I’m very happy to read this. I feel lucky that Lisa works at MIRI, and I feel major gratitude to her and everyone else who contributed behind the scenes.
KurtB
Karma: 369
I have some replies to Nate’s reply.
Overview:
I’m not asking anyone to modify their personality at all. I mainly wish I had been warned about what was in store for me when I joined MIRI, and I want such warnings to be a visible part of Nate’s reputation.
I feel some pressure to match Nate’s conciliatory tone, but something feels incongruous about it. I’m concerned that people will read Nate’s calm, kindly replies and come away with the wrong idea of how he presented himself at MIRI.
I find Nate’s additional context to be, well…missing some important context. See below...
More context and recollections
It’s true that I didn’t report directly to Nate, and there could be a reasonable expectation that I refrain from bothering him without at least talking to my manager first. My memory is that this was a practical emergence, and not an explicit rule. Regardless, it seemed that Nate was sort of having it both ways, because he did in fact sometimes directly ask me questions (while quite angry), for example why we had ordered lunch from a restaurant he didn’t like, or where the soy sauce was. I now have to wonder what would have happened if I had refused to answer those angrily posed questions on the grounds that I didn’t report directly to him. My guess is that I probably would have lost the job shortly thereafter (and been happier for having held my boundaries—such a story would have been the labor market functioning even more efficiently).
I told the story about how he got very angry when I didn’t inflate his tires properly and one went flat during his commute. He added mitigating context claiming that he was rushed and sweaty having just replaced the tire, and then felt his own (implicit) boundaries violated by my approach.
Well, my memory is that I had his tire professionally replaced. Maybe I misremember that detail. But I am quite sure I took his bike to the shop for some kind of repair in the immediate wake of the flat tire. After that, I messaged him to let him know the repairs were finished and his bike was back on the office bike rack, ready to use. It wasn’t necessary to send him that message, but I wanted to do the small professional courtesy as a gesture of respect and de-escalation. I regret that, because he didn’t reply at all.
In the next day or so, we ended up taking the elevator at the same time. We stood in cold silence. I got no signal that he felt anything other than annoyance and disregard. I would have accepted an apology or just some kind of thawing, but I sure as hell wasn’t going to risk initiating another conversation he didn’t want. When he says he has no recollection of wanting me to quit (as opposed to improve), I feel frustrated because I think this will give people a distorted mental image of what his outward demeanor was like.
Alternate timelines that I would have preferred
Nate says he feels trapped:
I have a sense that there’s some sort of trap for people with my emotional makeup here. If you stay and try to express yourself despite experiencing strong feelings of frustration, you’re “almost yelling”. If you leave because you’re feeling a bunch of frustration and people say they don’t like talking to you while you’re feeling a bunch of frustration, you’re “storming out”.
Perhaps I’m missing some obvious third alternative here, that can be practically run while experiencing a bunch of frustration or exasperation. (If you know of one, I’d love to hear it.)
Here are six alternatives, ranked by my preference. I personally think none of them seem impossible and all of them are reasonable to expect from an agentic, cooperative EA leader. Each scenario is meant to be imagined independently of the others.
In 2014 Nate writes his blog and crafts his public image in such a way that by the time I am recruited to work for MIRI, I have gotten a full and realistic sense of his emotional build, and quickly decline the job offer from MIRI (or at least bargain for modifications to the job description). This probably would have required placing several prominent disclaimers throughout the blog, as well as changing the tone and implication of many of the posts. Nate accepts the reputational costs that this modified narrative requires.
MIRI makes sure their ops recruitment checklist includes giving explicit warnings about Nate’s emotions, and I quickly decline the job offer with no hard feelings. This one seems the most realistic to me.
Nate firms up the organizational separation. Before having any angry outbursts, it is made highly explicit that I should not bring up e.g. bike tire questions with Nate, and furthermore that Nate should not e.g. ask me to explain perceived lunch errors. Those things are to be resolved with my manager first if possible. I am warned that he may not be able to calmly tolerate violations of these boundaries. It is acknowledged that this is kind of weird, but MIRI is a weird org trying to save the world.
If he had done this, I probably would have felt uneasy, still lost some respect for him, and told some friends about it. I probably still would have wanted to quit, and felt some mild skepticism toward the org ever after. But I would have felt much better about it, and I wouldn’t feel the need now to warn people that Nate was given unadvertised exemptions from social norms.
