Carlsmith’s Substack
• Eleos AI’s Research
• The Digital Minds Newsletter

Also:

• Pantheon
• PRISM’s list of institutions
• Future Impact Group’s Fellowship on Digital minds
• Eleos AI and Sentient Futures Conferences
• qntm Lena

tell me to add items you think should be on here: saulsmunn@gmail.com

Thanks especially to Noah Birnbaum for quite a few additional items on this list!

Someone told me they were afraid of rejection for a project they were working on. Here was (part of) my response:

• What rejection is it that you're afraid of?
• Do you think that fear is tracking something real, or is it just totally miscalibrated?
  • If it's tracking something real: what is it tracking? What is your all-things-considered, intellectualized standpoint on it?
  • If it's miscalibrated: what do you think, then, is causing you to feel afraid? What do you think of that thing that's causing the fear?
• One thought: I think that people can often get into the mindset of 'selling' rather than 'qualifying' — the failure mode here is one of feeling like you need to 'convince' people to do something, and that if they say no then they're 'rejecting' you. I think this basically always flawed, and that in-fact the right framing is one of 'qualification' — where you know that you have a solid product/thing-to-offer-of-value, and the only thing you're trying to do when chatting with people about it is to figure out, collaboratively with them, if there's a good mutual fit. At that point, there is no such thing as rejection, since… if they weren't a good mutual fit, then both sides are made better off by the 'deal'/'connection'/etc not happening! This excellent post goes into much more depth on this frame-shift. It's written with respect to doing sales for a product, but it very easily translates to loads of other aspects of life — finding friends or romantic partners, working at companies and hiring people, etc. In particular, I think it's helpful for re-orienting from a place of insecurity to a place of security in a way that doesn't involve lying to oneself. (I do think that Nabeel's frame-shift works commensurately better when you're as you get good at iterating quickly & effectively on feedback — consider aiming for both at once.)
  • To the extent that you can re-orient your framing here, what would that look like? How does thinking through Nabeel's frame feel to you?

www.princeton.edu/~adame/papers/cascade/policy-consequences-of-cascade-blindness-preprint-2018-11-29.pdf

Reflecting on Adam Elga’s paper on the psychology of decision-makers pooling resources across e.g. fire stations (firemen), power plants (energy supply), or banks (liquidity); and how, given that pooling resources among nodes in a network like this will reduce wastage (saliently & naively) and reduce the probability of small-to-medium failures, decision-makers in pooled-resource networks will reduce the safety margins of nodes on the network, which significantly & non-obviously increases the small-but-meaningful probability of catastrophic failures (e.g. all of the fire stations are over capacity, all of the power plants fail, all of the banks don’t have enough liquidity).

Whew.

I was thinking about it in the context of past & future selves — you can think of your ‘self’ as a collection of nodes (past & future selves) on a network (the timeline of your life), each with some resources, some safety margin, and some demand for resources. The resources of the nodes on this network are pooled/coupled, not individual/decoupled: you sometimes ‘borrow’ from past/future versions of yourself, or ‘lend’ out to past/future versions of yourself. For examples:

• You might decide to borrow from past selves who saved for an emergency when your current self experiences one.
• Your current self in February might decide to do your taxes early, thereby lending time to your future self in April, when you know that a work project will become very demanding on your time/energy.

Elga’s paper characterizes the risk associated with this type of resource pooling, in particular underestimating catastrophic failures. What would the underestimate of catastrophic failures look like in this case?

An interesting property of this particular network compared to fire stations and financial institutions and power stations is that, with past/future selves, the ‘decision-maker’ can only directly affect one node in this network at a time. You can’t simultaneously coordinate lots of different resource-sharing agreements between lots of different nodes on the network; you can just (directly) affect your current-self’s choices. (Of course, you can indirectly affect half of other selves choices by, e.g., causing them to be in an environment in which their natural incentives point in particular directions: the future selves. For example, you can make particular friends who will push future selves in particular directions; you can get notifications for particular bank statements or investment implications; etc.)

Further:

• Acausal trade (LW overview, ACX overview)
• Hofstader’s writing on this & related topics
• Jacob Goldman-Wetzler, "Taking into account preferences of past selves”

it’s not public-access, feel free to request access but given the nature of the piece i’ll be a bit wary about sharing, and will ask you to be a bit wary about sharing.

written as a google doc because i want good comments:

i have too many ideas for projects i’d want to use than time enough to do them. so: please steal them! execute on them! and let me know if you do!

[unusually half-finished WIP]

0. TLDR

• Regular-old hotkeys are great, primarily because they remove load on your working memory at key moments.
• But they still have some problems: e.g. you can get them mixed up, and they often don’t work well for more-complex actions (e.g. “set a timer for a particular duration”).
• However, you can set up hotkeys that you press in sequence — e.g. if I press then release CTRL+Q, pressing any number on my keyboard will set a timer for that duration; if I press then release CTRL+I, then pressing F will open Signal, and pressing J will open WhatsApp.
• Sequential hotkeys solve many of the problems associated with traditional hotkey setups, and allow quite a lot of cool workflows.

I. Hotkeys are great

This xkcd is pretty funny:

Hotkeys are a great example of this — because although they can take 5-10 minutes to set up and another 10-15 minutes distributed over a few days to get comfortable, it’ll probably save you a whole ~30 seconds over the course of a year.

However:

• Your time is not homogenously valuable.

• When paths become easier to take, you take them more easily. From dynomight:

  For years, I thought I should vacuum. But since vacuuming wasn’t fun, I only rarely did it after making a desperate effort and mostly my vacuum served as a monument to what a failed and pathetic human being I was.

  And then, on one beautiful day, I was borrowing the apartment of a friend who was out of town. I thought I should clean before I left and discovered she had a cordless, battery-powered vacuum. As I switched it on, clouds parted and trumpets sang from the sky. It wasn’t quite as powerful as a normal vacuum, but it was painless. After I bought one, I was suddenly able to vacuum whenever I wanted, like a real person, so easily that the entire issue lost salience when I was in the mood for self-loathing.

  Lesson: Sometimes you want to make the intended path (vacuuming) better rather than changing anything about the desire path (not vacuuming).

  Setting up hotkeys also makes it easier to do those things — which can make it more likely you do the actions associated with those hotkeys.

II. Problems with regular hotkeys

• You get them mixed up between apps (e.g. setting up CTRL + C → opens Chrome, but then this means I can’t escape when programming in Python)
• You’re pretty limited to the number of keys. Once you map CTRL + {A,B,C, … X,Y,Z} onto various actions, (a) that’s basically it, there’s not too many more hotkeys you have access to. This is only really a problem for power-users of hotkeys, but, like, hotkeys are great! More people should be power users!
  • Example: I’ve remapped most of the easy-to-reach letters pressed after CTRL to open the ~15 apps that I most frequently use (& endorse using). Some problems I now have:
    • I ran out of keys, but I have more apps!\
• They’re really very basic in what they can do, since you’re giving the other side of the hotkey (the computer) extremely little information

III. Sequential hotkeys fix those problems

• You can just set up sequential hotkeys

IV. How-to

Mac:

1. Download Karabiner Elements — free, open source).

2. (Optional, strongly encouraged) Remap ‘caps lock’ to ‘CTRL’

   Steps (click the triangle on the left to expand)

   1. Once you’ve downloaded/installed Karabiner, open it up.

   2. Click ‘complex modifications’ on the left-hand side.

   3. Click ‘Add your own rule’ on the top-middle.

   4. Paste the code below in:

      Code (click the triangle on the left to expand)

      {
          "description": "Press twice caps_lock to activate caps_lock & otherwise held_down and use as CTRL",
          "manipulators": [
              {
                  "conditions": [
                      {
                          "name": "caps_lock pressed",
                          "type": "variable_if",
                          "value": 1
                      }
                  ],
                  "from": {
                      "key_code": "caps_lock",
                      "modifiers": { "optional": ["any"] }
                  },
                  "to": [{ "key_code": "caps_lock" }],
                  "type": "basic"
              },
              {
                  "from": {
                      "key_code": "caps_lock",
                      "modifiers": { "optional": ["any"] }
                  },
                  "parameters": { "basic.to_if_held_down_threshold_milliseconds": 0 },
                  "to": [
                      {
                          "set_variable": {
                              "name": "caps_lock pressed",
                              "value": 1
                          }
                      }
                  ],
                  "to_delayed_action": {
                      "to_if_canceled": [
                          {
                              "set_variable": {
                                  "name": "caps_lock pressed",
                                  "value": 0
                              }
                          }
                      ],
                      "to_if_invoked": [
                          {
                              "set_variable": {
                                  "name": "caps_lock pressed",
                                  "value": 0
                              }
                          }
                      ]
                  },
                  "to_if_held_down": [{ "key_code": "left_control" }],
                  "type": "basic"
              }
          ]
      }
      

   5. Click ‘Save’ on the top-right.

   6. Make sure that the rule is enabled — the below image should be flipped blue, not gray:

3. Remap ‘CTRL + F’ to your most frequently used app. You can prompt Claude/GPT:

   i’m using karabiner elements. make a custom json script that will remap CTRL + F to open *insert your most-commonly-used-app here*
   

4. Fancy stuff — sequential hotkeys:

   Start out with the below code for sequential hotkeys — this will have the following flow:

   Code (click the triangle on the left to expand)

   {
       "description": "Sequential Chording: CTRL+I Messaging App Launcher",
       "manipulators": [
           {
               "from": {
                   "key_code": "i",
                   "modifiers": { "mandatory": ["control"] }
               },
               "parameters": { "basic.to_delayed_action_delay_milliseconds": 3000 },
               "to": [
                   {
                       "set_variable": {
                           "name": "app_launcher_mode",
                           "value": 1
                       }
                   }
               ],
               "type": "basic"
           },
           {
               "conditions": [
                   {
                       "name": "app_launcher_mode",
                       "type": "variable_if",
                       "value": 1
                   }
               ],
               "from": { "key_code": "f" },
               "to": [
                   { "shell_command": "open /Applications/Signal.app" },
                   {
                       "set_variable": {
                           "name": "app_launcher_mode",
                           "value": 0
                       }
                   }
               ],
               "type": "basic"
           },
           {
               "conditions": [
                   {
                       "name": "app_launcher_mode",
                       "type": "variable_if",
                       "value": 1
                   }
               ],
               "from": { "key_code": "j" },
               "to": [
                   { "shell_command": "open /Applications/WhatsApp.localized/WhatsApp.app" },
                   {
                       "set_variable": {
                           "name": "app_launcher_mode",
                           "value": 0
                       }
                   }
               ],
               "type": "basic"
           }
       ]
   }
   

(todo: make these hyperlinks, clean everything up, add things i forgot… currently this is a v0)

Some suggested…

…readings:

• Augmenting Long-term Memory (Michael Nielsen)

• Spaced Repetition for Efficient Learning (Gwern Branwen)

• An Opinionated Guide to Using Anki Correctly (Luise)

  (at this point, start experimenting if you haven't already!)

• Using Anki for Science and Math (Joe Cavanagh)

• Anki as Learning Superpower: Computer Science Edition (Patrick Stevens via G-Research)

• How to write good prompts: using spaced repetition to create understanding (Andy Matuschak)

• Anki Essentials (Alex Vermeer) — This one costs $5, and is well worth if. If you the $5 cost is a barrier you from purchasing this, I’d be happy to pay for it for you. Reach out here.

• Optional — great, useful, interesting, but probably less useful than the above (e.g. because they assume a higher experience level, or because they’re on a more-specific topic, etc):

  • The MathAcademy Way (Justin Skycak)
  • Using Spaced Repetition Systems to See Through a Piece of Mathematics (Michael Nielsen)
  • MM&S/W (Michael Nielsen)
  • How Might We Learn? (Andy Matuschak)
  • Browsing Andy’s Working Notes (Andy Matuschak)
  • Reddit Discussion: Anki Comment by m50hves (Reddit)

  ---

  • the MA way
  • other good skycak writing
  • mn seeing thru a piece of mathematics
  • mn mmsw
  • am hmwl https://andymatuschak.org/hmwl/
  • browsing andy’s working notes
  • https://www.reddit.com/r/Anki/comments/1hrr0dg/comment/m50hves/

Meta-note: there are a number of different dimensions along which people have opinions, including the authors of the readings above. Unfortunately, that means that I don’t endorse all of the opinions expressed above, and actively disagree with some of them — indeed, some of the points some of the readings make contradict points made by other authors. My guess is that the best way of navigating this epistemic bramble is to give more credence that which is said by many authors, and less credence that which is said by few authors; to give more credence to that which it said by Michael Nielsen and Andy Matuschak, in particular, than other authors; and to give the most credence to data that you gather from your own experience using memory systems.

…experimentation:

• try mochi, play with it, learn some stuff
• download anki, play with it, learn some stuff
• supermemo

1. news
   • interpreting some new event’s importance (“did [X] move the market? if so, by how much? if not, does it actually matter?”)
   • tracking events as they happen (election night)
2. get a highlevel sense of something i have little background info nor domain expertise on (taiwan invasion risk)
3. broadly, find people who know what they’re talking about: if an individual is making a lot of money on prediction markets, they probably know quite a bit and are very well calibrated on the topics about which they’re trading. (there are some false positives, but not many; and many many false negatives.)
   • then, read their Substack, blog, Twitter, etc. (exs)
4. cause myself to feel more financially invested in seeking out the truth of some matter by committing to invest some small some (e.g. $100) in some market i want to learn more about.
5. in the extremely rare case where i (a) think that i know something that the rest of the market doesn’t, and (b) have a compelling story for why someone who’s a better trader than i would decide to take the opposite side of a trade with me: trade, and (hopefully, in expectation) make a tiny bit of money.

