I don’t want to ruffle any feathers, but this has been bugging me for a while and has now become relevant to a decision since MIRI is fundraising and is focused on communication instead of research.
I love Eliezer’s writing - the insight, the wit, the subversion. Over the years though, I’ve seen many comments from him that I found off-putting. Some of them, I’ve since decided, are probably net positive and I just happen to be in a subgroup that they don’t work for (for example, I found Dying with Dignity discouraging, but saw enough comments that it had been helpful for people that I’ve changed my mind to think it was a net positive).
However, other comments are really difficult for me to rationalize. I just saw one recently on the EA forum to the effect that EAs who shortened their timelines only after chatGPT had the intelligence of a houseplant. I don’t have any model of social dynamics by which making that statement publicly is plausibly +EV.
When I see these public dunks/brags, I experience cognitive dissonance, because my model of Eliezer is someone who is intelligent, rational, and aiming at using at least their public communications to increase the chance that AI goes well. I’m confident that he must have considered this criticism before, and I’d expect him to arrive at a rational policy after consideration. And yet, I see that when I recommend “If Anyone Builds It”, people’s social opinions of Eliezer affect their willingness to read/consider it.
I searched LW, and if it has been discussed before it is buried in all the other mentions of Eliezer. My questions are:
1. Does anyone know if there is some strategy here, or some model for why these abrasive statements are actually +EV for AI Safety?
2. Does MIRI in its communication strategy consider affective impact?
Phrased differently, are there good reasons to believe that:
1. None of Eliezer’s public communication is -EV for AI Safety
2. Financial support of MIRI is likely to produce more consistently +EV communication than historically seen from Eliezer individually.
Note: I’ve intentionally not cited many examples here. I know that “abrasive” is subjective and am confident that many people don’t have the same reaction. None of this is intended to put down Eliezer, for whom I have great respect.
Somebody asked if people got credit for <30 year timelines posted in 2025. I replied that this only demonstrated more intelligence than a potted plant.
If you do not understand how this is drastically different from the thing you said I said, ask an LLM to explain it to you; they’re now okay at LSAT-style questions if provided sufficient context.
In reply to your larger question, being very polite about the house burning down wasn’t working. Possibly being less polite doesn’t work either, of course, but it takes less time. In any case, as several commenters have noted, the main plan is to have people who aren’t me do the talking to those sorts of audiences. As several other commenters have noted, there’s a plausible benefit to having one person say it straight. As further commenters have noted, I’m tired, so you don’t really have an option of continuing to hear from a polite Eliezer; I’d just stop talking instead.
It is a damn shame to hear of this tiredness, and I hope that your mood improves, somehow, somewhen, hopefully sooner than you expect.
This reply, though, I am forced to say, does not quite persuade, and to be totally frank even disappoints me a little. It was my understanding that both MIRI and to some extent one of your goals as of now was one of public outreach and communication (as a sub-goal for policy change) - this was at least how I understood this recent tweet describing what MIRI is doing and why people should donate to it, as well as other things you’ve been doing somewhat recently going on a bunch of podcasts and interviews and things of that nature (as well as smaller things such as separating out a ‘low-volume’ public persona account for reach as well as a shitpost-y side one).
Therefore, to put it maybe somewhat bluntly, I thought that thinking deeply and being maximally deliberate about what you communicate and how, and in particular how well it might move policy or the public, was, if not quite the whole idea, maybe a main goal or job of your organization and indeed your public persona. So, though of course a great many allowances are to be made when it comes to the tiredness, I don’t understand how to square the idea that
with what you and what MIRI are currently, well, for. You said, and I totally understand why, that the plan is to get other people to ‘take over’ that role, but this doesn’t really make it any less of a bad thing, just more of a hopefully temporary one. Is it truly such an all-or-nothing thing, that you’d just abandon all hope at such an important part of MIRI’s goals instead of trying to learn to be (as other commenters have put it) less abrasive?
I hope it is not too rude to hope that you come upon a different attitude when it comes to this at some point, because as it stands it seems kind of contradictory and self-defeating.
I’ve already tried, when I was younger and better able to learn and less tired. I have no reason to believe things go better on the 13th try.
I think it’s worth noting that I also have had times where I was impressed with your tact. The two examples that jump to mind are 1) a tweet where you gently questioned Nate Silver’s position that expressing probabilities as frequencies instead of percentages was net harmful, and 2) your “shut it all down” letter to NYT, especially the part where you talk about being positively surprised by the sanity of people outside the industry and the text about Nina losing a tooth. Both of those struck me as emotionally perceptive.