During the bike pump incident, Nate incurs an extra 30 seconds of wasted time, and hopefully no permanent emotional damage, in order to pad the angry interruption (“Learn!”) with a few extra syllables. Not even especially polite syllables, just enough to signal that he knows about and complies with the shared burden of emotional regulation. Something like, “Okay, I’m going cut you off right there–all I’m going to say is that I’m really unhappy and can’t deal with this right now. You should talk about this with your manager.”
Not long after the bike pump incident, Nate and I end up taking the elevator at the same time. He takes the opportunity to give at least a perfunctory apology and explain why he snapped. I get the sense that he genuinely regrets not establishing boundaries sooner. I also get the sense that he does not actively want me to quit (presuming that is actually true).
Nate doesn’t initiate any apologies, but he does send some kind of signal that things won’t go wrong if I ever try to bring it up again. He does not continue signaling a hair trigger while walking around the office. When I message him to let him know that the bike is back, he replies with something at least as cordial as “ok good to know”.
So I don’t think there’s any reasonable sense in which Nate was “trapped”.
What I want right now
I’m not especially bidding for him to change his emotional habits. In fact, I don’t really want much of anything from Nate himself.
When I called up my former trainee the other day, I did not say, “I’m sorry Nate was angry and toxic to you.” Rather, I said, “I’m sorry I didn’t make absolutely sure you knew what you were getting into. I’m sorry for not giving you the thorough disclaimer that I wish someone had given me.”
What I do really want is for Nate’s reputation to catch up to his behavior. I think this is already happening to some extent, and the community is appropriately deducting some prestige and bargaining power.
Some people I respect have already told me that they now think significantly differently about Nate, which I think is right and proper, as well as being some relief to me personally.
Looking Forward
I hope that if there is any more that ought to be revealed, that it is. I hope that bright-eyed young people entering the scene will be warned about him as thoroughly as I was warned about Eliezer (or more). I also hope that Nate does not suffer any unfair or unreasonable consequences[1]–I admit that I would feel some momentary satisfaction from that sort of thing, but it wouldn’t be worth diluting the valid complaints. Humans have an intrinsic drive for scapegoating, or so I have heard, and that’s the kind of overindulgence that is not affordable in the fight for the future.
I look forward to less often hearing Nate’s persona invoked, with no caveats, as an exemplar of agency and self-understanding. He has some notable virtues, and has done things worthy of commendation, but I hope that the halo effect around him substantially diminishes.
This risks derailment, but I hope that some new AI notkilleveryoneism researchers manage to loosen Nate’s monopoly on his niche at MIRI. By that I mean that it would be cool if, by some miracle, Eliezer found enough energy that he could invest some of it in a few final, dignified hail-mary apprenticeships. Instead of hearing that Eliezer and Nate said something, I hope I get to hear that The Cool New Eliezer-Approved AI notkilleveryoneism Team said something. I’m not holding my breath but that’s my hope.
I have some skepticism toward all of Nate’s self-reports, including the ones about his recent improvements. I am generally pessimistic about adults changing very much, especially when they are in their thirties or older. But...if Nate beats the odds in a hard-to-fake way, I will be duly surprised and impressed.
More miscellaneous notes
(Misc but still important)
I feel some pressure right now to match Nate’s conciliatory tone–pressure to realize that Nate was just a cooperative guy all along who honestly wanted the best and fell short of his own values. I also feel an opposite desire to shout, “No! He’s predictably performing gentleness now that he’s at a disadvantage! Onlookers will overweight this and walk away with the wrong overall impression!!!” I endorse voicing that feeling. I find Nate’s replies here just a bit too slippery, and I urge people to treat this as an occasion where actions speak louder than words. Questioning people’s motives is often considered bad form, so I will simply say that I believe that Nate is not outright lying about his thoughts or feelings, but also that Sarah Constantin’s Player vs Character model is weighing on my mind.
I want it to be known that during Nate’s outbursts, I was unsure about where his actual limits were. I obviously never worried that he would, like, punch me or something, since that’s quite illegal. But let me put it this way: If Rob had started shouting at me, I would have been utterly shocked, whereas if Nate had done that, I would have been merely surprised. Regardless of what Nate claims about his internal states, I think it’s not a coincidence that my uncertainty about potential escalation was so large, and that I was stressed into doing more emotional labor than I signed on for. To say that differently: I think it’s not a coincidence that his actions reliably gave him the benefits of brinkmanship. I find it entirely plausible that if Nate’s incentives had been different, then his emotional impulses would have adapted. But again, I would have been fine with just being forewarned during recruitment.
