Truthseeking is the ground in which other principles grow


First they came for the epistemology/​we don’t know what happened after that.

I’m fairly antagonistic towards the author of that tweet, but it still resonates deep in my soul. Anything I want to do, anything I want to change, rests on having contact with reality. If I don’t have enough, I might as well be pushing buttons at random.

Unfortunately, there are a lot of forces pushing against having enough contact with reality. It’s a lot of work even when reality cooperates, many situations are adversarial, and even when they’re not entropy itself will constantly chip away at your knowledge base.

This is why I think constantly seeking contact with reality is the meta principle without which all (consequentialist) principles are meaningless. If you aren’t actively pursuing truthseeking, you won’t have enough contact with reality to make having goals a reasonable concept, much less achieving them. To me this feels intuitive, like saying air is necessary to live. But I’ve talked to many people who disagree, or who agree in the abstract but prioritize differently in the breach. This was supposed to be a grand post explaining that belief. In practice it’s mostly a bunch of pointers to facets of truthseeking and ideas for how to do better. My hope is that people can work backwards from these to the underlying principle, or flesh out their own relationship with truthseeking.

Target audience

I think these are good principles for almost any situation, but this essay is aimed at people within Effective Altruism. Most of the examples are from within EA and assume a certain amount of context. I definitely don’t give enough information to bring someone unfamiliar up to speed. I also assume at least a little consequentialism.

A note on examples and actions

I’m going to give lots of examples in this post. I think they make it easier to understand my point and to act on what agreement you have. It avoids the failure mode Scott Alexander discusses here, of getting everyone to agree with you by putting nothing at stake.

The downside of this is that it puts things at stake. I give at least 20 examples here, usually in less than a paragraph, using only publicly available information. That’s enough to guarantee that every person who reads this will find at least one example where I’m being really unfair or missing crucial information. I welcome corrections and arguments on anything I say here, but when evaluating the piece as a whole I ask that you consider the constraints I was working under.

Examples involving public writing are overrepresented. I wanted my examples to be as accessible as possible, and it’s hard to beat public writing for that. It even allows skimming. My hope is that readers will work backwards from the public examples to the core principle, which they can apply wherever is most important to them.

The same goes for the suggestions I give on how to pursue truthseeking. I don’t know your situation and don’t want to pretend I do. The suggestions are also biased towards writing, because I do that a lot.

I sent a draft of this post to every person or org with a negative mention, and most positive mentions.

Facets of truthseeking

No gods, no monsters, no epistemic daddies

When I joined EA I felt filled with clarity and purpose, at a level I hadn’t felt since I got rejected from grad school. A year later I learned about a promising-looking organization outside EA, and I felt angry. My beautiful clarity was broken and I had to go back to thinking. Not just regular thinking either (which I’d never stopped doing), but meta thinking about how to navigate multiple sources of information on the same topic.

For bonus points, the organization in question was J-PAL. I don’t know what the relationship was at the time, but at this point GiveWell uses their data, and both GiveWell and OpenPhil give them money. So J-PAL was completely compatible with my EA beliefs. I just didn’t like the idea that there might be other good sources I’d benefit from considering.

I feel extra dumb about this because I came to EA through developmental economics, so the existence of alternate sources was something I had to actively forget.

Other people have talked about this phenomenon from various angles, but it all feels tepid to me. Qiaochu Yuan’s thread on the search for epistemic daddies has some serious issues, but tepidness is not one of them.

Reading this makes me angry because of the things he so confidently gets wrong (always fun to have a dude clearly describe a phenomenon he is clearly experiencing as “mostly female”). But his wild swings enable him to cut deeper, in ways more polite descriptions can’t. And one of those deep cuts is that sometimes humans don’t just want sources of information to improve their own decision making, they want a grown-up to tell them what is right and when they’ve achieved it.

I won’t be giving examples for this facet beyond past-me and Qiaochu. I don’t feel good singling anyone else out as a negative example, and positive examples are called “just being normal”, which most people manage most of the time.


Delegate opinions instead of deferring

There is nothing wrong with outsourcing your judgment to someone with better judgment or more time. There are too many things you need to do to have contact with all of reality. I’d pay less for better car maintenance if I understood cars better. When I buy a laptop I give some goals to my friend who’s really into laptop design and he tells me what to buy and when, because he’s tracking when top manufacturers are changing chips and the chips’ relative performance and historical sales discounts and… That frees up time for me to do lit reviews other people can use to make better decisions themselves. And then my readers spend their newfound energy on, I don’t know, hopefully something good. It’s the circle of life.