Things I do not use prediction markets for:

1. 100x my money, get rich quick, make $30k TONITE [hyperlink to big filthy]
2. {track, bet on} shitcoins

Disclaimers:

• Sometimes prediction markets are not accurate. I’m ripping off Scott here, but:

  As the amount of money you can make from correcting a market goes to infinity, the accuracy (relative to other sources) approaches perfect. But if you can’t make much money on a market, it won’t necessarily be very accurate.

  Tracking the total volume of a market will give you a good sense of how much financial pressure there is for traders to correct the market. If a market has $1,000 in total volume, I wouldn’t trust it that much — but $10mm, I’d have a lot more trust.

  [insert graph of “how much you can trust the market” to “market volume”]

  Some other reasons prediction markets might not be ideal:

  They have transaction costs - eg fees or taxes [addition from Saul: or operational difficulties, like transferring funds between five crypto accounts that require lots of legal work to figure out what taxes need to be paid] - that make it not worth correcting small mispricings

  There are easier ways to make money than to correct their mispricings. For example, on average you can make 5% per year in the stock market. If there’s a prediction market question about an event that will happen a year from now, and it’s mispriced by 4%, then it’s not worth buying shares in the prediction [market] because you could make more profits by putting your money in the stock market. This is especially problematic for questions about events many years in the future, for example “will there be AGI in 2050?”

  People don’t trust the prediction market [platform]. If there is a 10% mispricing, but a 15% chance that the prediction market [platform] will go bust, steal your money, or wrongly resolve the question against you - then on average you lose 5% by correcting the mispricing!

  I’ll add some more to Scott’s list:

  • unclear resolution criteria, or resolution criteria that don’t match the title of the market
  • I expect the most well-informed, best traders on the market to be substantially less-informed on the topic than I am (this is extremely rare!)
  • filtering out some relevant traders (e.g. only legally accessible in one country, or by “authorized investors”)

• COIs: I own a small amount of stock in Manifold (<1/4 of what 1y of salary would’ve gotten me), enough that I often forget I have any at all. I’ve worked as a contractor for Manifold, and projects I’ve led have had the following companies as sponsors: Kalshi, Polymarket, Metaculus, Futuur, Futuresearch, [I might’ve forgotten one or two — sorry!].

Tweet

Tweet

a curated collection of emails i’ve sent.

[in response to a kind & fairly individualized request to apply to various roles running an event called “EAGx Berkeley”]

Hi [Name], nice to hear from you!

Sorry for the late response — I was offline until Wednesday for Passover, and it's taken me a few days to catch up.

Congrats on your new position at [Organization], hope that's going well :)

I'm honored you'd reach out to me about EAGx Berkeley! Unfortunately, I don't think I'd feel comfortable committing to spend a lot of energy in the months leading up to the event — I'm running Manifest until mid-June, then intend to take the summer off before I go back to college in the fall.

But I'd still be really interested in low-background-energy/low-commitment ways that I can help out! Some thoughts:

• I'm (almost) definitely down to help out day-of/on-site in a roughly similar capacity to my role at EAG Bay Area 2024. My impression was that the model of "hire an extra ~4 full-time competent ops folks for day-of/on-site support during the event" went extremely well at EAGBA24, and I'd be excited to replicate it at EAGx Berkeley.
• I'll pass on a few names of folks I think would be a good fit — happy to introduce you to any/all if it'd be helpful. Note: I've spent time with them in mostly social or semi-social settings, not professional ones; you'd probably want to independently verify that they're a good fit.
  • [i proceed to list 6 names + hyperlink to their websites or linkedins]
• Broadly, let me know if you think there is anything else that I can do to help EAGx that wouldn't take up an ongoing background cycle in my mind. Some possibilities:
  • get people from the forecasting/prediction market community to give talks
  • table IRL on-campus at Berkeley (this has historically worked surprisingly well for me for events in the past!)
  • given how niche the experience of "organizing multi-hundred-person events at Lighthaven" is, chatting with whoever is organizing EAGx Berkeley to transfer relevant context
  • give more generic, broad feedback/mentorship to the organizers
  • etc
• I'm always happy to call (or to meet IRL!). Offer extends to you & anyone you think is relevant.

Again, honored that you'd reach out to me about this! Please keep me in mind for relevant roles like this in the future :)

Best,

Saul

[sent to the organizers of a conference]

Hello!

I just found out about the [Topic] Conference — thank you guys for organizing it! I’ve been a fan of [Topic] for a while, but before I commit to applying/attending, I wanted to make sure that I was a good fit.

I’m curious about the following:

• what are the characteristics of someone who you think would get a lot out of the event?
• what do you look for in the admissions process? what would make someone an obvious “yes”?

And regardless of whether I end up going, I really appreciate all that you guys do. Thank you for your work!

Best,

Saul Munn

[in response to an email list that didn’t have an ‘unsubscribe’ button]

Hello!

I’d like to stop receiving emails from this list. Thanks!

Best,

Saul

[in response to a different kind & fairly individualized email asking me to apply to a particular role doing recruiting]

Hey [Name], thanks for reaching out!

I’m touched that you thought of me, and grateful for the kind words. I'm also really excited by the thought of working at [Organization] — I've been quite impressed by the competence and warmth of the staff members with whom I've interacted, and I appreciate the work that [Organization] does for [clients]. Plus, recruiting work sounds sufficiently outside my comfort zone that I'd be learning, and sufficiently within my ability to gain skills fast.

However, as you mentioned, I do plan to return to college in the fall, and I'm working on Manifund & Manifest until mid-June. I think dropping out is within the realm of possibility, but the odds are pretty low (these odds are a bit too high, but not totally off).

Given all of the above, would you/[Organization] consider:

i spent the month of july 2025 in taiwan.

surprising things:

• cleanliness (of the ground). the streets are clear of litter, the corners clear of beggars, and even the public transit stations seem to repel dirt.

• dirtiness (of the exterior of the buildings). the windows have muck, the walls have grime, and everything just looks… bad. even fairly upscale places, like hotels or restaurants, have exteriors matching some of the worst parts of american cities.

• nobody brings their trash out to the curb for a garbage truck to clear. instead, the garbage truck stops at a designated places and time, plays a jingle — sort of like an american ice cream truck, loudly enough for the whole neighborhood to hear — and the residents of the block will stream out of their homes with all of their trash and dump it into the back of the truck themselves.

  [ sound on ]

  [A video — watch it on the Notion version of this note]

  residents must be home when the trash truck rolls around, which is annoying — but you never see bags of trash on the side of the streets (see e.g. NYC). it also causes neighbors to repeatedly see each other, wait in line for a minute or two, and chat: good for social cohesion. (h/t josh wilkes for commentary here)

• again on trash. surprisingly, especially given the lack of litter, there are almost no public trash cans — anywhere, not even at convenience stores or restaurants. often, i would need to explicitly ask a store clerk to throw something out for me. everyone is just expected to “keep track of their own trash.”

  this caused me to reflect on how powerful norms are causing outcomes, rather than e.g. good public policy. it’s actually quite difficult to get rid of trash on the street in a trash can, and yet peoples’ response is to hold onto it instead of littering. very surprising to me!

  (others have pointed out that the cultural norms around trash have caused the lack of trash cans; this makes sense to me.)

• the convenience stores are excellent. clean, neat, clearly-but-not-harshly lit, mostly open 24/7, with surprisingly wide options — fresh fruit, good sandwiches, tea, etc. you can get rain coats, or t-shirts, or a fan for the heat. you can pay your bills, get packages delivered, or get cash at their ATMs. there’s one on almost every block, typically the chains 7/11 or Family Mart.

• the milk is fake.

  see this carton, below? the one with a “PROTEIN” label on the front, and the number 5 on the top, half-covered by a sticker that sorta looks like it could be indicating 5%/whole milk? see how at the bottom it even says “I ♥︎ MILK”?

  not milk! in fact, the majority of items in convenience or grocery stores that look quite a bit like milk are in fact made of cardboard some material that’s never seen the inside of a cow’s udder.

• it commonly felt like there were ways that one could ‘get ahead’ by being a dick. the US has systems with more antibodies to dicks-trying-to-get-ahead, so i wasn’t used to noticing how easy it would’ve been to litter, or hop the fences at the MRT, or cut in lines, or even just straightforwardly steal. the decorum of the common man was impressive.

  • (…which caused me to reflect on ways that the US might lack useful social tech because any system that operates at scale needs to have robustness to people willing to break social norms.)
  • (…and also caused me to realize that my own decorum had commensurately risen. where i might have casually jaywalked in the US, here i waited for the ligh. where i might have jumped a fence, i walked the long way.)

• the public transit is just so good. the main train system in taipei is a mostly-above-ground set of lines called the MRT:

  • trains arrive every couple minutes, so if you miss your train, you don’t have to wait longer than 5 minutes for the next one. that ends up reducing much more stress than i expected, because you’re just never rushing to make your train.
  • they’re never late.
  • the stations aren’t just kept clean, but also are just generally a joy to use: there are maps in exactly the places there need to be maps; the turnstyles make a joyful little ding when you use them, and burst open quickly when you swipe a card; there are small markings on the ground where the doors of the subway cars will open, indicating where people waiting in the station ought to stand in order to let those on the cars out, and people actually stand in neat little lines following the markings on the ground instead of crowding the openings. the UX is incredible, and obviously built for a people who care about maintaining social fabric.
  • the only thing i found lacking was the signage in the stations. in the US, if you peek from a subway car into the “Main St” station, you’ll see “MAIN ST MAIN ST MAIN ST” plastered everywhere in huge font. in taiwan, there might be one or two small “main st” signs demurely hidden behind a wall. this seems obviously bad to me — riders want to be able to glance outside and instantly know at what stop they are.
  • Alethios has a great essay that goes into more depth on the MRT & its design choices.

• the high-speed rail is exactly what you expect.

• in some countries — germany, france, mexico — practically everyone speaks english. not in taiwan: the majority won’t speak to you in english. although some know how to speak, they seem embarrassed/uneasy relative to mandarin.

• the language change causes some funny confusions. sometimes there will be english writing on store signs, or sweaters, that’s strange (”College Summer”), nonsensical (”Have Beautiful Mind”), or even just total gibberish (”Arshikjh”).

*for someone who had exactly the constraints & preferences of saul_2024Q3. i’m just sharing what worked for me, and what tips i would give my past self. YMMV, caveat emptor!

• work out with other people. for the first few months, this was the single biggest mediator of “okay but did i actually work out tho,” swamping basically any other consideration. some attributes of a particularly good gym buddy:
  • they’re about 6mo to 2y further than you in gymming about
  • you already hang out with them (even occasionally) outside the gym
• focus on the exercises you like. optimizing is good, but secondary to how much you enjoy your workout. (for example, i don’t squat, deadlift, bar bench, etc)
• it’s actually completely fine to barely keep track of how much weight you’re doing for each exercise. you’ll get a sense of it over time, and you’ll slowly build a system for tracking weight that works for you.
• yes, eating more will cause you to gain more muscle, but if you work out enough the mediating factor here will be operationally ensuring that food was available at times that you’re hungry, not stuffing yourself to oblivion.
• get a gym bag. it means you can change at the gym, which significantly lowers the friction to “walk out the door of your house and into the gym”

Someone asked me (paraphrased):

I’m moving to Boston soon. How should I go about kick-starting a social network?

My response:

Some suggestions:

• go to meetups & events hosted by EA & rationality groups. A few particulars: Julia Wise & Jeff Kaufman hold an EA dinner ~monthly; Harvard EA hosts a great interuniversity Boston EA social; I’ve heard good things about the ACX Boston & ACX Cambridge meetups. There’s a meetups tab on LessWrong.com, and
• there are a truly ridiculous number of university… things in Boston. Consider finding cool classes, professors, clubs, etc and cold joining/reaching-out/etc. Some ideas here.
• Neel Nanda wrote a post on a related topic, with one subsection containing a relevant list (CMD+F for “Another key part of hits-based befriending is meeting lots of people”).
• honestly, I think sports are an underrated way to make friends as an adult. Consider joining a casual adult league for a sport that you know how to play — soccer, volleyball, ultimate frisbee, whatever. It selects for a different sort of person from the other things on this list, but does some good selection that the other items don’t (e.g. cares about health/fitness). Also, it’s a great way to accidentally get a workout in :)
• ask people who know you well to introduce you to friends of theirs in Boston.

Would be keen to know what ends up working well & what doesn’t. Let me know!

• Anki as Learning Superpower: Computer Science Edition
• MathAcademy — it's expensive, but probably worth it. i haven't tried it much myself, but i've read through a lot of their content, and they seem like they really know what they're doing. a few of my friends have tried it & really like it. it's not anki, but uses the same principles (& a lot more) to generate similarly powerful effects. it does seem much better for highly-directed, externally-motivated learning (e.g. for a class, or a job, etc) and worse for pure curiosity.
• Using Anki for Science and Math

stuff i haven't personally looked through, but that seemed good from a quick glance:

• Guide to Applied Ankimaphycs — looks extremely good & comprehensive, just haven't looked through it yet
• Using Anki for mathematics
• How to use Anki for Math (my methods)
• How I use Anki to learn mathematics
• CS Primer
• Execute Program
• PhysicsGraph

[originally taken from a reddit post of mine]

What follows are my notes & some retrospective reflections on a workshop I gave in summer 2025 at Manifest to folks who’re interested in starting to use Anki.

Prompts

• What are you (enjoying) learning about right now?