There were other highly regarded researchers at MIRI whose lunch orders I messed up, and who are not known for empathy or people skills, but who reliably managed to express mild disappointment instead of hangry hostility.
Nate said:
”My recollection of the thought that ran through my mind when you were like ‘Well I couldn’t figure out how to use a bike pump’ was that this was some sideways attempt at begging pardon, without actually saying ‘oops’ first, nor trying the obvious-to-me steps like ‘watch a youtube video’ or ‘ask your manager if he knows how to inflate a bike tire’[...]”Nate has a point here—it’s entirely plausible that the particular bike-related anger could have been resolved on my end the way he suggests. But I also find it entirely plausible that the difficulty could have persisted even after some googling and YouTubing, and I have to wonder what would have happened in that case. If I had managed to say the word “YouTube” early enough in that interaction, would Nate have let me finish my sentence? Well, that’s possible, and there are other possibilities too. So maybe Nate feels like he was tracking things cleanly enough that no one would trigger his rage unless they definitely had a serious lapse in their responsibilities (which is what most people think of themselves). But personally I still doubt that.
“[...] the entire hypothesized time-save of somebody else inflating bike tires is wiped out by me having to give tutorials on it.” This seems like an overstatement and I want to defend my common sense. I continue to think that the trouble I was having with the pressure-checking might possibly have been entirely solved by having me come watch him do it once.
The quotation, “Well, I couldn’t figure out how to use a bike pump” is a misrepresentation of what happened, and I’m disappointed to find myself defending against it. I had owned and maintained multiple bikes without running into this problem. Different equipment sometimes has subtleties. I still don’t know what was going wrong, and indeed it might have been a silly error on my part, but I claim that I was not as stupid, incompetent, or lazy as Nate makes me sound here.
It’s very dramatic, but I have to say I understand why TurnTrout said he would burn 10% of his liquidity in exchange for never having a certain conversation with Nate. My own experiences with angry Nate were weirdly unsettling. Again, I’ve endured objectively harsher treatment from superiors and customers in previous jobs, but those didn’t feel as bad. I think the difference had something to do with everyone I respected praising him, nuancing away his flaws, leaving me wondering what the hell was going on.
Alex Turner (TurnTrout) is my friend, we regularly talk and hang out and get food. In real life I think Alex frequently shows more pro-sociality and self-control than I do. When I think of Alex taking that kind of psychic damage from Nate,
I feel…just…intolerably scandalized, and need to do deliberate mental tricks to shake it off. (EDIT: I no longer feel intolerably scandalized. That’s because I have since had multiple, separate, surprisingly heated and frustrating in-person arguments with Alex, of a sort that never happen with my other friends. It’s still entirely plausible that Nate spontaneously lost his temper at Alex with very little provocation, but I now have some decent credence that Alex significantly contributed to the escalation. I feel obligated to add this minor change here, out of fairness to Nate and respect for truthseeking. My overall stance remains unchanged: Nate has a history of especially bad anger management, multiple friends of mine have been on the receiving end of it, his years of public image-crafting and blogging about his own self-understanding flagrantly omitted it, and while it’s not a crime to have anger problems, it’s definitely good that MIRI adopted a policy of specifically warning new recruits about it.)
Some positivity
I’m really happy overall with the replies I received from everyone. I expected to encounter a lot more resistance and skepticism, and I wouldn’t have blamed people for it. I have no regrets about anything I wrote. Without identifying anyone, let me say I’m also happy with what people have told me in private. It’s quite nice to have it confirmed that I’m not crazy and that I’ve helped make things better. And to those of you who have taken decisive actions: I honestly consider you to be heroes.
Even though it comes quite late, I’m glad to hear that there is some kind of effort being made to give people the warning I wish I had received. It’s maybe also good that Nate stopped being the executive director.
I sat on my complaints for 5+ years. But then last week, Alex Turner said the bravest public and true thing about Nate I had yet seen: that people are failing to enforce norms on him. Shortly thereafter, I overheard some people I respect having an actually nuanced conversation about it, not just the spurious, unbalanced “nuance” that had been protecting the status quo for years. Alex’s lonely dissent and the reverberations it caused inspired me to break my silence. Hero.
My experiences elsewhere in the ecosystem have been much better. I’ve been at my current job for over 18 months, and have witnessed none of the problems that Nate had while I was at MIRI. My coworkers clearly put in the effort every day to cooperate, and I feel good about the future of the org.