I accept your correction that I misquoted you. I paraphrased from memory and did miss real nuance. My bad.
Looking at the comment now, I do see that it has a score of −43 currently, and is the only negative karma comment on the post. So maybe a more interesting question is why I (and presumably several others) interpreted it as insult when logical content of “Intelligence(having <30y timeline in 2025) > Intelligence(potted plant)” doesn’t contain any direct insult. My best guess is that people are running informal inference on “do they think of me as lower status”, and any comparison to a lower intelligence entity is likely to trigger that. For instance, I actually find the thing you just said suggesting that I could have an LLM explain an LSAT-style question to me, to be insulting because it implies that you assign decent probability to my intelligence being lower than LLM or LSAT level. (Of course, I rank it less than “calling someone out publicly, even politely”, so I still feel vague social debt to you in this interaction.) I also anticipate that you might respond that you are justified in that assumption given that I seem to not have understood something an LLM could, and that that would only serve to increase the perceived status threat.
The “polite about the house burning” is something I have changed my mind about recently. I initially judged some of your stronger rhetoric as unhelpful because it didn’t help me personally, but have seen enough people say otherwise that I now lean toward that being the right call. The remaining confusion I have is over the instances where you take extra time to either raise your own status or lower someone else’s instead of keeping discussion focused on the object level. Maybe that’s simply because, like me, you sometimes just react to things. Maybe, as someone else suggested, its some sort of punishment strategy. If it is actually intentionally aimed at some goal, I’d be curious to know.
I’m sorry to hear about your health/fatigue. That’s a very unfortunate turn of events, for everyone really. I think your overall contribution is quite positive, so I would certainly vote that you keep talking rather than stop! If I got a vote on the matter, I’d also vote that you leave status out of conversations and play to your strength of explaining complicated concepts in a way that is very intuitive for others. In fact, as much as I had high hopes for your research prospects, I never directly experienced any of that—the thing that has directly impressed me, (and if I’m honest, the only reason I assume you’d also be great at research) has been the way you make new insights accessible through your public writing. So, consider this my vote for more of that.
It’s actually been this way the whole time. When I first met Eliezer 10 years ago at a decision theory workshop at Cambridge University, I asked him what his AI timelines were over lunch; he promptly blew a raspberry as his answer and then fell asleep.
My model of Eliezer thinks relatively carefully about most of his comms, but sometimes he gets triggered and says some things in ways that seem quite abrasive (like the linked EA Forum comment). I think this is a thing that somewhat inevitably happens when you are online a lot, and end up arguing with a lot of people who themselves are acting quite unreasonably.
Like, if you look at almost anyone who posts a lot online in contexts that aren’t purely technical discussion, they almost all end up frequently snapping back at people. This is true of Gwern, Zvi, Buck, and to a lesser degree even Scott Alexander if you look at a bunch of his older writing, and most recently I can see even Kelsey Piper who has historically been extremely measured end up snapping back at people on Twitter in ways that suggests to me a lot of underlying agitation. I also do this not-too-infrequently.
I feel pretty confused about the degree to which this is just a necessary part of having conversations on the internet, or to what degree this is a predictable way people make mistakes. I am currently tending towards the former, but it seems like a hard question I would like to think more about.
I dispute that I frequently snap at people. I just read over my last hundred or so LessWrong comments and I don’t think any of them are well characterized as snapping at someone. I definitely agree that I sometimes do this, but I think it’s a pretty small minority of things I post. I think Eliezer’s median level of obnoxious abrasive snappiness (in LessWrong comments over the last year) is about my 98th percentile.
I think your top-level answer on this very post is pretty well-characterized as snapping at someone, or at least part of the broader category of abrasiveness that this post is trying to point to (and I was also broadly pointing to in my comment).
I also think if you look at all of Eliezer’s writing he will very rarely snap at people. The vast majority of his public writing this year are in If Anyone Builds It and the associated appendices, which as far as I can tell contain zero snapping/abrasiveness/etc. My sense is also approximately zero of his media interviews on the book have contained this thing (though I am less confident of this, since I haven’t seen them all).