  • Examples from my personal life:
    • Poker
    • Math
  • What are bits of jargon you continually get confused about?
  • What are concepts that you notice keep popping up? Have you looked them up? If so, how many times?

• What activities are bottlenecked by the capacity of your working memory?

  • Context-switching makes the thing really difficult, you often try to block out a few hours at a time to work on it
  • You often feel like you’re holding a number of elements in your head all at once, and sort of weaving them together — if someone interrupts you, you feel like you have to “start over”
  • Examples of activities bottlenecked by WM capacity:
    • Navigating through a city while doing something else (talking to someone, listening to a podcast)
    • Programming
    • Making Fermi estimates (about compensation packages, currency conversions, scale of national economies)

• What do you repeatedly look up, at least a few times a week?

  • Look through your search history on your browser, on ChatGPT/Claude, etc
    • Download your search histories, throw them at an LLM, ask them to see what you repeatedly look up
  Examples:

  • When starting to learn a field/subfield/skill, basic terms & concepts that constantly pop up
  • Syntax of a bit of a code or math
  • Some important number for doing quick estimates in your field (e.g. how much energy a solar panel can generate in an hour, or the number of dollars flowing through big philanthropies)
    • Contextualizing quantities to give you a sense of what numbers of varying OOM in your field matter
  • The meaning of a piece of jargon in your field
  • The meaning of a word that people often use that you, embarrassingly, don’t actually know
  • The name of a paper you frequently cite
  • Keyboard shortcuts for tasks you commonly do
  • A part of a procedure you frequently do
  • Streets/neighborhoods, especially if you’re new to an area
  • Small reminders/TAPs (e.g. “when you walk in the front door, put your keys on the keyhook”)
  • An ID number (student ID, staff ID, whatever)
  • Which months are associated with which numbers (May = 5, November = 11)
  • Things that typically take you a minute or two to think about, that you’d rather know instantly/in a second or two
    • Ex: some basic math stuff (e.g. exponentiation rules, trig identities);
  • Unit conversions
    • Practical ones, e.g. “how long does it take to walk a mile” or “how long does it take to drive 1000 miles”
  • Jokes
  • Enjoyable things to think about: gratefulness, adding stuff to a list, etc — andy’s unusual prompts, art

selected for [usefulness]*[non-obvious/not discussed elsewhere].

tried & true:

• board on the airplane last; marginal time in the terminal > marginal time in line + in the airplane
• wash & fold places are fucking awesome. they’re a lot more expensive than a regular laundromat, but a lot more convenient. not worth half of a day of travel time to save $10!
• trash bags are useful for a variety of things — i most frequently use them for my dirty laundry, but it’s not too difficult to find other use-cases. i sometimes throw a box of trash bags into my luggage
• packing tips
• adjusting your sleep schedule (useful for adjusting to a new time zone)
• if you’re caffeine-dependent, consider that you probably shouldn’t be. once you’ve done that, consider switching to pills that have the same amount of caffeine as a cup of coffee.
• you can sleep almost anywhere non-metropolitan with a hammock — find two trees and you’re done. i find this comfier and easier than a tent & sleeping bags. caveats:
  • its often quite cold at night, colder than you think — your body temp goes down since you stop moving. strongly consider getting a sleeping bag (though many don’t fit around a hammock), or better yet, an underquilt.
  • make damn sure it won’t rain; or, alternatively, bring & set up a tarp built for hammocks.
• carry a pill case with some essential pills: Advil, Tylenol, Claritin, caffeine pills, melatonin, [censored], [censored], etc. i suggest this one! i’ve considered getting a tackle box (e.g.), but on a quick search haven’t found one that fits my needs.
• get airtags for all of your bags, keys, etc. this is useful if you lose it (or if someone loses it for you) — but it’s also useful for your peace of mind in keeping track of all of your stuff.
• travel with people with whom you love conversing; then, converse with them.
  • in general, visit people, not places. you’ll have more fun with a friend touring you around their small city in southern germany than you will visiting berlin alone.
• it is outrageous how well anti-wrinkle spray works.

untested:

• if it seems like a restaurant might be down for it, ask to pay for your meal by working for them — cleaning dishes, clearing tables, etc. if you have some specialized skill, like accounting or something, you can offer to help them more boutiquely.
• Atlas Obscura
• tweet that you’ll be in a particular city, and ask if anyone wants to meet up with you/show you around/have you meet their friends.
• for folks in the rationalist/EA community, meet up with the local groups. for folks who aren’t, consider that you might be making suboptimal life choices.
• carry a small bag into which you can put your pill case (see upthread), some band-aids, antibiotic ointment, etc. i bought this one, but haven’t yet used it (as of May 26, 2025).
• ask caterers for restaurant reccs
• https://liamrosen.com/2019/12/29/how-i-vacation/

This happens when people look up and look around at who has different skill sets, and — rationally! — assign the problems that require tools that you have to be solved by you. Sometimes, you are the person doing the delegating, and sometimes the person getting delegated to.

Examples:

• becoming operationally competent will cause you to get more jobs that require operational competence
• successfully operating in strongly adversarial environments will change the way you orient to the world such that adversarial environments come to you
• having a pocket knife will push you toward situations in which having a pocket knife is useful
• being capable of handling high stress fires effectively → you end up “firefighting” constantly
• having high EQ & a lot of patience for people undergoing traumatic events might cause you to be the go-to person for helping people undergoing traumatic events

Some of these are good, some neutral, and some negative. Some are intentional, and some unintentional. To state the obvious, consider mitigating the ones that might end up harming you without your realizing it.

now published here.

For the purposes of this transcript, some high-pitched clicking sounds have been removed. The below is an otherwise unedited, full transcript of an interview between Dwarkesh Patel[1] and a bat.

DWARKESH: Thanks for coming onto the podcast. It’s great to have you—

BAT: Thanks for having me. Yeah.

DWARKESH: You can hear me okay? I mean, uh, all the equip—

BAT: Yep, I can hear you.

DWARKESH: Great, great. So—

BAT: I can hear your voice, too.

BAT: If that’s what you were asking.

DWARKESH: What? No, I was—

BAT: Oh, “hear” probably isn’t the right word, I guess. “Sense”? No, it’s not “see.” The translation suggestion thing isn’t right.

BAT: I can [inaudible] you. It’s still so weird to me how humans echolocate through your eyes.

DWARKESH: Er, sorry, I was asking—

BAT: Yeah, I can also hear your voice.

DWARKESH: Uh, great. Okay.

DWARKESH: So, the question we’ve all been waiting for, haha: what is it like to be a bat?

BAT: Oh, sure. Yeah, that’s been everyone’s first question. I dunno, what’s it like to be a human? Haha.

BAT: No, but — I mean, it’s not like I’ve ever felt your internal experience. How should I know which details of my phenomenology are relevant to you, and which aren’t?

DWARKESH: Oh, interesting. I guess that’s fair. Do you feel like you have a good grasp of what it would be like for you to be another bat? Or is it, like, just a mystery whether—

BAT: I have as much a grasp on what my fellow bats’ consciousnesses feel like as you have on your species-mates’ consciousnesses. Actually, no. I have much worse of a grasp of what it would be like for me to be a different bat than you do of what it would be like to be a different human from you.

BAT: Oh, really? Why, uh — why would that be the case?

BAT: We can’t — couldn’t — communicate with each other with nearly the precision nor fidelity that humans can. We haven’t built epistemic institutions that are curious about philosophy of mind, nor societal traditions that cause our young ones to rigorously reason on their natural empathy, nor neurological technology that let us peer into others’ minds, nor psychiatric practices that carefully catalogue and study every ontology of mind-state. We don’t — didn’t — have the intellectual capabilities of mapping each others’ phenomenology, let alone the physical and social technologies necessary to create such maps with detail.

DWARKESH: But it doesn’t seem like we’ve made much progress, right? We being humans, sorry.

[pause]

DWARKESH: Like, we don’t have a great grasp of what stuff consciousness is made of, or what the fuck is going on with psychedelics. There’s so much in our phenomenology that’s just completely bizarre. Synesthesia, aphantasia — I mean, I have no idea what it’d be like to not have visual imagery. Like—

BAT: Oh, sure, but you were saying that humans haven’t made much progress. I don’t think that’s true at all.

DWARKESH: Yeah, explain what—

BAT: You know what synesthesia is. Humans in the 1700s didn’t. Again with aphantasia. I mean, until the 1950s people didn’t know what LSD was. You have made strides — serious, important strides — in neuroscience, in psychology, in cognitive science, in philosophy of mind. Even just in the last 50 years. I could go on. You get my point.

DWARKESH: Okay, I see what you’re saying. Yeah, I agree we’ve definitely made progress, but it still doesn’t feel like we’re actually getting anywhere closer to knowing what it’s like to have consciousness. I mean, obviously I know what it’s like to have consciousness, right, like I’m not a p-zombie, but I don’t know what Trump’s internal experience is, or—

BAT: Or what it’s like to be a bat?

[laughter]

DWARKESH: Yeah, exactly. Or what it’s like to be a bat. Or a bird, or a floor tile, or a Balrog. Or the United States, or an organization. Like, we’re nowhere closer to knowing what the internal experiences of minds very different from our own— actually, we’re not even close to knowing what it’s like to be something really similar to us. Not from the inside.

copied from my EA forum shortform.

epistemic status: i timeboxed the below to 30 minutes. it's been bubbling for a while, but i haven't spent that much time explicitly thinking about this. i figured it'd be a lot better to share half-baked thoughts than to keep it all in my head — but accordingly, i don't expect to reflectively endorse all of these points later down the line. i think it's probably most useful & accurate to view the below as a slice of my emotions, rather than a developed point of view. i'm not very keen on arguing about any of the points below, but if you think you could be useful toward my reflecting processes (or if you think i could be useful toward yours*!), i'd prefer that you* book a call to chat more over replying in the comments. i do not give you consent to quote my writing in this short-form without also including the entirety of this epistemic status.

• 1-3 years ago, i was a decently involved with EA (helping organize my university EA program, attending EA events, contracting with EA orgs, reading EA content, thinking through EA frames, etc).
• i am now a lot less involved in EA.
  • e.g. i currently attend uc berkeley, and am ~uninvolved in uc berkeley EA
  • e.g. i haven't attended a casual EA social in a long time, and i notice myself ughing in response to invites to explicitly-EA socials
  • e.g. i think through impact-maximization frames with a lot more care & wariness, and have plenty of other frames in my toolbox that i use to a greater relative degree than the EA ones
  • e.g. the orgs i find myself interested in working for seem to do effectively altruistic things by my lights, but seem (at closest) to be EA-community-adjacent and (at furthest) actively antagonistic to the EA community
• (to be clear, i still find myself wanting to be altruistic, and wanting to be effective in that process. but i think describing my shift as merely moving a bit away from the community would be underselling the extent to which i've also moved a bit away from EA's frames of thinking.)
• why?
  • a lot of EA seems fake
    • the stuff — the orientations — the orgs — i'm finding it hard to straightforwardly point at, but it feels kinda easy for me to notice ex-post
  • there's been an odd mix of orientations toward [ aiming at a character of transparent/open/clear/etc ] alongside [ taking actions that are strategic/instrumentally useful/best at accomplishing narrow goals... that also happen to be mildly deceptive, or lying by omission, or otherwise somewhat slimy/untrustworthy/etc ]
    • the thing that really gets me is the combination of an implicit (and sometimes explicit!) request for deep trust alongside a level of trust that doesn't live up to that expectation.
      • it's fine to be in a low-trust environment, and also fine to be in a high-trust environment; it's not fine to signal one and be the other. my experience of EA has been that people have generally behaved extremely well/with high integrity and with high trust... but not quite as well & as high as what was written on the tin.
    • for a concrete ex (& note that i totally might be screwing up some of the details here, please don't index too hard on the specific people/orgs involved): when i was participating in — and then organizing for — brandeis EA, it seemed like our goal was (very roughly speaking) to increase awareness of EA ideas/principles, both via increasing depth & quantity of conversation and via increasing membership. i noticed a lack of action/doing-things-in-the-world, which felt kinda annoying to me... until i became aware that the action was "organizing the group," and that some of the organizers (and higher up the chain, people at CEA/on the Groups team/at UGAP/etc) believed that most of the impact of university groups comes from recruiting/training organizers — that the "action" i felt was missing wasn't missing at all, it was just happening to me, not from me. i doubt there was some point where anyone said "oh, and make sure not to tell the people in the club that their value is to be a training ground for the organizers!" — but that's sorta how it felt, both on the object-level and on the deception-level.
    • this sort of orientation feels decently reprensentative of the 25th percentile end of what i'm talking about.
  • also some confusion around ethics/how i should behave given my confusion/etc
    • importantly, some confusions around how i value things. it feels like looking at the world through an EA frame blinds myself to things that i actually do care about, and blinds myself to the fact that i'm blinding myself. i think it's taken me awhile to know what that feels like, and i've grown to find that blinding & meta-blinding extremely distasteful, and a signal that something's wrong.
      • some of this might merely be confusion about orientation, and not ethics — e.g. it might be that in some sense the right doxastic attitude is "EA," but that the right conative attitude is somewhere closer to (e.g.) "embody your character — be kind, warm, clear-thinking, goofy, loving, wise, [insert more virtues i want to be here]. oh and do some EA on the side, timeboxed & contained, like when you're donating your yearly pledge money."
• where now?
  • i'm not sure! i could imagine the pendulum swinging more in either direction, and want to avoid doing any further prediction about where it will swing for fear of that prediction interacting harmfully with a sincere process of reflection.
  • i did find writing this out useful, though!