But delegating your opinion is a skill. Some especially important aspects of that skill are:

  1. Be aware that you’re delegating, instead of pretending you came to the conclusion independently.

  2. Make that clear to others as well, to avoid incepting a consensus of an idea no one actually believes.

  3. Track who you’re delegating to, so you can notice if they change their opinion.

  4. The unit of deference is “a person, in a topic, while I’m no more than X surprised, and the importance is less than Y”

    1. I was very disappointed to learn people can be geniuses in one domain and raving idiots in another. Even within their domain they will get a few things critically wrong. So you need to be prepared to check their work when it’s particularly surprising or important.

  5. Keep track of how delegating to them works out or doesn’t, so you’re responding to their actual knowledge level and not the tone of their voice.

  6. Separate their factual judgment from emotional rewards for trusting them.

  7. Have multiple people you delegate to in a given area, especially if it’s important. This will catch gaps early.

  8. The person ultimately in control of your decisions is you. You can use other people’s opinions to influence your decisions to the exact degree you think is wise, but there is no escaping your responsibility for your own choices.

Stick to projects small enough for you to comprehend them

EA makes a very big push for working on The Most Important Problem. There are good reasons for that, but it comes at a high cost.

If you have your own model of why a problem is Most Important, you maintain the capability to update when you get new information. As you defer, you lose the ability to do that. How are you supposed to know what would change the mind of the leader you imprinted on? Maybe he already had this information. Maybe it’s not a crux. Or maybe this is a huge deal and he’d do a 180 if only he heard this. In the worst cases you end up stuck with no ability to update, or constantly updating to whichever opinion you last heard with a confident vibe.

You will also learn less pursuing projects when you’re deferring, for much the same reason. You’ve already broken the feedback loop from your own judgment, so how do you notice when things have gone too off track?

There are times this sacrifice is worth it. If you trust someone enough, track which parts of your model you are delegating, or pick a project in a settled enough area, you can save a lot of time not working everything out yourself. But don’t assume you’re in that situation without checking, and be alert to times you are wrong.

Seek and create information

I feel like everyone is pretty sold on this in the abstract, so I won’t belabor the point. I don’t even have real suggestions for actions to accomplish this, more categories of actions. But I couldn’t really make a whole essay on truthseeking without mentioning this.

Shout out to GiveDirectly, whose blog is full of posts on experiments they have run or are running. They also coordinate with academics to produce papers in academic journals. Points for both knowledge creation and knowledge sharing.

Additional shoutout to Anima International. AI used to have a campaign to end home carp slaughter in Poland. They don’t any more, because their research showed people replaced carp with higher-accumulated-suffering fish. I would take off points for the formal research being sparked by a chance news story rather than deliberate investigation, but I’d just have to give them back for the honest disclosure of that fact.


The world is very big and you can’t know everything. But if you’re not doing some deep reading every year, I question if EA is for you. For bonus points you can publicly share your questions and findings, which counts as contributing to the epistemic commons.

Make your feedback loops as short as possible (but no shorter).

I argued with every chapter of the Lean Start-Up book but damned if I didn’t think more experimentally and frontload failure points more after I finished it. This despite already knowing and agreeing with the core idea. The vibes are top notch.

Protect the epistemic commons

Some things are overtly anti-truthseeking. For example, lying.

But I don’t think that’s where most distortions come from, especially within EA. Mustache-twirling epistemic villains are rare. Far more common are people who know something and bias their own perception of reality, which they pass on to you.

E.g. a doctor knows his cancer drug works, and is distraught at the thought of people who will suffer if the FDA refuses to approve it. He’d never falsify data, but he might round down side effects and round up improvements in his mind. Or that doctor might have perfect epistemic virtue, but fails to convey this to his assistants, who perform those subtle shifts. He will end up even more convinced of his drugs’ impact because he doesn’t know the data has been altered.

If the doctor was deliberately lying while tracking the truth, he might discover the drug’s cost benefit ratio is too strong for even his tastes. But if he’s subtly and subconsciously suppressing information he won’t find out unless things go catastrophically wrong. At best the FDA will catch it after some number of unnecessary deaths, but if it’s subtle the falsehood may propagate indefinitely.