Some Links
Here are some links that have affected how I think about human affairs in general, especially interpersonal drama. If anyone finds my takes here to be too cynical, then these links will at least show where I picked up a lot of that cynicism. (However, I also fully agree with the cynical about cynicism post.)
[1] It was pointed out to me that no one hopes for unfair or unreasonable consequences, so my words here are vague. Yeah okay. People can debate what’s fair and reasonable, and I may join the debate or not.
Oct 11 & 12 EDITS: Restructured some sentences, swapped out some words, and added a whole bullet point to the Misc section.
Oct 3, 2024: Updated the final misc note to include the possibility that Turntrout was significantly at fault.
Alright, I’ll say it.
I did office operations at MIRI from Sep 2017 to June 2018 as a contractor and it finally feels right to share. All views herein are my own and not meant to represent anyone else. I intended to write a few paragraphs here but ended up with several pages.
Okay, so...my gut wants me to shout, “He’s not simply overly blunt in math arguments! He’s mean and scary[1] toward ops workers! Doesn’t anyone notice this?! I feel like I’m taking crazy pills!” That’s my gut reaction. My reasoned words below will be longer and more nuanced[2].
I wish some people at MIRI had explicitly told me up front something like, “Hey, if you mess up a lunch order, you may want to avoid Nate until the next day. He is a very high-performing researcher, and you should not expect typical levels of patience or anger management from him. Also, if you try to stand up for yourself, he may simply cut you off and storm out of the room. Such is the price of having world-savers…do you have thick enough skin to work here?” And I would have said, “Ah, I appreciate the candor and respect. Seems like you guys are making a reasonable tradeoff—after all Newton was notoriously prickly too. But I’m probably a bad fit for the role. Thanks for your consideration, and I look forward to seeing y’all at the next house party :]”
I spent over a week training my replacement at MIRI, and I feel pretty bad that I didn’t give him that exact warning. I think I said something like, “yeah, by the way, make sure to include such-and-such when you place the lunch order, or else Nate will get real mad at ya, haha. Anyway, moving on…”
(I really feel like I failed my fellow opsman here, so I called him a few hours ago and apologized. He doesn’t feel like I failed him, but he understands why I feel that way, and he appreciated the apology. Also, for the record, some of my fellow ops team said they liked the work I did and were sad to see me go, and I think they were sincere. MIRI was ready to take me on as a full-time employee but I backed out. I did like them too, and wished I could have been around them more and around Nate less, but that wasn’t what the org needed.)
I’ll try to sharpen up my point: I believe that different people are subject to different rules, regardless of official messaging. I claim that while I was at MIRI, Nate was not subject to the same behavioral regulation norms that almost everyone else in the community[3] is. He got away with angry outbursts and defections against norms of cooperative communication, over small stakes such as running out of sourdough bread too often. I think lots of people (especially new people) are not properly warned about this situation, which allows Nate to tilt the usually shared responsibility of self-control onto others in ways that they would not have agreed to if they had been better informed. Of all the things that have increased my cynicism toward the EA ecosystem over the years, none has disturbed me quite as much as the ongoing euphemisms and narrative spin around Nate’s behavior.
I feel like I’ve said what needs to be said, but for the sake of thoroughness, and because people will reasonably want them anyway, here are my object-level complaints. (I feel a little more exposed writing this part than the rest of it. You know how these things tend to be.) :
Once I was delayed a day or two in one of my biweekly grocery runs for the kitchen. Nate had already been grumpy about his favorite bread running out too often, and when a couple other things dried up that week, he...flipped out. He didn’t exactly yell at me and my fellow ops coworker, according to my imaginary decibelmeter, but he was indisputably hostile and aggressive, and obviously uninterested in 2-way communication. I did not have control over the food budget, and if I did, I would have been happy to just buy more food so we would run out less. (This was in the days of much less funding.) I speculate that this particular episode was splash damage from Nate’s conflict with the higher-up ops staff, but I don’t know, and that doesn’t really matter from my perspective.
My manager asked if it would be fine for me to add bike tire inflation to my task list. I was happy to. Nate later complained a couple times that I was leaving his tires underinflated. I asked him if there was some special trick for using his bike pump, because it felt harder to use than the ones I had owned, and I couldn’t detect any user error on my part. He said no, just do it in the straightforward way every so often. I was confused, but further questions felt unwelcome and it seemed possible that I really did just lose track or something. Anyway, one day Nate got a flat tire out on the road, which is dangerous, and he was understandably upset. He complained to my manager who informed me. I saw Nate in the office kitchen later that day (a Saturday) and thought it was an appropriate time to bring up again that I was having trouble with our available pump. I didn’t know how to–“Learn!” he snapped and then stormed out of the room. I considered this moment a major…update about the organization culture, communication standards, and my value in the org. After this incident, the ops team authorized me to purchase a reliable pressure gauge (I don’t remember if I specifically asked them if I could before that, but overall I felt like I couldn’t simply buy things to make my tasks easier without a very clear and legible need). Ever after, I have rolled my eyes at the suggestion that Nate has anything approaching normal levels of self-control or self-understanding. I am containing the urge to use much stronger language about this.