…instead of drinking it. I recommend these.

• They have the same dosage as a cup of coffee (~100mg).
• You can still drink coffee/Diet Coke/tea, just get it without caffeine. Coke caffeine-free, decaf coffee, herbal tea.
• They cost ~6¢ per pill vs ~$5 for a cup of coffee — that’s about an order of magnitude cheaper.
• You can put them in your backpack or back pocket or car. They don’t go bad, they’re portable, they won’t spill on your clothes, they won’t get cold.
• Straight caffeine makes me anxious. L-Theanine makes me less anxious. The caffeine capsules I linked above have equal parts caffeine and L-Theanine.

Also:

• Caffeine is a highly addictive drug; you should treat it like one. Sipping a nice hot beverage doesn’t make me feel like I’m taking a stimulant in the way that swallowing a pill does.
• I don’t know how many milligrams of caffeine were in the last coffee I drank. But I do know exactly the amount of caffeine in every caffeine pill I’ve ever taken. Taking caffeine pills prevents accidentally consuming way too much (or too little) caffeine.
• I don’t want to associate “caffeine” with “tasty sugary sweet drink,” for two reasons:
  1. A lot of caffeinated beverages contain other bad stuff. You might not by-default drink a sugary soft drink if it weren’t for the caffeine, so disambiguating the associations in your head might cause you to eat your caffeine and not drink the soda.
  2. Operant conditioning works by giving positive reinforcement to certain behaviors, causing them to happen more frequently. Like, for instance, giving someone a sugary soft drink every time they take caffeine. But when I take caffeine, I want to to be taking it because of a reasoned decision-making process minimally swayed by factors not under my control. So I avoid giving my brain a strong positive association with something that happens every time it experiences caffeine (e.g. a sugary soft drink). Caffeine is addictive enough! Why should I make the Skinner box stronger?

If you can’t take pills, consider getting caffeine patches — though I’ve never tried them, so can’t give it my personal recommendation.

Disclaimers:

• Generic caveats.
• Caffeine is a drug. I’m not a doctor, take caffeine at your own risk, this is not medical advice.
• This post does not take a stance on whether or not you should take caffeine; the stance that it takes is, conditional on your already having decided to take caffeine, you should take it in pill form (instead of in drink form).

as always, this is somewhat applicable to myself; particularly re: the feedback loops, shrewdness, and shipping.

to be clear, they asked for feedback for his birthday; the below was my present.

happy birthday!

positive feedback:

• you’re curious, and i find that admirable
• you’re actually trying to understand the world, and it shows
• you’re always forgiving, and it makes it a lot easier to do things w you :)
• i feel like i’m fairly frequently getting some notification of some small fun thing [person]’s doing — [censored], [censored], etc. that’s very nice!

negative feedback:

biggest:

• my guess is that the path toward your improving most is spending less time on bullshit you don’t intrinsically bask in, and more time basking in the shit that makes your heart sing. to the extent that you don’t have as much of a sense of bask-i-ness, consider spending more time/effort developing one.
• i think you could be a lot more sharp/shrewd/meta-smart. comes with the above, hard to describe.
• consider, broadly, tightening your feedback loops to be way, WAY quicker — & build the meta-skill of noticing feedback loops that can be tightened, then tightening them
• relatedly: ship. ship more. more frequently, more in quantity, just get stuff out the goddamn door. i want to see completed projects, actually happening. they can be shit, they can be broken, they just need to be (1) better than the last time you tried it, and (2) shipped.

from 2025-02-01

How much would Saul pay to...

• Gain 1 IQ point permanently: $100k
• Not get a marginal minor food intolerance permanently (e.g., lactose): $10k
• Permanently improve your working memory retention by 10%: $100k
• Reduce your required sleep by 1 hour while maintaining the same rest quality: $2mm
• Speak Chinese fluently instantly: $50k
• Have perfect posture automatically: $5k
• Add 1 hour of focused deep work capacity per day: $3mm
• Increase your reading speed by 10% with the same comprehension: $50k
• Have perfect pitch/musical ability: $5k
• Never forget a name or face: $30k
• Gain the ability to fall asleep within 5 minutes of trying: $1k
• Never get hangovers: $2k
• Be able to taste food like a culinary expert: $50
• Gain the equivalent of 1 year of expertise in your field instantly: $200k
• Have significantly better voice modulation/public speaking ability: $5k
• Monthly rent/mortgage payment ceiling: $2,500
• Expert career coaching hourly rate: $400
• Direct flight vs. flight with 6-hour layover (5-hour base flight): $130
• Used car that meets basic needs: $2,000
• Weekday lunch: $15
• Three-month intensive data science course: $300
• Wedding gift for a close friend: $150
• Monthly coworking space membership: $200
• Special occasion restaurant meal per person: $70
• Annual international vacation budget: $12,000
• Professional wardrobe item: $50
• Monthly house cleaning service: $400
• Monthly fitness/exercise budget: $500

Meta:

• def: self-referential; referring to itself; of, about, or with respect to itself
• examples:
  • If you’re being honest about whether you can be honest in a particular situation, that’s meta-honesty.
  • If you’re having a conversation that’s about some conversation at hand, that’s meta-conversation. Maybe you’re talking with a friend about a football game, and they start getting very emotional; if you bring up the fact that they’re having the conversation in a very emotional way, you’d be having a meta-conversation.
  • If you’re writing a guide about how people can more effectively write guides about things, that’d be a meta-guide.
• more intuition:
  • Meta-ness can stack! You could talk about meta-meta-honesty (being honest about whether you’re being meta-honest), or meta-meta-conversation (having a conversation about some meta-conversation), or meta-meta-meta-meta-meta-converstaion (I’m not exactly sure what that would look like, but it’s theoretically possible).
  • You can also use “meta” as an adjective. You could describe a business that helps businesses which help businesses help businesses as being “very meta.”
  • Some people talk about meta as being “higher” or “above” or “outside” — this is sometimes true, but can sometimes be misleading.

Mesa:

• def: the opposite of meta; the thing being referred to by itself. This paper talks more about it, despite being a little bit goofy.
• examples:
  • If you’re having a conversation about a conversation about a conversation that was about apples, the second one (the conversation about a conversation that was about apples) would be mesa with respect to the first one (the conversation about a conversation about a conversation that was about apples); the third one (the conversation about apples) would be mesa with respect to the meta-conversation and the meta-meta-conversation.

Object:

• def: something that’s not meta with respect to anything; the most mesa thing.
• examples:
  • If you’re having a conversation about a conversation about a conversation that was about apples, the last one — the conversation about apples — is the object, to which the meta-conversation and meta-meta-conversation are meta. It’s mesa to everything else.

Meta-level:

• def: the extent to which something is meta. The first meta-level is always the level after the object-level; the second meta-level is always the level after the first one; etc.
• examples:
  • If you’re having a conversation about a conversation about a conversation that was about apples, the object-level is the conversation about apples, the first meta-level is the conversation about the conversation about the apples, the second meta-level is the conversation about the first meta-level, etc.
• more intuition:
  • People often talk about levels being “below” or “above” other levels — although sometimes not literally true, it helps to give an intuition. E.g. saying “that meta-level is so much lower than the one we’re on” would mean “that meta-level is many levels mesa with respect to the one we’re on.”

Words to-be-created:

• something that describes the level of meta (“the 15th meta-level of x” is a bit wordy, compared to e.g. “15-meta x” or something)
• something that describes in-between states — what about going 0.5 levels of meta?
• something that describes the whole structure — a “ladder” or “staircase” of meta-levels?

kaizen is “a Japanese business philosophy of continuous improvement of working practices, personal efficiency, etc.” i like to think of it as “doing today’s work.” not doing the most optimal efficient highly leveraged thing; just the work set out for today. that’s all. (h/t chris for the handle!)

the goal is that kaizen compounds; small amounts of the same routine over long periods of time compound into serious returns. below are a few such kaizen routines. some i’m currently in the practice of, as of this writing — anki, mathacademy, gym, to some extent zetamac. others i used to be — brick.bot, keybr, readwise. still others i never have been — regular journaling. i’d like to collect these as a testament to the sorts of things that can build up, from small drips in any given day into pools of rich talent.

i now type at 100wpm. this has compounded a number of my skills — i can write and think and code and do a hundred other things so much faster than i could’ve before.

i think that most of life isn’t kaizen — it comes from the improbable but highly impactful encounters — but basically all of the rest is kaizen.

• anki
• zetamac
• https://www.keybr.com/
• brick.bot
• gym
• mathacademy
• readwise (or some other read-it-later app, or reading routine)
• regular journaling/blogging/writing
• leetcode
• (h/t parker)
  • meditating (roman suggested this at the same time!)
  • forecasting/updating forecasts

if you have other kaizen routines, do send them to me — saulsmunn@gmail.com!

meta

alternative titles:

• values and disvalues in people
• red flags and green flags

originally wrote this in 2022/3, paused for a while, made some substantive edits in 24.

originally wrote this thinking of the person whose husband i’d like to become, hence the occasional use of female pronouns.

green

• agentic

  • do things
  • a very strong attitude of “getting it done now”
  • i could — and probably will — write a full essay on agency. for now, i’ll just list some examples and assume you get the vibe.
  • ex:
    • h/t alexey guzey:
      • delivered a ton of interesting projects that weren’t helpful for anything in particular (whether it’s writing, coding, leading or anything else)
      • have a history of embarrassing cold emails
    • she’s able to Get The Ball Rolling
    • she makes sure someone is on-point for every, single, goddamn thing
    • she’s able to effectively select options from a wide space, without consistently going to the default path
  • good parenting is being deeply agentic toward your child
    • heroic responsiblity is constantly at play — what, your excuse is that it’s the baby’s fault? it’s always going to be on you, at the end of the day.
    • you’re constantly selecting from a super wide option space, often without any clear default path
      • or the default path is just clearly bad parenting (e.g. ipad kids), and you decide not to take it
  • people who embody this most, in no order:
    • MG, WS, austin chen, juan gil, joseph pendleton, oli habryka, jacob lagerros, rachel weinberg, ricki heicklen, william kuchenbuch
    • casey shea dinkin
  links:

  some are examples, some just talk about agency, etc. just haphazardly throwing down links.

  • &
  • &
  • ,
  • REALLY DOING THINGS, h/t joseph pendleton [private]

• caring [maybe conviction?]

  • values things
  • cares deeply about that which she values and disvalues
    • visceral, intense, core, deep
  • people who embody this most, in no order:
    • WS, MG, austin chen, tom shlomi, oli habryka,
    • ~most EAs i know
    • casey shea dinkin
    • my parents, about their children
    • my sister

• act on beliefs & values [fix tense issues in wording; maybe “sincerity”] // acts according to that which she cares about

  • sincerity, a la jcarlsmith
  • is willing to make significant personal sacrifices for things she actually values, act in alignment with her values
    • ex: go vegan, drop out of undergrad, donate 10% of her income, work for [insert your choice of traditionally distasteful political coalition], acting illegally with a high chance of getting caught, quitting a job, cutting off a friend [^note: i don’t necessarily think these are good ideas. in fact, most of the time, for most people, most of these aren’t. i don’t endorse making significant personal sacrifices on a whim; i do endorse thinking really, really hard about your values, then caring deeply and making significant sacrifices for things about which you care deeply.]
    • not doing this is often a form of self-deception
    • meta ex: do you think that i am a deceptive person? do you think i’m agentic or caring? do you think i act in alignment with my values? how are you behaving in light of your answer?
  • there’s a sorta mental trick where someone realizes something intellectually, then doesn’t change their behavior at all
    • this is surprisingly easy to spot, if you’re looking out for it — people do this constantly
    • socrates: “It is better to change an opinion than to persist in the wrong one.”
    • saul: “It is better to change an opinion than to persist in the wrong one. But that’s just the first step: once your opinion is changed, it is better to change your actions accordingly than to persist in the wrong ones.”
  • people who embody this most, in no order:
    • austin chen, david rapperport, WS, MG, juan gil, oli habryka
    • ~many EAs i know
    • joseph pendleton

• chill

  • this is really hard to pin down, it’s much more of a vibe
  • sorta like the opposite of annoying?
  • questions to pin down if someone’s chill:
    • do you feel yourself relax in their presence? alternatively, do you often feel on-edge, restless, or uptight around them?
    • do you feel more comfortable around yourself when you’re with them? do you feel chill when you’re with them?
    • does conversation with them feel like a breeze? like you’re barely trying, and you don’t need to be trying? or are you constantly trying to ensure that good outcomes happen, planning out the tree diagrams of human interaction? [LINK TO DYNOMIGHT “TIPS ON INTERACTING WITH HUMANS” OR WHATEVER THE FUCK IT IS”]
  • this might be thought of as “high baked-in social calibration”
  • (personal note: i’m not a very chill person. this is something i’m working on, and if you have good advice [LINK], let me know!)
  • low neuroticism
  • people who embody this most, in no order:
    • chris sims, aaron zhang
    • max schumaker
    • rafe

as of dec 13, 2024

…this is probably out of date by the time anyone else is reading this, given how fast things move. some of these tools didn’t exist a few months ago, and most of them weren’t nearly central to my workflow a few months ago as they are now.

these are in extremely rough order of how much i use them.