Or they might put up subtle barriers to others’ truthseeking. There are too many methods to possibly list here, so let’s talk about the one that most annoys me personally: citing works you will neither defend, nor change your views if they are discovered to be fundamentally flawed, but instead point to a new equally flawed source that supports your desired conclusion. This misleads readers who don’t check every source and is a huge time cost for readers who do.


Care less about intent and more about whether something brings you more or less contact with reality.

Some topics are inherently emotional and it’s anti-truthseeking to downplay that. But it’s also anti-epistemic to deliberately push others into a highly activated states that make it harder for them to think. This is one reason I hate the drowning child parable.

If you see something, say something. Or ask something. It’s easy to skip over posts you see substantial flaws in, and pushing back sometimes generates conflict that gets dismissed as drama. But as I talk more about in “Open sharing of information”, pushing back against truth-inhibiting behavior is a public service.

Sometimes saying something comes at great personal risk. One response to this is to do it anyway, whatever the cost. This is admirable (Nikolai Vavilov is my hero), but not something you can run a society on. The easier thing to do is get yourself in a position of lower risk. Build a savings cushion so you can afford to get fired. Hang out with friends that appreciate honesty even when it hurts. This lets you save the bravery for when nothing else can substitute.

Managers, you can help with the above by paying well, and by committing to generous severance no matter what terms the employee leaves on.

As a personal favor to me, only cite sources you actually believe in. They don’t have to be perfect, and it’s fine to not dump your entire evidence base in one post. All you have to do is disclose important flaws of your sources ahead of time, so people can make an accurate assessment. Or if it’s too much work to cite good sources, do even less work by explicitly noting your claim as an assumption you won’t be trying to prove. Those are both fine! We can’t possibly cite only perfect works, or prove an airtight case for everything we say. All I ask is that you don’t waste readers’ time with bad citations.

Sometimes it’s impossible to tell whether an individual statement is truthseeking. It’s a real public service to collect someone’s contradictory statements in public so people can see the bigger picture with less work. Ozzie Gooen’s recent post on Sam Altman and OpenAI is a good example. It would be better with sources, but not so much better I’d want to delay publication.

In most cases it’s anti-epistemic to argue with a post you haven’t read thoroughly. OTOH, some of the worst work protects itself by being too hard to follow. Sometimes you can work around this by asking questions.

You can also help by rewarding or supporting someone else’s efforts in truthseeking. This could be money, but there are very few shovel ready projects (I’ve offered Ozzie money to hire a contractor to find evidence for that post, although TBD if that works out). OTOH, there is an endless supply of epistemically virtuous posts that don’t get enough positive attention. Telling people you like their epistemic work is cheap to provide and often very valuable to them (I vastly prefer specifics over generalities, but I can’t speak for other people).

Contact with reality should (mostly) feel good

Some of the most truthseeking people I know are start-up founders asking for my opinion on their product. These people are absolutely hungry for complaints, they will push me to complain harder and soften less because politeness is slowing me down. The primary reason they act like this is because they have some goal they care about more than the social game. But it doesn’t hurt that it’s in private, they get a ton of social approval for acting like this, and the very act of asking for harsh criticism blunts the usual social implications of hearing it.

I think it’s fine not to act like this at all times in every area of your life- I certainly don’t. But it’s critical to notice when you are prioritizing social affirmation and accept what it implies about the importance of your nominal goal. If you object to that implication, if you think the goal is more important than social standing, that’s when you need to do the work to view criticism as a favor.


“Cultivate a love of the merely real” is not exactly an action but I can’t recommend it enough.

Sometimes people have trauma from being in anti-truthseeking environments and carry over behaviors that no longer serve them. Solving trauma is beyond the scope of this post, but I’ll note I have seen people improve their epistemics as they resolved trauma so include that in your calculations.

There are lots of ways to waste time on forecasting and bets. On the other hand, when I’m being properly strategic I feel happy when I lose a bet. It brings a sharp clarity I rarely get in my life. It reminds me of a wrong belief I made months ago and prompts me to reconsider the underlying models that generated it. In general I feel a lot of promise around forecasting but find it pretty costly; I look forward to improved knowledge tech that makes it easier.

I found the book Crucial Conversations life altering. It teaches the skills to emotionally regulate yourself, learn from people who are highly activated, make people feel heard so they calm down, and share your own views without activating them. Unlike NVC it’s focused entirely on your own actions.

Open sharing of information

This has multiple facets: putting in the work to share benign information, sharing negative information about oneself, and sharing negative information about others. These have high overlap but different kinds of costs.