He got really angry at me when the rest of the office outvoted him on the choice of lunch catering. He had veto power on restaurant selection, and that was fine with me, but the anger was apparently not helping him remember to use it when it mattered.
There were other incidents. It was a pattern. I didn’t feel like there was anything I could do except quit, which I did, and which I think he wanted.
Here are a few miscellaneous notes:
With Eliezer, my experience has been the opposite. When I showed up in Berkeley, people who knew Eliezer tripped over themselves to tell me how arrogant and difficult they found him. I’ve talked to him for 5-10 minutes on 5-10 occasions, and every single time he was somewhere between neutral and friendly. I guess I should admit that this was never before lunch, and he was hardly ever around the office anyway, so the comparison is weak.
I once did operations at an early-stage tech startup. On a few occasions the cofounders yelled at us. It honestly didn’t feel too bad at the time, and I don’t hold it against them. The stakes felt real—the company was not yet profitable and we needed to change that ASAP or the company would stop existing. At no point were my expectations painfully violated, and at no point did I feel like my default communication norms were choked for no reason.
I actually had heard of or observed a few of Nate’s scary moments, going back to 2015. I feel ashamed that I was so slow to update on them. My best guess, based on fallible memory, is that by mid-2017 my overall impression of him was confused and uncertain: he had so much social approval, he had written an inspiring and uplifting blog, hell, he had done some quite nice things for me personally!…but there was also this disturbing thread of bad behavior. I didn’t interact with him much and I didn’t work for MIRI yet, so the cognitive dissonance didn’t feel urgent and I simply didn’t think about it much. I count this as a strike against my own judgment: however adaptive my cognition might be at any given time, I now question my ability to notice and call out bad behavior until it personally impacts me.
I’m highly uncertain about how Nate perceives his relationships to operations staff. Will all of my claims simply hit his ear as lies? Understandable confabulations of a disgruntled former contractor? Just a bunch of wild misunderstandings? I honestly don’t know, and I’ll withhold my more detailed speculations.
I could say a lot more, but this is already way too long.
Until now I have mostly kept all these things to myself, for the following reasons:
I didn’t want the trouble. Given that Nate was benefiting what I saw as a flagrantly false public image of self-control, I worried about what I would be getting myself into. It seemed like I would be initiating a conflict with someone who at best was not subject to the same rules as everyone else.
I didn’t think people would listen. After all, I didn’t listen until it directly impacted me.
Didn’t want to interfere with saving the world. But in light of Kwa’s post I now believe this was wrong.
I dunno when I’ll next check this comment thread. Might be tomorrow. Might be never. I wrote all of the above because I honestly believe it, and it finally feels worth the potential trouble to say.
[1]I considered the word “coercive”. But I exist in a free-ish labor market, no one coerced me into working for MIRI in particular. I left due to multiple reasons, including that I didn’t like how Nate conducted himself. That’s the labor market functioning properly. This is really important to me. There is a sense in which people with low market power are “forced” to work at lower compensation/conditions than people with high market power. But that “force” is being applied by every single potential employer who does not offer them a better deal, and I believe that any moral responsibility for a worker’s dissatisfaction is commensurately diffuse.
[2]I have some concern that discussions of Nate’s behavior often have a unusually high level of nuance and circumspection, in a way that makes me suspect motivated continuation or similar.
[3]Even given his rank and station.
- 's comment on Related Discussion from Thomas Kwa’s MIRI Research Experience by (10 Oct 2023 22:40 UTC; 12 points)
- 's comment on Related Discussion from Thomas Kwa’s MIRI Research Experience by (8 Oct 2023 17:58 UTC; 10 points)
- 's comment on TurnTrout’s shortform feed by (10 Jan 2026 20:50 UTC; 9 points)
- 's comment on TurnTrout’s shortform feed by (10 Jan 2026 21:26 UTC; 8 points)
In human interactions at any scale, it is net good to at least momentarily consider the Elephant and/or the Player, with very few exceptions.