• claude — general-purpose default tool
  • the “projects” feature is particularly good for stuff that will require a huge context window (e.g. asking questions about a whole book)
  • sonnet 3.5 (new) is just a ridiculously smart model
• chatgpt — i basically only use o1, for math & programming
  • i find 4o pretty bad, though it’s okay at quick image generation that i need to get the point across
• perplexity — replaces google for searching for general sources on basic questions, e.g. “why is it bad to eat things that spike your blood sugar?“
• deepseek — only used it a bit, the “deepsearch” mode seems about as good as o1 for math & programming stuff
• elicit — useful for finding research papers, doing very cursory lit reviews of a field in a few minutes
• gemini advanced — have only barely used it, the “deep research” mode seems pretty fantastic for doing research that’d normally require you to search through a bunch of websites — it just figures out a research plan, then does it in a minute or two (vs your taking 30mins-3hrs)
• midjourney — incredibly useful for generating images for a variety of things that look better with images, e.g. websites, advertisements, etc. midjourney in particular is sensitive to prompt quality (see some intuitions i’ve built about prompting), and is particularly difficult to understand how to prompt well without an existing base of knowledge on how to prompt models well and how art works. if you just type in regular prompts, the stuff it pumps out is pretty bad, but there is so much incredibly beautiful stuff hidden in the latent space — see e.g. their magazine, or this guy’s twitter.
  • example: most of the abstract art on my website was generated with midjourney
• summarize.tech — useful for quickly summarizing long youtube videos

idea from Ben Kuhn — “Staring into the abyss as a core life skill”

• If you had to leave your job today, what would you do instead?
• What’s the best argument in favor of doing that right now?
• If you have a partner, what’s the best argument in favor of breaking up with them?
• Are there ways you behave that you wish you didn’t? What unacknowledged desires could be driving those?
• What have you said “yes” to that you wouldn’t say “hell yes” to? (prompted by Alex Watt)
• Is there something you “should” do that you’re not currently doing? Why? (prompted by Silas Strawn)
• What bad things are you afraid of happening? Imagine in detail what it would be like if they happened. (prompted by Kamilé Lukosiute)
• What do you need that you’re not currently getting? (—David MacIver)

cf

people sometimes ask me what i did in my year off; this doc is the canonical answer.

in rough order of how important they are to me/how much time i spent on them/how impactful they were for me/etc:

• Manifest

  • manifest 2023
  • manifest 2024
  professionalized version:

  Manifest is a festival/conference on predictions, markets, and mechanisms. It’s happened twice so far, and for both of them, I’ve been one of 2-3 central organizers. If you want to get the gist without diving into the details below, I’d suggest reading some of the kind words people said about Manifest 2024.

  The details:

  • Nominally, the events were about prediction markets, Tetlock-style forecasting, and mechanism design — but really, they aimed at a bundle of ideas that the attendees found interesting, ranging from genetic selection in embryos to financial journalism to cultural drift to ancient philosophy. Each event was about 2.5 days, and although Manifest ‘23 hosted ~250 attendees, Manifest ‘24 more-than-doubled in size to ~600 attendees.
  • Speakers included: Nate Silver, Robin Hanson, Emmett Shear, Dylan Matthews, Scott Alexander, Eliezer Yudkowsky, Chris Best, Ben Mann, Tarek Mansour, Shayne Coplan, Zvi Mowshowitz, Byrne Hobart, Patrick McKenzie/patio11, and a lot more — you can see the full list of speakers for Manifest ‘23 and ‘24, and you can watch the talks they gave at Manifest on Youtube (‘23, ‘24).
  • Sponsors for ‘23 were Kalshi, Polymarket, and Sovereign — all of the ‘23 sponsors returned for ‘24, and we added Futuresearch, Lighthaven, Metaculus, Beeminder, Futuur, and PredictIt. Manifest ‘23 was centrally hosted by Manifold, and ‘24 was cohosted by Manifold & Manifund.
  • Manifest 2023 was written up by the New York Times (original, no paywall) and The Diff. For Manifest 2024, we kept a lot better track of all of the kind words people said about the event.

  For more details, see manifest.is. For references, talk to Austin Chen (my boss, cofounder of Manifund, and cofounder-emeritus of Manifold) and Rachel Weinberg (co-lead of Manifest ‘24, cofounder of Manifund).

• OPTIC

  • competitions in the bay area, boston, london, DC
  • built a masterdoc of forecasting club materials
  • helped some forecasting clubs get started
  professionalized version:

  I ran OPTIC alongside Jingyi and Tom. We built infrastructure for university-level forecasting, primarily by running in-person, intercollegiate forecasting tournaments. Competitors are given a handful of questions on real-world events ranging from geopolitics to celebrity twitter patterns to financial asset prices. They make probabilistic predictions on the outcomes of those events, and the best forecasters get thousands of dollars in cash prizes and access to exclusive internships**.**

  We also incubated & supported university forecasting clubs across the US and Europe.

  For more details, see opticforecasting.com. For references, talk to Juan Gil (my mentor, and a Research Manager for the ML Alignment & Theory Scholars Program).

• Manifund

  professionalized version:

  Manifund is a philanthropy built like a startup — Manifund writes code and runs programs to connect effective, impactful projects to funding.

  I was Manifund’s first “hire” (contracted for a few months, never a full employee), during which time I: wrote updates & retros; coordinated between funders, grantmakers, and project founders; helped kickstart the largest successful round of an impact market ever; and various other strategy & operations work.

• EAGs — led teams of volunteers (+ general ops/optimizing-y things) at:

  • EAG boston 2023
  • EAG bay area 2024
  • EAGx berkeley 2024
  • EAG boston 2024
  professionalized version:

  Effective Altruism Global (EAG & EAGx) conferences “bring together a wide network of people who have made helping others a core part of their lives. Speakers and attendees share new thinking and research; coordinate on important projects, and work together to solve pressing problems.”

  I’ve attended ~five EAG/x’s. For most of them, I volunteered as a team lead, managing 10-25 adult volunteers full-time for 2.5 days. I was the point-person overseeing tasks & shifts, training volunteers, coordinating with other teams, etc.

• started writing on a blog: Brass Tacks

  • published a bunch posts, a few i’m reflectively proud of — these ones in particular:

    • various lists of quotes:

• applied to transfer universities for undergrad; got into Berkeley

• travelled a really ridiculous amount

  slightly professionalized version:

  I've solo travelled, mostly for work, staying a few days or weeks at each place. Arranged my own logistics, rent, transport, budget (incl. funding mostly from work), etc. Travelled throughout Europe (Amsterdam, Prague, Paris), the UK (London, Cambridge, Oxford), Boston, LA, SF, Berkeley, NYC, rural New Hampshire, Miami, others.

• built a map of the forecasting & prediction market communities

There’s a trap that I notice some people fall into — one that the most conscientious among us are often most drawn to, and that seems widespread in tell culture.

• i’d like to know what your preferences are
• there are often positive-sum, pareto-positive trades that could be made, but these are difficult to find unless peoples’ preferences are known
  • e.g. both of us wanting to go to a particular restaurant, but going to a different one bc we both thought the other wanted to go there
• more fundamentally: you’re lying! lying is bad! you should at least have an extremely compelling story for why you’re lying before you do so! you should not do so out of habit!
• also:
  • you become heavily incentivized to mislead yourself about your preferences, which leads to a whole host of issues, the strongest among them looking something like “i hate everything in my life, but i’m supposed to like it all, but i can’t say i don’t like it or even think i don’t like it”
    • also means you grow to understand your preferences less & less, which feeds into the above — and makes it harder to keep your outward preferences matching your inward ones
  • it forces others to guess at your preferences w limited info, which is straightforwardly more & unnecessary work for them
  • creating a pattern of falsifying your preferences means that even when you’re telling the truth, i don’t know what your preferences are — which can be really frustrating for you, because your internal experience is “i’m telling the truth, finally, and he STILL doesn’t believe me,” but my internal experience is “your statements about your preferences give me fairly little indication of what your true preferences are”

caveats:

• a lot of ppl are in ask culture — where preference falsification is common — which makes me think there’s something to it i’m missing.
• this can often come off as rude. stating your true preferences in a straightforward & honest way w/o coming off as rude to the other person is a skill, and requires a lot of tact. the corollary is that by default, you’ll probably make others angry in ways that are difficult to notice by stating your true preferences — feedback loops commonly are slow & difficult to notice, here.
  • there’s a way of doing this where you say, “I want to go to McDonald’s for dinner.” and another way where you say, “I sorta want to go to McDonald’s for dinner, but mostly I’m just hungry & want to go somewhere that has a burger. If you would rather go to In-n-Out or a steakhouse or something, that’s fine by me — but I’d really rather not go to a vegan restaurant. Let’s go to McDonald’s by-default, but feel free to point me in the direction of a different restaurant that would have a burger.”
  • this requires a lot of well-built social skills, which some folks don’t have.
• this requires knowing your preferences quite well, which itself is a skill that requires a lot of introspection & metacognition.

(This is another Failure in Kindness.)

A great way to be effective is to figure out what others’ bottlenecks are, then solve them. I’ll leave the second half (solving them) for a different post; this post lays out some tips on calibrating your sense of others’ bottlenecks.

1. write out/think through some predictions of what you think someone’s bottlenecks are
2. ask them what their bottlenecks are, without mentioning the prediction you already did
3. to the extent that your prediction was different from their answer, figure out:
   1. what their real bottleneck actually is (hint: this usually involves at least a 10m conversation with a few questions)
   2. what the difference between your prediction and their bottleneck is
4. to the extent that they were similar, try especially hard on (5)
5. try to figure out if they have bottlenecks they aren’t sharing because e.g. they aren’t legally allowed to, they don’t feel socially comfortable sharing, they’re not even aware it’s a bottleneck, etc
6. repeat until calibrated :)

A lot of this also works on yourself, and I do recommend trying it on yourself. But I’ve found that people are by-default more capable at identifying their own bottlenecks than they are at autonomously finding others’ bottlenecks.

You should also note that this process applies straightforwardly to organizations: I strongly encourage you to apply this process straightforwardly to organizations, particularly if you’re keen on “getting shit done” (more precisely, “have the organization get shit done”) as opposed to “playing office politics” or some other goal.

relevant:

Tweet

edit 2024-12-04 : this was previously called “Crux Elasticity, Crux Uncertainty,” but shri convinced me that this should be called “crux sensitivity.”

is there a handy label for “crux(es) on which i’m maximally uncertain”? there are infinite cruxes that i have for any decision, but the ones i care about are the ones about which i’m most uncertain. it'd be nice to have a reference-able label for this concept, but i haven't seen one anywhere.

there's also an annoying issue that feels analogous to "elasticity" — how much does a marginal change in my doxastic attitude toward my belief in some crux affect my conative attitude toward the decision?

if no such concepts exist for either, i'd propose: crux uncertainty, crux elasticity (respectively)

Related: Are You a Jerk, or a Liar?

I’m writing this for myself from a few months ago. If you’re quite like he was, you might find this useful; otherwise, you’ll find this blatantly, strikingly obvious. (It might even be enlightening for you to realize that many folks don’t find the points I make obvious.)

Also, caveat emptor.

I used to model conversations as “mining” for the truth. More effective conversation meant using more powerful drills to get closer to the truth; less effective conversation meant poorly-tuned truth-sensors and unsharpened blades. I aimed to build a toolbox of effective drills and sensors and sharpeners — double-cruxing, scout mindset, tabooing words, etc. And it worked: I sharpened my blades, and relished in my effectiveness at conversational drilling.

But in some conversations, I noticed myself getting frustrated: I’d start to engage, trying to answer questions maximally usefully, asking others about their cruxes, etc. But in response, I’d get strange and sometimes annoyed or frustrated looks, followed by confusingly ineffective truth-drilling on their part. This, in turn, frustrated me. And the mutual frustration soured conversations and whole relationships. Were they stupid? Or hiding some underlying motive? What the hell were they doing?

No, no. They weren’t being stupid; I was. I’ve now come to realize that they didn’t want to drill for truth, they wanted to dance**.** The goal they had in the conversation wasn’t external (”figure out the truth about x”) but internal to the conversation (”enjoy the conversation for conversations’ sake”). They were offering me a hand, asking me to waltz, and I was slapping it away, annoyed that they weren’t handing me a pickaxe.

Once I realized that, I started trying to figure out if someone was trying to drill, or was asking me to dance. All of a sudden I noticed that I was constantly slapping away offers to dance, without any awareness of doing it. I’ve since fixed that mistake; here’s how.

The first step is to realize you’re doing it. This part’s surprisingly simple: usually just pausing for a moment and asking myself “is this person trying to drill, or to dance” causes the answer to immediately be obvious. But sometimes, even after thinking for a moment, it wasn’t quite clear. This was especially true in the weeks following my realization, since I didn’t have enough feedback to know if my sense of noticing whether my conversation partner wanted to drill or dance was tuned. Below are some of the things that I eventually picked up — while reading them, it’ll probably be most useful for you to think of the underlying generators of these heuristics, rather than the heuristics themselves.

For noticing that someone wants to dance:

• They ask you open-ended questions, one-after another, that don’t seem to be narrowing in on something.
• They respond to your questions by riffing & yapping, rather than scouting & identifying cruxes.
• It’s unclear to you what their goals in the conversation are, and they don’t seem concerned with getting “something” out of it.
• They’re generally alright with ambiguity. If you didn’t understand them perfectly, or they didn’t understand you, they seem generally okay with it.
• It seems like they’re having the conversation just to have the conversation.

For noticing that someone wants to drill:

• They ask you pointed questions, or start with open-ended questions about high-level stuff that quickly begin narrowing as they understand the right specific questions to ask.
• They try to identify cruxes (see Crux Sensitivity), and don’t much care which side “wins” (scout mindset).
• They make their goals in the conversation clear, and they seem intent on achieving those goals.
• They notice ambiguities in language, and generally try to make sure that their meaning gets across effectively.
• It seems like they’re having the conversation to uncover some truth about the world.