The trait all three share is that the benefits mostly accrue to other people

Even the safest post takes time to write. Amy Labenz of CEA mentioned that posts like this one on EAG expenses take weeks to write, and given the low response her team is reducing investment in such posts. I’ll bet this 4-part series by Adam Zerner on his aborted start-up took even longer to write with less payoff for him.

Sharing negative information about yourself benefits others- either as by providing context to some other information, or because the information is in and of itself useful to other people. The downside is that people may overreact to the reveal, or react proportionately in ways you don’t like. Any retrospective is likely to include some of this (e.g. check out the comments on Adam’s), or at least open you up to the downsides.

For examples, see the Monday morning quarterbacking on Adam’s posts, picking on very normal founder issues, or my nutrition testing retrospective. The latter example was quite downvoted and yet a year later I still remember it (which is itself an example of admitting to flaws in public- I wish I was better at letting go. Whether or not it’s a virtue that I got angry on Adam’s behalf as well, when he wasn’t that bothered himself, is left as an exercise to the reader).

Publicly sharing information about others is prosocial because it gets the information to more people, and gives the target a clear opportunity to respond. But it rarely helps you much, and pisses the target off a lot. It may make other people more nervous being around you, even if they agree with you. E.g. an ingroup leader once told me that my post on MAPLE made them nervous around me. I can make a bunch of arguments why I think the danger to them was minimal, but the nervous system feels what it feels.

Criticizing others often involves exposing your own flaws. E.g. This post about shutting down the Lightcone co-working space, or Austin Chen’s post on leaving Manifold. Both discuss flaws in entities they helped create, which risks anger from the target and worsening their own reputation.

It is the nature of this facet that it is hard to give negative examples. But I think we can assume there are some departing OpenAI employees who would have said more, sooner if OpenAI hadn’t ransomed their equity.


Public retrospectives and write-ups

Spend a little more time writing up announcements, retrospectives, or questions from you or your org than feels justified. The impact might be bigger than you think, and not just for other people. Austin Chen of Manifund shared that his team often gets zero comments on a retrospective; and some time later a donor or job applicant cites it as the cause of their interest. Presumably more people find them valuable without telling Manifund.

Which brings up another way to help; express appreciation when people go through the work to share these write-ups. Ideally with specifics, not just vague gratitude. If a write-up ends up influencing you years later, let the author know. Speaking as an author who sometimes gets these, they mean the world to me.

Beware ratchet effects

Gretta is a grantmaker that works at Granty’s Grantmaking Foundation. She awards a grant to medium-size organization MSO.

Granty’s has some written policies, and Gretta has some guesses about the executives’ true preferences. She passes this on to fundraiser Fred at MSO. She’s worried about getting yelled at by her boss, so she applies a margin around their wishes for safety.

Fred passes on Gretta’s information to CEO Charlotte. Communication is imprecise, so he adds some additional restrictions for safety.

CEO Charlotte passes on this info to Manager Mike. She doesn’t need some middle manager ruining everything by saying something off-message in public, so she adds some additional restrictions for safety.

Manager Mike can tell Charlotte is nervous, so when he passes the rules down to his direct reports he adds on additional restrictions for safety.

By the time this reaches Employee Emma (or her contractor, Connor), so many safety margins have been applied that the rules have expanded beyond what anyone actually wanted.

New truths are weird

Weird means “sufficiently far from consensus descriptions of reality”. There’s no reason to believe we live in a time when consensus descriptions of reality are 100% accurate, and if you do believe that there’s no reason to be in a group that prides itself on doing things differently.

Moreover, even very good ideas in accord with consensus reality have very little alpha, because someone is already doing them to the limits of available tech. The actions with counterfactual impact are the ones people aren’t doing.

[You might argue that some intervention could be obvious when pointed out but no one has realized the power of the tech yet. I agree this is plausible, but in practice there are enough weirdos that these opportunities are taken before things get that far.]

Weirdness is hard to measure, and very sensitive to context. I think shrimp welfare started as a stunning example of openness to weirdness, but at this point it has (within EA) become something of a lapel pin. It signals that you are the kind of person who considers weird ideas, while not subjecting you to any of the risks of actually being weird because within EA that idea has been pretty normalized. This is the fate of all good weird ideas, and I congratulate them on the speedrun. If you would like to practice weirdness with this belief in particular, go outside the EA bubble.