I don’t super want to litigate this, though happy to talk with you about this. I do think you are basically #2 in terms of people who do this in my mind who I am socially close to (substantially above everyone else except maybe Eliezer in that list, and I don’t know where I would place you relative to him). You do this much less in public, and much more in person and semi-public.
My intuition is that if our in-person conversations left a trail of searchable documentation similar to our internet comments, it would be at least similarly unflattering, even for very mild-mannered people.
(Unlike real life it’s more available to conscious choice to be mild-mannered all the time, if you set your offense-vs-say-something threshhold in a sufficiently mild-mannered direction. I doubt one can be sufficiently influential as a personality though without setting that threshold more aggressively, however. I haven’t gotten in a stupid fight on the internet in a long time (that I can recall; my memory may flatter me) but when I posted more, boy howdy did I.)
I think it’s important to distinguish irritation from insult. The internet is a stressful place to communicate. Being snappish and irritable is normal. And many people insult specific groups they disagree with, at least occasionally.
What sets Eliezer apart from Gwern, Scott Alexander and Zvi is that he insults his allies.
That is not a recipe for political success. I think it makes sense to question whether he’s well suited to the role of public communicator about AI safety issues, given this unusual personality trait of his.
Your conception of “allies” seems… flawed given the history here. I don’t super want to litigate this, but this feels like a particularly weak analysis.
You don’t think EA is an ally of the AI safety movement?
Eliezer definitely doesn’t think of it as an ally (or at least, not a good ally who he is appreciative of and wants to be on good terms with).
Yeah, that’s the problem. EA’s the most obvious community clearly invested and interested in the kind of AI safety issues Eliezer focuses on. There’s huge overlap between the AI safety and EA movement. To fail to recognize that, and carve time out of his day to compose naked, petty invective against EA over his disagreements, seems quite unpromising to me.
As a relevant point, he also writes things like this where he tries to reduce EAs unnecessarily beating themselves up. (I disagree with him on the facts, but I think it was a kind thing to do.)
I get why you read it as “kind.” But I have an alternative thesis:
Functionally, the essay erects a firewall between Eliezer and the FTX scandal.
While superficially “kind,” the essay is fundamentally infantilizing, absolving the community while denying them agency. This infantilization is crucial to building the firewall.
If you’re interested, I can expand on this.
Edit: Clarifying changes, especially to emphasize that I interpret the essay as containing motivated reasoning and self-interested spin, not that Eliezer is lying.
I’m not interested in making such a request for expanding on it, thanks for the offer. (I’m not asking you not to, to be clear.)
To respond to your point, you may be aware that there’s a large class of Singerian EAs that are pathologically self-guilting and taking-personal-responsibility-for-the-bad-things-in-the-world, and it was kind to some of them to point out what was believed to be a true argument for why that was not the case here. I don’t think it is primarily explained by self-serving motivation; and as evidence you can see from the comments that Eliezer was perfectly open to evidence he was mistaken (via encouraging Habryka to post their chat publicly where Habryka gave counterevidence), so I think it’s unfair to read poor intent into this, as opposed to genuine empathy/sympathy for people who are renowned for beating themselves up about things in the world that they are barely responsible for and have relatively little agency over.
I don’t see evidence in the post comments that it was received that way, though it’s possible those who read it as a true, helpful and kind didn’t respond, or did elsewhere.
I don’t think he’s a schemer or engaging in some kind of systematic project to silence dissent.
What do you mean by “ally” (in this context)?
Institutional support, funding, positive and persistent community interest, dialog, support, and professional participation. Examples:
Open Phil
FTX Future Fund (extremely bad allyship, but still was regarded as allyship until it went down in flames)
80,000 hours, MATS
MIRI has been heavily supported by EA donors
Anthropic safety influences
FHI (now closed) and CSER gave AI safety intellectual credibility and were staffed and funded by EAs
Take the above as my beliefs and understanding based on years of interaction, but no systematic up-to-date investigation.
I think that statement is tricky (the AI Safety Movement is not a monolithic entity, and neither is EA). It seems more clear that most of EA is not much of an ally of Eliezer.
No collective entity is a monolith.
If it wasn’t obvious, I meant the term “ally” not in the sense of a formally codified relationaship, but to point out the uniquely high level of affinity, overlap, and shared concerns between the AI safety movement and EA.
There is a reason I said “ally,” rather than literally identifying EA as part of the AI safety movement or vice versa.
Yep, I am not trying to insist on a particularly narrow definition of “ally”.