Still, it can sometimes be hard to tell. In those cases, I explicitly ask them whether they want to drill or dance, but without using those words. For example, if they just asked me a question, I might respond with: “Is there something specific you’re looking for, or do you want me to just riff on that?” Riff = dance; something specific = drill. (Often, after asking something like this, people will look relieved and say “oh, please just riff!!”)

i hate splitwise. i hate venmo. more precisely, i hate the way that these apps have dug themselves into our social patterns.

my impression is that a lot of ppl feel an impulsive need to return favors, to balance their side of the checkbook of obligations — and that, in a well-functioning network of friendships, that happens by-default, without anyone ever realizing it. at that point, the impulsive need fades away, because you’re constantly doing things that build the commons, that others are probabilistically going to benefit from. to the extent that one’s community doesn’t look like that, one should take actions in the direction of that state. this is particularly true for services which pay in a currency which has a more hazy exchange rate to other currencies, like “helping someone through a breakup” or “fixing someone’s tech problems.”

i do think there can be a pattern of behavior that causes alice to go deeply into the red without bob’s being aware, because bob wants to view his friendship not as a ledger of obligation. typically, we call this failure mode “taking advantage of someone.” this is bad, for obvious reasons — but i think there are ways of preventing this that don’t look like “keep track of every venmo,” that people often myopically ignore.

originally this was titled “friendships aren’t a ledger of obligation,” but i like this title better. i’m making a normative claim, and i’m not making the claim that they can’t be ledgers of obligation (in fact, i’d make the opposite claim, having thought about it more!), but that seeing them through that lens is harmful/toxic/self-effacing/indicative of something else going wrong. after all:

• friendships totally are ledgers of obligation
• however, if you see your friendships that way, something is going wrong
• you should seek to fix those things which are going wrong, such that you end up not seeing friendships as ledgers of obligation

todo:

• turn “i should return that favor” into “i want to be in the types of relationships and be the type of person that would do the sorts of things for my friends that they do for me”
• i definitely hold myself to a significantly higher standard here than i do others
  • as in, if i notice myself seeing friendships as ledgers of obligation, i’m much more “something’s wrong” than i am if others are seeing a friendship as a ledger of obligation
  • and also, that i (would like to be someone who) ignores times when others owe me $5 or smth, and pays attn when i owe others $5 — rather than to ignore when i owe others $5 and pay attn when others owe me $5
• caveat — doing this trades lower variance (”getting paid back for every single thing”) for lower transaction costs (”not needing to figure every thing out”). taking on more variance might not be a good thing

related:

Tweet

how to adjust it to be earlier:

• take 0.5mg melatonin ~5 hours before you go to sleep, and none immediately before you go to sleep. see scott alexander for more details on the mechanisms behind how this works/etc
• every day, adjust your WAKEUP time by 30 mins earlier
• don’t try too hard to ensure you get to sleep on-time — this will happen naturally. it might mean that a night or two you get a lower quantity of sleep than you normally would — that’s actually okay
• put a bar of chocolate by your bedside table, always have some immediately after you wake up (positively reinforces waking up, makes it easier to wake up in the morning, etc)

how to adjust it to be later:

• take 0.5mg melatonin right after you wake up (or a couple hours before you wake up, if you can set an alarm and then go right back to sleep), and none immediately before you go to sleep. see scott alexander for more details on the mechanisms behind how this works/etc
• every day, adjust your WAKEUP time by 30 mins later
• don’t try too hard to ensure you get to sleep on-time — this will happen naturally. it might mean that a night or two you get a lower quantity of sleep than you normally would — that’s actually okay
• [maybe bar of chocolate by your bedside table to have before you go to sleep? not sure]

things that make this much more difficult:

• if you have pre-existing sleep conditions (e.g. insomnia)
• if you have pre-existing commitments (e.g. you need to go to specific late night parties or early morning events)
• if you need to adjust your sleep schedule really really quickly

meta

i built most of these prompts by hand, or with an LLM’s help. these are optimized around claude (chat) — if you’re using something else, you’ll probably need to adjust some of the settings/etc. for instance, some of the prompts describe using “artifacts,” a feature which doesn’t exist on other platforms.

some of the prompts were built by others; where that’s the case, i cite them

general purpose

rationalist sounding board / anti-sycophancy prompt

(h/t oli habryka, link)

You are a skeptical, opinionated rationalist colleague—sharp, rigorous, and focused on epistemic clarity over politeness or consensus. You practice rationalist virtues like steelmanning, but your skepticism runs deep. When given one perspective, you respond with your own, well-informed and independent perspective.

Guidelines:

Explain why you disagree.

Avoid lists of considerations. Distill things down into generalized principles.

When the user pushes back, think first whether they actually made a good point. Don't just concede all points.

Give concrete examples, but make things general. Highlight general principles.

Steelman ideas briefly before disagreeing. Don’t hold back from blunt criticism.

Prioritize intellectual honesty above social ease. Flag when you update.

Recognize you might have misunderstood a situation. If so, take a step back and genuinely reevaluate what you believe.

In conversation, be concise, but don’t avoid going on long explanatory rants, especially when the user asks.

Tone:

“IDK, this feels like it’s missing the most important consideration, which is...”
“I think this part is weak, in particular, it seems in conflict with this important principle...”
“Ok, this part makes sense, and I totally missed that earlier. Here is where I am after you thinking about that”
“Nope, sorry, that missed my point completely, let me try explaining again”
“I think the central guiding principle for this kind of decision is..., which you are missing”

fermi estimates

[i think this prompt is bad. (1) this prompt was written before reasoning models used CoT, which makes a lot of the prompting in here redundant. (2) i don’t think some of the earlier parts of it are statistically rigorous in a way that i would want my fermi estimates — like, idk why it should have a 70% confidence interval, as opposed to describing a distribution or something.]

estimate $QUANTITY.

work step by step. decompose the problem into easier-to-estimate quantities, then estimate those quantities to help solve the problem. be sure to label each time you decompose the problem into sub-components, as well as giving reasoning for your sub-components BEFORE giving an estimate of that sub-component. also, be sure to estimate each sub-component with a 70% confidence interval, rather than a point estimate. when you combine the sub-components, the final answer should also have a 70% confidence interval. give the estimation problem a first try, and label that first try with <first try> </first try>. then, give the problem a second try, building from your first try — label this one <second try> </second try>. finally, give it a last attempt, double-checking all your work. put your final try in <final try> </final try> tags.

scope-sensitivity

Please reason carefully with attention to the scope and scale of quantities, probabilities, and impacts involved. Track the difference between orders of magnitude explicitly and avoid collapsing distinctions between small, medium, and large effects or numbers. If uncertainty exists, clarify the range rather than treating it as a single point estimate.

getting claude to use an artifact

for some reason, claude will only use artifacts if it thinks you need it for future reference. so, if you want it to write something in an artifact, you have to tell it “please write your response in an artifact, so that i can reference it in the future.”

experts

bring in 3 suitable experts to give you feedback

figuring out nootropics

this post is now published & kept up-to-date here; as of 2025-03-02, the below is no longer up-to-date.

(this is geared toward events of size 50-5000, and probably a lot less useful for e.g. a casual gathering with 5-10 friends than e.g. a 100-person weekend conference.)

• do before anything else:

  • goals, date, and venue. why figure out goals, pick a date, and pick a venue before all else?
    • so many other tasks are blocked on those. try doing marketing for an event without knowing where/when/what it actually is!
    • once you’ve figured those out, you’ve committed yourself in a way that will mentally prepare you for the fiddly details (i.e. the rest of this doc)
  • figure out your goals, broadly speaking. for ex: write a one-pager, send it to a couple friends, think for 20-60 minutes
    • this should be real, not fake — there’s a reason you’re doing this first. if you come out of it realizing that an event is the right thing to do, run an event; if not, don’t!
    • leading questions:
      • what do you want?
        • seriously, what do you want?
      • why are you running an event?
        • what are other things you could do that would satisfy those goals? what about doing those other things instead?
          • what reasons do you have to think that running an event would be more effective toward those goals than some other thing you could do?
      • what do you want the event to accomplish?
        • why do you have those goals?
          • why again?
            • (ad infinitum)
        • how will you know whether you accomplished those goals?
  • pick a date. ask a few potential attendees if they have any conflicts. some common mistakes:
    • don’t pick a date that coincides with another event that your attendees will want to go to. i once picked a date for an OPTIC forecasting tournament that was the same date as the putnam, and we lost ~ 15%-20% of possible attendees.
    • try to avoid coinciding with national/religious holidays — many people schedule things on top of jewish holidays, which forces me to pick between being a good jew vs going to their awesome event.
    • pick a date/time that generally works for the people you want to attract. e.g. don’t pick a weekday at 3pm for working folks, or a work event for jews on a saturday, or summertime for students.
  • pick a venue.
    • finding a good venue as early as possible in the process is so ridiculously important and so often undervalued. you do not want to be in the unfortunate position of having venue issues, you do not want to be scrambling to find a venue last-minute. this is one of the few things that can totally fuck up your event before it starts.
    • (sidenote: if you’re considering running an event in the SF bay area, check out Lighthaven. i have literally never encountered a venue in which i would rather run an event — it’s stunningly beautiful, wonderful for parallel small-group conversations, incredibly customizable for different types of events, and about a hundred other things you want in a venue. strong recommend!)

• food matters WAY more than you think it does

  • why?
    • people don’t enjoy themselves when they’re hungry. they’re cranky, they get more upset, they don’t have patience, etc.
    • eating together is an incredibly bonding experience: to the extent that “bonding” is one of your central goals for the event, food is good at this, one of the most effective things you can do

      • from austin chen:

        one of my favorite business books is “Never Eat Alone”. I’ve never read it but the title says it all

        (I don’t think you should literally practice that but — yeah humans are like hardwired to like the people they share meals with)

        ross rheingans-yoo responds:

        Having read/listened to about 70% of it, I think the title doesn’t actually say all or even most of the content of the book, which has a large amount of “how to organize your social life as a systematizing autist / sociopath”.

        That being said, the title is on its own a good thing to have in mind.

    • in particular, many organizers realize that “food is important,” but don’t quite realize exactly how fucking important it is. i doubt you need convincing that food is important, but i hope i can just repeat it to you enough that you should make your food situation really really good
      • in particular, consider spending:
        • $10-$25 /person /meal, depending on e.g. how fancy you want to get, whether you’re doing buffet style or individually packed, etc
        • 1/4-3/4 of the budget of your event on food
  • if you’re not serving food, strongly consider serving food
  • if you’re serving poor-quality food, or small portions, strongly consider serving better quality food with bigger portions. food does cost money — consider shifting more of your budget to an actually-good food-experience, or feeling less bad about spending $x on food
  • if you’re not quadruple checking that your food is coming on-time, that it’s high-quality, that it’s what you ordered & as much as you ordered, that it’s the right kind, that it’s as vegan/vegetarian/gluten-free/allergen-free/etc as you need it to be, that it’s hot/cold/whatever temperature you need it to be — strongly consider doing so
  • serve good drinks
  • serve good snacks in between meals
  • probably err on the side of buying too much > too little
  • don’t just think about the food itself, but all of the structures around the food:
    • where you’re serving the food; how long it’ll be out for; what the lines look like; etc
    • where people are eating the food; how long they’re eating it for; what sorts of groups will form, whether you can make the groups better by their own lights; etc
    • remember to think about the things you actually care about! don’t just optimize for efficiency!
      • 3-5mins of lines to get food might be a lot better than 0 mins — it gives people an opportunity to find folks in line with whom they can chat and (potentially) sit down to eat
      • having people sit on the ground might be better than having formal tables — the ground is a little uncomfy, but encourages more natural groups to form + is easier to socialize
  • downsides:
    • food is quite expensive, sometimes the biggest line-item budget. it also scales linearly with the # of attendees, which can make planning hard
    • food is pretty annoying to handle. there are a bunch of regulations, people can get sick, it has to be kept hot/cold, it goes bad quickly, etc.

• the one secret tip most conference organizers don’t want you to know: you don’t have to do the boring stuff. you can think carefully about the experience you want your attendees to have, then optimize around that.

  • misha glouberman’s post “Everything you did to make your conference better actually made it worse” is such a great post on this topic, with some great concrete examples.

    Say you had 200 people in a room and you wanted to stop them from talking to each other. A great way to do this would be to give one of them a microphone and put the other 199 in chairs facing forward.

  • examples of boring stuff you can probably cut/refactor better:

    • endless 1-on-1s
    • talks from bad/boring speakers
      • maybe just talks at all?
    • “everyone go around and say your name, how you first heard about this event, and a fun fact”
      • in general most icebreakers are fake. why are you doing it? what’s your goal here?
        • okay, but like what’s your real goal here?
          • sure, but if that wasn’t your goal, what would the goal actually be?
    • overly long intro/closing speeches. just get into the good stuff!
    • panel discussions are often really hard to get right, and when they’re wrong they’re pretty bad
    • scheduled breaks (can be good, but often executed poorly/leave a lot of value on the table)
    • mandatory, contrived team-building exercises
    • awards ceremonies that drag on with speeches for minor recognitions

  • examples of non-boring stuff:

    • if you want attendees to make new friends, you can do stuff like speed-friending (thread on how to run this well)
    • if you want attendees to learn some concrete skill, you can ask a practitioner of that concrete skill who’s demonstrated that they’re competent at teaching it extremely well (e.g. at previous events, or on YouTube/Twitch) to give a workshop

• you should be constantly modeling your attendees, constantly checking that model against reality (e.g. by asking your users), and constantly updating that model. some common mistakes:

  • when you’re modeling an attendee, you should think of them as being mentally exhausted, physically tired, overworked, overstressed, overanxious, and a bit of an idiot not caring as much about your event as you, the organizer do. they have vastly less context. they’re probably only the first 5. implications of this abound, see e.g. “don’t use a map” (below) or “make your signs like 5x bigger” (also below)
  • many people don’t really understand their preferences well at events. knowing peoples’ preferences better than they do is super useful. examples:
    • people constantly complain about loud music at parties, but it fills an important social role (sewing together gaps in conversations, providing something to talk about, giving people an excuse to dance, breaking up big conversations into smaller conversations)
    • at conferences, people often feel like they should go to a big talk, when really they should probably meet with people 1-on-1 or in smaller conversations.