On the negative side: I can make an argument for any given inclusion or exclusion on the 80,000 hours job board, but I’m certain the overall gestalt is too normal. When I look at the list, almost every entry is the kind of things that any liberal cultivator parent would be happy to be asked about at a dinner party. Almost all of the remaining (and most of the liberal-cultivator-approved) jobs are very core EA. I don’t know what jobs in particular are missing but I do not believe high impact jobs have this much overlap with liberal cultivator parent values.

To be clear, I’m using the abundance of positions at left-leaning institutions and near absence of conservative ones as an indicator that good roles are being left out. I would not be any happier if they had the reversed ratio of explicitly liberal to conservative roles, or if they had a 50:50 ratio of high status political roles without any weirdo low status ones.

High Decoupling, Yet High Contextualizing

High decoupling and high contextualizing/​low decoupling have a few definitions, none of which I feel happy with. Instead I’m going to give four and a half definitions: caricatures of how each side views itself and the other. There’s an extra half because contextualizing can mean both “bringing in more information” and “caring more about the implications”, and I view those pretty differently.

High decoupling (as seen by HD): I investigate questions in relative isolation because it’s more efficient.
Contextualizing (as seen by C): The world is very complicated and more context makes information more useful and more accurate.
HD (as seen by C): I want to ignore any facts that might make me look bad or inhibit my goal.
C-for-facts (as seen by HD): I will drown you in details until it’s impossible to progress
C-for-implications (as seen by HD): you’re not allowed to notice or say true things unless I like the implications.

My synthesis: the amount of context to attach to a particular fact/​question is going to be very dependent on the specific fact/​question and the place it is being discussed. It’s almost impossible to make a general rule here. But “this would have bad implications” is not an argument against a fact or question. Sometimes the world has implications we don’t like. But I do think that if additional true context will reduce false implications, it’s good to provide that, and the amount that is proper to provide does scale with the badness of potential misinterpretations. But this can become an infinite demand and it’s bad to impede progress too much.

Hope that clears things up.


Get good.

Willing to hurt people’s feelings (but not more than necessary)

Sometimes reality contains facets people don’t like. They’ll get mad at you just for sharing inconvenient facts with them. This is especially likely if you’re taking some action based on your perception of reality that hurts them personally. But it’s often good to share the truth anyway (especially if they push the issue into the public sphere), because people might make bad decisions out of misplaced trust in your false statements.

For example, many years ago CEA had a grantmaking initiative (this was before EA Funds). A lot of people were rejected and were told it was due to insufficient funds not project quality. CEA was dismayed when fewer people applied the next round, when they hadn’t even met their spending goal the last round.

In contrast, I once got a rejection letter from Survival and Flourishing Fund that went out of its way to say “you are not in the top n% of applicants, so we will not be giving further feedback”. This was exactly the push I needed to give up on a project I now believe wasn’t worthwhile.

To give CEA some credit, Eli Nathan has gotten quite assertive at articulating EAG admissions policies. I originally intended to use that comment as a negative example due to inconsistent messaging about space constraints, but the rest of it is skillfully harsh.

My favorite example of maintaining epistemics in the face of sadness is over on LessWrong. An author wrote a post complaining about rate limits (his title refers to bans, but the post only talks about rate limits). Several people (including me) stepped up to explain why the rate limiting was beneficial, and didn’t shy away from calling it a quality issue. Some people gave specific reasons they disliked the work of specific rate-limited authors. Some people advocated for the general policy of walled gardens, even if it’s painful to be kept outside them. I expect some of this was painful to read, but I don’t feel like anyone added any meanness. Some writers put very little work into softening, but everything I remember was clear and focused on relevant issues, with no attacks on character.


Multiple friends have recommended The Courage To Be Disliked as a book that builds the obvious skill. I haven’t read it myself but it sure sounds like the kind of thing that would be helpful.

To the extent you want to resolve this by building the skill of sharing harsh news kindly, I again recommended Crucial Conversations.


Deliberately creating good things is dependent on sufficient contact with reality. Contact with reality must be actively cultivated. There are many ways to pursue this; the right ones will vary by person and circumstance. But if I could two epistemic laws, they would be:

  • Trend towards more contact with reality, not less, however makes sense for you.

  • Acknowledge when you’re locally not doing that.

Related Work

Thanks to: Alex Gray, Milan Griffes, David Powers, Raymond Arnold, Justin Devan, Daniel Filan, Isabel Juniewicz, Lincoln Quirk, Amy Labenz, Lightspeed Grants, every person I discussed this with, and every person and org that responded to my emails.

Crossposted to EA Forum (98 points, 16 comments)