I think that snapping back at people is most likely caused by belief that the person at which one snapped did something clearly stupid or hasn’t bothered to do a basic search of related literature.
Does it help?
A bunch of points that are kind of the same point:
Not suffering fools gladly is a common way to punish people for wasting your time or strategically misunderstanding you. Expecting that you will lose status for doing this to someone is an incentive to, if you’re going to engage, actually engage with what they said, rather than to misread them. (It’s also an incentive not to engage with them.)
Similarly it’s a common dynamic that powerful people will pretend to like people who are interesting but don’t cost them political points to endorse, while kind of ignoring them, but once they’re actually costing you politically then the pretense will go away, and I do think this pretense is bad for actually engaging with ideas and positions (e.g. I think many religious people might have said that Richard Dawkins was a polite debate partner if he stuck to just saying that, and didn’t have more fiery rhetoric that was causing a movement to build, even though his position didn’t change).
I think Eliezer has said he took a more polite and hopeful tone in the past (e.g. this and this and this versus this and this and this), and felt that people basically didn’t engage with him and did whatever they were going to do anyway. So he thought he should try something different, and it indeed has been more successful at causing people to engage with the perspective that AI is by-default going to be an extinction-level bad event.
I think that broadly, Eliezer is more hostile due to feeling like he’s in a hostile epistemic environment. Personally I take this as a cue to more seriously consider that the hostility is appropriate to the world we find ourselves in (which also is a world where people have seemingly had their head in the sand for over a decade and very powerful people are racing to build doomsday machines while largely not openly engaging with this fact nor the arguments that this is an atrocity).
Some other factors that are relevant:
I think it’s accurate to model Eliezer as having a chronic illness to do with fatigue, which means he just doesn’t have a lot of time/effort to put into lots of public dialogues. This is a factor that makes him less invest in back-and-forths and understanding others.
Eliezer’s internal thoughts are also in a pretty different language to others, which can lead to miscommunications. I think in the example you give, with the potted plant, the questioner kind of indicated that they hadn’t clearly read the post, where explicitly says he’s asking about pre-ChatGPT, and the person asks whether a statement made today would count; if you assume that they did read the post then his reply makes more sense (it wasn’t intended primarily as an insult, as much as an attempt to to give a short answer to a somewhat pointless and irrelevant question).
To be clear I think he could do a better job of understanding people he’s writing with via text format, and I am still confused about why he seems (to me) below average at this.
I’m not sure if his approach is actually productive for this, but for the longest time, the standard response to Eliezer’s concerns was that they’re crazy sci-fi. Now that they’re not crazy sci-fi, the response is that they’re obvious. Constantly reminding people that his crazy predictions were right (and everyone else was wrong in predictable ways) is a strategy to get people to actually take his future predictions seriously (even though they’re obviously crazy sci-fi).
I think this makes sense as a model of where he is coming from. As a strategy, my understanding of social dynamics is that “I told you so” makes it harder, not easier, for people to change their minds and agree with you going forward.
(Again, not trying to excuse pointlessly being a dick. Plausibly Eliezer is not infrequently a big pointless dick, I do not know, no strong opinion.)
Another hypothesis: It’s possible that he thinks some people should be treated with public contempt.
As an intuition pump for how it might be hypothetically possible that someone should be treated with public contempt, consider a car salesman who defrauds desperate people. He just straightforwardly lies about the quality of the cars he sells; he picks vulnerable people desperate for a cheap way to juggle too many transport needs; he charms them, burning goodwill. He has been confronted about this, and just pettily attacks accusers, or if necessary moves towns. He has no excuse, he’s not desperate himself, he just likes making money.
How should you treat him? Plausibly contempt is correct—or rather, contempt in reference to anything to do with his car sales business. IDK. Maybe you can think of a better response, but contempt does seem to serve some kind of function here: a very strong signal of “this stuff is just awful; anyone who learns much about it much will agree; join in on contempt for this stuff; this way people will know to avoid this stuff”.
When Eliezer says that something is just awful, I interpret that to mean “this is just awful.” Period. I’m sure that from long experience he has observed that in fact very few people who learn much about it will agree, because they are not able to follow his reasoning, let alone derive his conclusions for themselves. I also doubt he would find it useful to have a crowd of potplants dogpiling on the target in imitation of his excoriation but without his understanding.