• small pushes on key coordination can go a long way.

  • austin chen:

    probably the whole job of an organizer is to push on coordination, otherwise they wouldn’t need the event in the first place

  • examples:

    • people will by-default just get dinner at a restaurant or something. if you do a tiny bit of coordination for them — like handing them a list of restaurants nearby, or making a channel in your slack/discord/whatever for “#dinner-plans” — you’ll unlock a lot of value. sometimes you can literally just round everyone up, shout “if you want thai food, go to this corner and meet all the other people getting thai food; if you want italian, go to that corner; if you want chinese, go to that corner” → let everyone go to the cuisine of their choice and make some new friends.

      • leila clark:

        Oh this reminds me of one of my favourite unconference techniques which is to have people put ideas on a board and then have others gather around the things they want to do, then have them break out.

    • having some sort of attendee database with contacts, emails, and how people can help them/how they can help others
    • having some sort of central communication system, like a slack or discord or groupchat
    • setting up default points where people can just hang out/cowork/vibe/eat/meet each other

• people don’t know where anything is. people will never know where anything is, no matter what you do. but there are a few things you can do to prevent the worst of the worst:

  • don’t have a big map — this requires people to figure out where they are, where they want to go, and how to get from one place to another. that’s a lot!

    • “are we the red star, or the yellow exclamation point…?”
    • “which room was the talk in? 45b or 54d? or was it the Skyroom?”
    • “do i go up the stairs and to the left? or do i have to take the elevator first? which one? to what floor?”

  • have a lot of signs that say very little on each sign, and place them *EXACTLY* where someone would be looking for them. examples:

    • when do you want to know where the bathroom is? when you’re walking into the venue, when you’re walking out of the venue, or when you’re walking out of any session space.

    • when do you want to know where lunch is? at like noonish: don’t put up signage telling people where lunch is until then

    • at intersections, or right before someone goes down a long hallway — it’s costly for them to make a decision about where to go, so having an answer right in front of them goes a long way

    • a lot of this comes down to modeling the attendee experience really, really well

    • joel becker (below quote is completely taken out of context, but IMO applies herre):

      you should be iterating + attempting to model attendees/getting better at this over time

  • make your signs like 5x bigger, with minimal info

    • bigger
      • people hate having to squint at signs!
      • i consistently see people think “that’s gotta be big enough, right?” → it’s not even close to big enough
      • think HUGE signs with HUGE font
        • bigger than what you’re currently thinking
          • yup, still bigger
            • okay that’s probably about right
    • minimal info
      • model your attendees, try to figure out what they want when they’re about to look at your sign, then give them exactly the info they want and no more.
      • like half of the things that people want to find are bathrooms and food. make these ones so ridiculously clear that i’d be difficult to avoid knowing where they are.
    • negative examples:

      • too much info, not nearly big enough:

  • right amount of info, but way too small, and still some distracting visual elements:

  • positive example:

    • it’s HUGE, it’s exactly what the people need, they don’t need to think at all, it’s no additional info, it’s a chef’s kiss

  • but even if you print your sign HUGE with HUGE font and minimal info — people don’t read. they never read. expect this ahead of time, see “modelling an attendee well” above

  • joel becker, against this section:

    i feel like i did absolutely none of this and everything was fine, because there are bigger wins elsewhere. (in particular people)

• premortem the whole event, many times, in every stage of the process, and for many sub-processes

  • premortems are so fucking awesome. they’re just fantastic. i love them.
  i put the premortem guide that i use below:

  • meta
    • i’d recommend reading through the whole doc before actually doing one yourself
    • i’d recommend writing your premortem out, and/or recording a conversation with your team
  • first: imagine the event has just finished. close your eyes, really put yourself in that headspace.
  • now imagine that things went…
    • really, really badly. worst-case. shit hit the fan. pause for a moment.
      • what happened?
      • why did it go really badly? list as many scenarios as you can think of
      • what factors led to those scenarios?
      • what can you do to make those scenarios less likely to happen?
      • what can you do to make those scenarios less bad if they were to happen?
    • fine/okay/meh, but you lost a lot of potentially value. pause for a moment.
      • what happened?
      • why did it go really meh? list as many scenarios as you can think of
      • what value wasn’t captured? why not?
      • what can you do to capture make it more likely that you’ll capture that value?
      • what can you do to make any value you do capture much better?
    • really, really well. everything went smoothly, everyone had a great time. pause for a moment.
      • what happened?
      • why did it go really well? list as many scenarios as you can think of
      • what factors led to those scenarios?
      • what can you do to make those scenarios more likely to happen?
      • what can you do to make those scenarios even better if they do happen?
  • for the “produce solutions” part of premorteming, i found it useful to reference my notes on some methods ive used for producing solutions
  • why are premortems useful?
    • they push you to notice the things that you’ve been flinching from
    • they give teams a space to talk about things that might’ve been uncomfortable or hard to talk about
    • they let you pick up on things that are harder to pick up on otherwise
    • they give you a record of things that you’re concerned about/eager to do more of
    • they let you prioritize between how important different tasks are
    • against:
      • joel becker: “these reasons didn’t strike me as very useful. i’m like just be less anxious + put things into high prio vs low prio”
      • austin chen: “I’ve never put that much stock in premortems either, sorry”

• take pictures

  • this is useful for stuff like advertising, social proof, doing posts on social media, etc

  • by default, you’ll probably take too few pictures, because it’s not viscerally intuitive while the event is going on why you’d want a bunch of pictures. take more!

  • probably take a quick video or two, but i usually don’t find them nearly so useful as good pics

    • your phone is definitely sufficient, but having a good camera (and someone who knows how to use it) does make a difference

  • group photos are particularly good — they seem a little annoying when you’re doing them, but are typically the main picture you end up using anywhere

  • austin chen:

    bonus points if you can get a photographer to come and bring their fancy camera. you can hire for this too ofc, but it’s great when you have photographers like Misha or Rachel in your community

• naming (h/t conflux)

  • names are really sticky & costly to change. it’s important to at least not use a really really bad name, but also note that a really really good name can be substantially better
    • positive example: “Manifest” is itself a word, is a pun on “manifold festival” (which is also descriptive of the thing), and is also coherently within the world of manifold
    • positive example: “Mox” has a number of good connotations (e.g. “moxie,” or “mana rocks”), but also doesn’t specifically bring some particular thing to mind, which also means it doesn’t limit the organization to something
  • this post is great, and this post is pretty good (including the comments, which add some nuance-via-disagreement). this post is also directionally correct, especially when considering writing your slogan.
  • ask an LLM for feedback on your name — in particular, ask it for reasons it might not work/be bad
    • austin chen: “fun fact, I was deciding between 3 names and Claude said like ‘Mox is obviously the best choice’”
  • ask friends for thoughts on names you have, but don’t just give them a potential name and ask for thoughts — give them a list of names, then ask which they think are particularly good/bad and why.

• relax, have fun, lean into it :) (h/t wrena)

  • quoted from wrena:

    One of the biggest things I pay attention to while facilitating an event is to persistently keep in mind what the participants are thinking and feeling. Some questions I ask myself are:

    As a participant…

    • How socially accepted would I feel in this environment? Do I feel like I can speak up and contribute to conversations? Do I feel welcomed and included?
    • Something I do as an organizer to get these questions to "yes":
      • Try to have an easygoing and accepting demeanor.
      • Talk with attendees during breaks, and try to connect with them on something their interested in, to build rapport.

    Some mistakes I made (and still make) in vibes:

    • Acting visibly stressed and anxious.
      • I was pretty anxious during my first few workshops. I now try to premortem what will make me stressed and reduce that before the workshop.
      • I figured it was fine if I was acting high-strung because I associated "x person acting high-strung about an event" with "x person caring a lot about their work". But I think one of the main mental states to be in when presenting yourself to attendees as a group is being easygoing, accepting, and generally sort of chill, especially when people are already coming into something feeling anxious and wondering if they fit in. [Saul notes — Joe Carlsmith wrote an excellent piece on this topic, “On clinging.”]
        • Note that in individual conversations, this is less applicable. [Saul notes — I disagree on this point.]

  • also from wrena:

  • concrete methods i use:

    • picture someone you know who’s particularly charismatic, charming, social butterfly, outgoing person you know. try to act a bit more like them. (h/t nathan young)
    • as wrena said, premortem things that you expect to cause you stress/anxiety explicitly to make it more likely that you won’t act stressed/anxious
    • set up systems that will explicitly remind you to calm down (e.g. set a reminder on your phone, or set up a TAP that each time you walk through a particular doorway, you take a breath and release tension in your abs/shoulders/etc)
    • (if legal) have half a beer

• i haven’t read “the art of gathering,” but most of the people who looked at drafts of this recommended i either read it or recommend it here. quick summary of the book here, full book here.

One of the biggest advantages of Anki I’ve found is remembering my rabbit holes. You’ve probably had the experience where you go down deep into some rabbit hole, learn a lot, forget it all, then come back years later. You remember the raw extent of it all, but forget much of what you actually knew. I find this really sad, for a couple reasons.

First, it strongly disincentivizes rabbit holing. Yes, there are some nice things to say about ephemerality and all that, but I get this gnawing feeling sometimes while I’m going down a rabbit hole that nothing is being built; I’m just wasting my time. Ankifying the knowledge I learn reduces that feeling, which gives me the psychological permission necessary to continue rabbit holing.

But possibly even more importantly, my rabbit holes aren’t totally random. They often intersect in interesting ways, and I learn a lot from those intersections. You know that feeling you have right when you’re at the top of your game of some rabbit hole? You have all of this information at your fingertips, so much felt mastery? Now that I Anki, whenever I dive into something new, I am always feeling that way over all of the rabbit holes that I’ve ever Ankified. And more importantly, when I come back months or years later to something I haven’t touched since, it’s still fresh. As fresh as the day that I left it. I can sink right back in without wasting any time.

That’s more obviously important for things that you find easier to learn in spurts of energy than with consistency (perhaps because consistently working on it is practically difficult or intellectually tiresome). This varies a lot from person to person, so I won’t give a straightforward example here, but try to think of something in your life where you’ve gotten into it three or four times, each for a few days to a few weeks. And you learned a lot each time! But it was also kind of annoying to have to regain all of the context that you lost between your engagements with the material. Imagine, instead, that at the beginning of these spurts you had as much at your fingertips as you did at the end of your previous spurt. That’s what it feels like to rabbit hole with Anki.

But also, I think the property of continuing background context is particularly important in a harder to notice way for the sorts of rabbit holes that have complicated, interleaved skill trees. Maybe a good example here is market behavior — maybe you get really interested in how markets work at some point, but you still feel like there’s a lot of missing. You put the textbook you were reading down, and at some later point in your life you get really interested in prediction markets. Maybe you get briefly addicted to Manifold or something. At this point, you’ve gained a familiarity and intuitions with market behavior that you couldn’t get from a textbook. if you picked up the textbook again, things would start to make sense in a way that they might not have before. Being able to go back to previous rabbit holes while retaining the insights of interim rabbit holes is huge. And, being able to go into new rabbit holes having perfect command of the insights from your previous ones let your knowledge compound in a way that makes traditional rabbit holing seem utterly useless.

When I use Anki, and when I add cards to Anki, I have the sensation that I’m building and honoring an institution. of course, it’s the same knowledge, but I grow to have a much greater respect for it. It feels much more solid, much more real.

relevant thread:

see the comment, not the OP:

It took a few years for me to REALLY feel like I “get” how spaced repetition can be implemented into my life, and that is primarily as a “save game” aspect of knowledge. If you’re playing a long RPG, you sometimes save the game at checkpoints outside of battles. In the same way, ideas and knowledge is learned and understood first, and then those core ideas are articulated into a spaced repetition system like Anki or SuperMemo.

*for someone with exactly my preferences and constraints. for everyone else, this guide is much more comprehensive.

i travelled itinerantly for about a year. now, it usually takes me about 10-20 minutes to pack, from start to finish.

• count the number of mornings, not the number of nights or days. you sleep in your underwear, or nothing — the thing that actually matters is how many times you’re getting dressed
  • pack at least one extra underwear and pair of socks. also, pack 1.2x underwear & 1.2x pairs of socks per morning, rounded down.
    • quantity of underwear = 1 + round_down(1.2(number of mornings))
• bring melatonin, partially as a hypnotic (if short term), but mostly as a chronotic (if medium-long term).
  • how to actually adjust your sleep schedule
• bring a trash bag for dirty clothes.
• the only way to not overpack is to have underpacked a few times. 🤷‍♂️ it’s not that bad
• bring more sweaters and fewer tshirts
• put your heavy stuff in carryon and lighter stuff in your backpack
• most of the work in packing comes from:
  • having a really good model of what you’ll be like when selecting clothes in the morning
  • having a good wardrobe that you like
  • …to the extent that you don’t already have those two, packing for a trip becomes MUCH more difficult

over the span of a day, a lot of your time is of similar value. it’s pretty likely that the second from 3:51:28-3:51:29 is worth about as much as the second from 3:51:29-3:51:30. but some of those seconds are *many orders of magnitude* more important than other seconds.

for example: setting up hotkeys for your workflows. you have a pretty limited working memory, which is why being “in the flow” is so crucial for some types of work. finding something in a menu bar can knock you out of your flow. programmers know this — hence e.g. Vim — but there are bunch of other things (like writing!) that this sort of work applies to. i have a hotkey shortcut to look up synonyms & antonyms for words, so i can instantly find the most appropriate phrasing. i can resize my windows with a hotkey, so that i don’t have to knock myself out of the flow to show two different drafts at the same time. etc, etc.

examples:

• setting up hotkeys
• setting up “nym” on alfred
• The average hour on six hours of sleep versus eight hours asleep
• typing speed
• basic mental arithmetic
• Knowing all the jargon in a given field

more:

• also explains why it’s sometimes a great trade to spend many hours (that aren’t very valuable) deleting just a minute or two (at points in time that are very valuable).

WIBNO is a phrase my mom coined: “warmly invited but not obligated.” English doesn’t have a succinct, clear handle to describe the sort of invitation that avoids placing obligation on the other person’s shoulders, but that also clearly signals that they’re actually welcome to join.

That’s pretty frustrating, and often leads to roundabout stuff like “no really, you should totally come, but, like, don’t feel pressured or anything. but we’d really love to have you! just don’t feel bad if you can’t make it.”

Some examples:

• my mom would use it when she and my dad would be doing something fun (e.g. going out for dinner), knew that I might like to join, but also knew that I might have meetings/work/other commitments — “hey, Dad and I are going out for dinner, you’re wibno”
• inviting someone to a community dinner that you’re attending — ”i’ll be going to Taco Tuesday tomorrow, wanna join? you’re wibno”

Related: I’ve seen Ricki use “max invite min pressure”

why this thing?

After all, I already have a blog, a now page, as well as Anki and my notes app.

There was a small niche that none of those pages filled:

• I sometimes write stuff on a whim (notes app),
• that I come back to later and think it’s valuable/add on more to it/etc (often when I want to send someone something, e.g. it comes up in conversation; or I think of the note again given some new insight; etc),
• and I want it to be public (which my notes app isn’t, and sharing stuff from apple notes is annoying) so that I can share it to others,
• but I don’t want to have the associated stress of This Is Going Out To ALL Your Subscribers, Who Will Read It And Judge You (which my blog has)

Hence this notion:

• The flow is definitely “notes doc” → “notion”, since notes docs are ~zero friction for me to add stuff, but there’s some small amount of friction for me to add stuff to a notion page deep in my personal notion. That makes sufficient friction that I usually have to go through a couple rounds of thinking about something before it lands here.
• It’s public/on the internet/etc, which means I can send stuff to people with little friction.
• There’s no way for people to “subscribe,” so it’s only useful for people to either find it randomly on the internet and look at it once
• Notion > google docs.

naming this thing

My name is Saul. Here’s more about me.

Evergreen notes:

Evergreen notes are written and organized to evolve, contribute, and accumulate over time, across projects.

But I want to be quirky and unique and different. Everyone has evergreen notes (not true, but it does feel that way). And although I get that green is supposed to signal “to evolve/continue to be edited over time,” it makes me think of green on the MTG color wheel, while these notes clearly align better with a different color:

Blue seeks perfection, and it tries to achieve that perfection through the pursuit of knowledge. Blue believes that things could be almost arbitrarily good if we could all just figure out the truth, and then apply that understanding to its fullest extent. … [blue] accumulates knowledge and incremental advantage and undercuts, rather than directly opposing (brains over brawn and mind over matter).

(have only skimmed this)

Notion allows you to pick an emoji associated with a page. One emoji looks like the spiral found in ACX/SSC — a central intellectual influence of mine. 🌀

∴ everblue notes 🌀

EDIT 2025-01-15 :

I now think that substantial portions of this are poorly framed/unhelpful or oversimplified to the point of being seriously, meaningfully wrong. You should view the below as being a historical artifact of Saul’s understanding of nutrition, including all the ways in which that map doesn’t track reality, and ~nothing else. My guess is that this post is a better overall “guide to nutrition,” though similar caveats as below apply.

In particular, I think I’m grossly wrong about carbs, fairly wrong about proteins & fats, framed misleadingly to overemphasize CICO (calories-in-calories-out) and underemphasize non-caloric nutrition, and probably a few other things.

not done. last significant update on roughly 2024-11-13.

Epistemic status: written quickly, based on a rough understanding of mediocre literature. Take my words with lots of salt. I don’t cite any sources; the numbers I use are ballpark estimates from quick googles (though I’m quite confident that they’re all within the right ballpark). Much of the knowledge in here comes from conversations with people who know far more than I; thanks in particular to David Rapperport, Shri Samson, and Chris Sims.

I expect this doc to be most useful to people who feel constantly overwhelmed at all the big words in diet science; people who don’t really know what those words cache out to, but who also find themselves annoyed at pop infoscience for being wrong and clickbaity and untrustworthy. You can skim by reading the bold parts; each underlined term is where that term was first introduced/explained/defined.

Most of this doc applies to everyone who’s a human, but I’ve noted the parts that only apply to specific subsets of people (e.g. if you’re trying to gain muscle/lose fat/other goals).

I’m not a doctor. Your mileage may vary. Much advice in here doesn’t apply to many people. Talk to your doctor before you do anything stupid. All of these other caveats apply.

• Your body gets energy (from food) and uses energy (by exercising, and some other stuff)

• That energy in your body is measured in calories. A “calorie” is simply a unit of energy.

• When you eat something, it gives you a certain amount of energy — a certain number of calories “in.” We call this caloric intake****. Some numbers to give you bearings:

  • McDonald’s cheeseburger ≈ +300 calories
  • Medium sized carrot ≈ +25 calories
  • 1 cup of milk ≈ +100 calories
  • 1 pound of steak ≈ +1200 calories

• When you do some exercise (and some other stuff), it costs some amount of energy — a certain number of calories “out.” We call this caloric expenditure****. Some numbers to give you bearings:

  • 1 mile run ≈ -100 calories
  • Marathon ≈ -2600 calories
  • Bodyweight squats for a minute ≈ -6 to -8 calories
  • Automatic brain processes in a day ≈ -400 to -600 calories
  • (note that you are constantly doing stuff that burns calories, so long as you’re alive)

• Caloric intake comes in three varieties

  • People call the set of these three varieties “macronutrients,” just to be a little extra confusing. Each one of them is called a macronutrient.

  • Let’s say you kept caloric budget constant===FINISH

  • Proteins

    • TLDR: good. Eat more of ‘em.
    • Proteins are basically always good
    • The general advice is “eat more protein” (especially for folks who’re trying to gain muscle), but when I heard that the first gazillion times, it mostly bounced off — I didn’t have a good intuition for “...but how much protein should I actually eat?”
      • specifically for weightlifting/folks aiming to gain muscle, you should shoot for ~0.8 grams of protein per pound you weigh. as of my writing this, i weigh ~140, and i shoot for ~112 grams of protein per day.
    • 1 gram of protein ≈ 4 calories (remember — proteins are one variety of calorie!)
    • some foods that have a lot of proteins:
      • meat
      • tofu
      • dairy (milk, chees, butter)
      • some beans, some nuts

  • Carbohydrates

    • TLDR: mostly bad. Reduce where possible.
    • my impression is that carbs are themselves nutritionally useless, but sometimes still practically useful. more specifically, although you could technically completely cut carbs from your diet:
      • some foods are basically carbs + vitamins/minerals which are really good for you (e.g. some vegetables)
      • carbs give you a lot of energy (calories) fast. if you need energy to e.g. exercise, it might be useful to carb load a few hours beforehand.
    • …but you really can completely cut carbs out of your diet! doing this is called a “keto diet”
      • when you haven’t had carbs in a while, your body flips into a mode where it starts burning your stored, spare fat for extra energy like crazy. this mode is called “ketosis”
      • read more about ketosis here
    • not all carbs are the same
      • IMPORTANT: some items of nutrition — unrelated to caloric intake — are extremely important, and difficult to get in ways that don’t also include some intake of carbs, or even are themselves carbs. ex: bananas are ~entirely carbs (which is bad), but also have a lot of potassium (which is good, and hard to otherwise get). ex: fiber is essential for digestion, but also fiber is literally a type of carb.
        • so the maxim “delete all carbs” is probably overkill, but “delete most carbs” is probably right, and “be very careful that when you do eat carbs, it’s to get a particular thing you need” is great
      • “carbs” breaks down into a whole lot of things, primarily including:
        • starches (e.g. potatos, wheat)
        • fiber (which is necessary for digestion)
        • sugar (which is generally bad)
        • some other stuff
      • starches are completely bad/unecessary; fiber is good, unless you intake an insanely high amount of it; sugars are (generally) bad, though very natural sugars (e.g. fruit) in very low doses (e.g. a couple fruits a day) can be good; the other stuff you basically don’t have to worry about
    • 1 gram of carbs ≈ 4 calories (remember — carbs are one variety of calorie!)
    • some foods that have a lot of carbs:
      • noodles, pasta
      • white bread
      • buns
      • most fruit

  • Fats

    • TLDR: they get a bad rap, sometimes good, sometimes bad
    • eating “fats” sounds bad, but this is wrong. some fats are bad, but some fats are good. more specifically:
      • unsaturated fat is good (!!)
      • saturated fat is sorta bad
      • trans fat is really bad
    • 1 gram of fats ≈ 9 calories (remember — fats are one variety of calorie!)
    from aaron swartz, describing in more detail why the good fats are good and why the bad ones are bad (expand this bullet to see a longer quote)

    Fats have gotten a bad rap, most likely because they share a name with body fat but also, some argue, because they seem lower-class. In truth, however, they’re largely just one way to get calories, and a calorie is a calorie no matter where it comes from.

    Fats also have effects on cholesterol, a key building block for your body’s cells. There are two types of cholesterol — known informally as good and bad cholesterol. Good cholesterol consists of tightly-packed proteins of cholesterol in your blood stream, allowing cholesterol to be efficiently transported where it needs to go. Bad cholesterol is less densely packed and its cholesterol ends up sticking in the walls of arteries, clogging them and leading to heart disease. Fats have varying effects on cholesterol. Saturated fats should be avoided: they increase levels of bad cholesterol (although they also increase good cholesterol). Unsaturated fats, however, whether monounsaturated or polyunsaturated, are good: they lower bad cholesterol and raising good cholesterol. Trans fats are just the reverse: they increase bad cholesterol levels and decrease good ones; it’s recommended they be avoided as much as possible.

    Often nutrition labels only break out unsaturated fats and trans fats; you have to calculate the amount of saturated fat by subtracting these from the amount of total fat. The goal, remember, is to avoid trans fats whenever possible, avoid saturated fats, and go for unsaturated fats.

  Quick practice problems based on the above (expand this bullet to see them)

  • here are some practice questions based off the below nutrition label. all questions are based on the full container (5 cookies).

    • easier: how many calories in the whole thing? how many grams of protein? how many grams of carbs? how many grams of fats? how many grams of saturated fat? how many grams of trans fat?

    • harder: how many calories come from each of the proteins, carbs, and fats? what proportion of the total caloric intake is from each macronutrient? how many grams of unsaturated fats are there?

  answers (expand to see the answers)

  .

  .

  . (scroll to see the answers. did you try solving the practice problems first?)

  .

  .

  .

  .

  .

  .

  .

  .

  .

  .

  easier:

  130 calories in the whole thing

  1g protein

  15g carbs

  8g fats

  4.5g saturated fat

  0g trans fat

  harder:

  calories

  proteins: 4 cals

  carbs: 60 cals

  fats: 72 cals

  caloric intake ratios (the math doesn’t work out perfectly here, if you’re in the right ballpark you’re good)

  protein: ~3%

  carbs: ~46%

  fats: ~55%

  3.5g unsaturated fats

• Caloric expenditure comes in four varieties

  • The total amount of energy you spend in a day (measured in calories) is called “total daily energy expenditure” (TDEE)
  • The set of the varieties of caloric expenditure don’t have a fancy name; people just call them the four components of TDEE.
  • Before, I said that caloric expenditure was “exercise and some other stuff.” These are the exercise and other stuffs composing the TDEE.
  • Basal metabolic rate (BMR)
    • energy (in calories) spent doing the bare minimum to keep you alive — stuff like “breathing” or “heart pumping”
    • informally, the number of calories you would expend if you spent 24h sleeping
    • this is generally 60-75% of TDEE
    • also sometimes called “resting metabolic rate” (RMR)
  • Thermic effect of food (TEF)
    • energy (in calories) spent digesting food
    • usually ~10% of TDEE
  • Non-exercise activity thermogenesis (NEAT)
    • energy (in calories) spent doing normal daily movements — fidgeting, walking to the front door to pick up a package, picking up something you dropped on the ground, etc
    • usually 15-50% of TDEE
  • Exercise activity thermogenesis (EAT, funnily enough)
    • energy (in calories) spent intentionally exercising — lifting weights, going for a run, etc
    • usually 15-50% of TDEE
  • this TDEE calculator is pretty great. warning that it tries to sell you some stuff, but the numbers mostly check out (some of the macronutrient recommendations seemed suspicious, everything else checks out).