The world needs all types of activism: from the firebrands to the bridge-builders. I too find his tone to be abrasive at times. He can be self-aggrandizing, pompous, and downright insulting. In my experience this is not uncommon for people (most typically men) who believe they’re the smartest person in the room.
Personally, I try to live by, “first, be kind.” I’ve found the most success with leading with empathy, but it’s not effective in every circumstance. Some people you can reach better by showing them our commonalities. Some people need to be shocked into thinking about what the implications of their beliefs are. Sometimes a sit-in is effective in bringing about needed change, other times it takes a riot.
(This is not a defense of poor behavior; people are responsible for not pointlessly being dicks.) A hypothesis I keep in mind which might explain some instances of this would be The Bughouse Effect.
To give this hypothesis a bit more color, I think people get invested in hope. Often, hope is predicated on a guess / leap of faith. It takes the form: “Maybe we [some group of people] are on the same page enough that we could hunt the same stag; I will proceed as though this is the case.”.
By investing in hope that way, you are opening up ports in your mind/soul, and plugging other people into those ports. It hurts extra when they don’t live up to the supposed shared hope.
An added wrinkle is that the decision to invest in hope like this, is often not cleanly separated out mentally. You don’t easily, cleanly separate out your guesses about other people, your wishes, your plans, your “just try things and find out” gambles, and so on. Instead, you do a more compressed thing, which often works well. It bundles up several of these elements (plans, hopes, expectations, action-stances, etc.) into one stance. (Compare: https://www.lesswrong.com/posts/isSBwfgRY6zD6mycc/eliezer-s-unteachable-methods-of-sanity?commentId=Hhti6oNe3uk8weiFL and https://tsvibt.blogspot.com/2025/11/constructing-and-coordinating-around.html#flattening-levels-of-recursive-knowledge-into-base-level-percepts ) It’s not desirable, in the sense that any specific instance would probably be better to eventually factor out; but that can take a lot of effort, and it’s often worth it to do the bundled thing compared to doing nothing at all (e.g. never taking a chance on any hopes), and it might be that, even in theory, you always have some of this “mixed-up-ness”.
Because of this added wrinkle, doing better is not just a matter of easily learning to invest appropriately and not getting mad. In other words, you might literally not know how to both act in accordance with having hope in things you care about, while also not getting hurt when the hope-plans get messed up—such as by others being unreliable allies. It’s not an available action. Maybe.
An author is not their viewpoint characters. Nevertheless, there is often some bleed-through. I suggest you read Harry Potter and the Methods of Rationality. [Warning: this is long and rather engrossing, so don’t start it when you are currently short of time.] I think you may conclude that Eliezer has considered this question at some length, and then written a book that discusses it (among a variety of other things). Notably, his hero has rather fewer, and some different, close friends that J.K. Rowling’s does.
This ACX post, Your Incentives Are Not The Same, explains it adequately. He is prominent enough that what might be obviously negative-value behavior for you might not be for him.
I’m not aware of good reason to believe 1. 2 seems likely; MIRI has a number of different people working on its public communications, which I would expect to produce more conservative decisions than Eliezer alone, and which means that some of its communications will likely be authored by people less inclined to abrasiveness. (Also, I have the feeling that Eliezer’s abrasive comments are often made in his personal capacity rather than qua MIRI representative, which I think makes them weaker evidence about the org).
As far as I understand, Eliezer is abrasive for these reasons:
If a misaligned ASI is created without an aligned counterpart, the ASI will have no reason to keep us alive.
Mankind isn’t on track to solve alignment BEFORE creating the ASI.
Normally, THAT high stakes would prompt anyone to think twice of anything related to the issue. But Eliezer has seen things like many obviously stupid approaches to alignment[1], a high-level official at GDM building a model which predicts that mankind will be left with jobs[2] after the ASI’s rise or OpenPhil (seemingly?) basing its position about AGI timelines on the approach of Ajeya Cotra[3]
As evidenced by him claiming that an approach is “Not obviously stupid on a very quick skim” and congratulating the author on eliciting a THAT positive review. Alas, I also have seen obviously stupid alignment-related ideas make their way at least to LessWrong.
However, it would be possible if the ASIs required OOMs more resources per token than humans. In this case applying the ASIs would be too expensive. Alas, this is unlikely.
IMO Eliezer also believes that the entire approach is totally useless. However, a case against this idea is found in comments mentioning Kokotajlo (e.g. mine)