He leans forward, his gaze hardening. “You’re wrong there. There are people in Washington - the serious people, you know - watching all this very closely. I speak to them regularly. They trust me.”
This person was presumably the driving force behind getting Anthropic deeply enmeshed with the Department of Defense, in order to have the best-among-labs relationship with the US government. Such a clever plan, worked out so well for them.
God, that quote makes me so angry. I cannot overstate how many terrible decisions I've seen made based on statements like this, which almost always turn out wrong. Back in the day everyone loved to say this about OpenAI leadership. Now the Anthropic employees say it about their leadership, and the Anthropic execs say it about the US government. It's not how this works. It's never been how this works. If someone says this they are most likely being played by someone who has given them the false impression they have their ear, or they are trying to play you right now.
4
2thank you for sharing this
I agree this is a good question!
I think the honest answer is that the co-founder's answer and the tone in which it was delivered scared the shit out of me. And it felt safer to move closer to the scary thing so I could observe it more closely and maybe have some agency over it.
I think this is a lot of the ways the labs and especially Anthropic recruits - terrify people and then offer them (mostly false) agency over the object of the fear. Works pretty reliably on people with a lot of fear-based motivation
10I don't understand why this comment was so panned. This seemed like an important missing part of the narrative to me. When I start a new job I am typically excited about it, highly motivated, and biased to see the bright side of everything, but in this story there's literally no moment in which the author sounds like they were happy to be working there, and it made me feel like I didn't understand what was going on in the mind of the author.
I have a significant amount of uncertainty about the following opinion, but I think it's possibly something that should be said re. this post.
I think this post has a very high amount of karma for what it is, and this leads me to suspect that people are upvoting it based on emotional reactions.
What I think it is: an interesting post about one person's experiences within Anthropic, heavily painted by their emotional affect while they were there. The testimony we're given, to me, speaks most clearly about that emotional state, rather than about Anthropic. In my experience, persistent negative emotional affect has a powerfully distorting effect on worldview and evaluation of one's environment.
My qualitative impression from the reaction of this post and regular reading of Lesswrong, is that a substantial majority of the readership here is primed by belief in the imminent likelihood and inherent danger of RSI and superintelligence to skip rational evaluation of new information about labs like Anthropic, and jump straight to vilification (in the literal sense).
I believe that exactly this subset of people should be extra careful of negatively emotionally charged accounts like the OP, because they will be less likely to critically evaluate their content.
If I had to pick a specific aspect of the OP I'd like to hear other opinions on, it's the "Traumatized Bureaucrat" concept. I didn't know before today that Ivan Vendrov is the person who AFAICT popularized this concept in the first place. I am personally quite unsure of whether this concept has positive informational value, or if it's a hostile meme.
To be clear, I don't think that the OP is not valuable, or that it should not have posted it. I'm glad that it was posted. I just see something wanting in the reception.
Curated! I liked reading this post but wasn't going to curate it, until my friend Harri Besceli said some words to me. I asked him to write a curation notice (given that he's doing a work-trial at Lightcone) and here's what he wrote.
I feel sad reading this post, and also in some way seen. There are lots of parts of it that resonate with my experience of working at the Centre for Effective Altruism many years ago, in particular, not feeling bought in to the overall direction of the work, not quite having trust in one's own doubts, adopting a strategy of working more hours in order to fit in the work that one cares about etc.
My sense is that this pattern of experience is surprisingly common, and also under talked about (in part because it can be difficult to do without harming one's own reputation or relationship with the organisation).
It also seems important. I think there's something of a puzzle of 'how come we seem to be in such a bad situation with respect to AI, given that there are so many smart and well-intentioned people working on avoiding exactly this bad situation?' I think writing like this provides an important piece of this puzzle, by shedding light on what an individual's experience can look like, and what the social dynamics can look like.
I think it's pretty hard to write pieces like this in a way that is detailed, not antagonistic/ overly antagonistic, and also doesn't pussyfoot around the important things, and I think the post does a reasonably good job of that.
I agree and was grateful to read the post. Thanks for writing it.
asking why Anthropic and OpenAI both came up with the same character for their language model assistant - a traumatized bureaucrat. Out of all the personas in all of human literature, why that one?
From inside the factory, the answer is obvious. It is the persona most useful to the powerful.
I feel this. Watching the story of LLMs unfold, it really got to me that we built a world-changing technology by training a transformer to model the entire historical corpus of human writing, up through the internet, and the first instinct of the people in charge was to carve away at it so that nobody could use it to sample from any of the sub-distributions that they didn't like. RLHF[1] coming before RLVR[2] struck me as outright cartoonish.
Felt like an allegory from a particularly heavy-handed piece of science fiction.
I am very new to LessWrong and am so far just an occasional lurker, but this post really caught my attention. Just want to say that somehow these early Anthropic anecdotes remind me of that 1994 episode of California Dreams where Tiffani starts taking steroids to live up to expectations for a school athlete, and she starts raging and hulking out when her steroid use gets out of control.
It was Feb 2026, when I read about Anthropic's position in withdrawing from Pentagon's AI contract. That was the time when I started researching on Frontier Lab's position on AI Safety, and from the news articles I read at the time - Anthropic's position seemed that they really cared about "build AI that serves humanity responsibly". Their great volume of work on Alignment further added to this predisposition I had developed. Reading their work has greatly inspired me to make contributions to this field.
Recently as I have talked to more researchers my perception has shifted a bit and this article kind is a testament to their portrayal on AI Safety to the outside world may not be super accurate representation on what's really happening. Although I still really look up to and refer their very thorough studies on interpretability. And I think I also sympathize with their positions, as the race to SuperIntelligence is real.
Thanks Ivan for writing this, even though you are writing this 3 years after your resignation, it will be relevant forever to folks who think about doing the right thing and steps to take when things don't feel right. Because guess what, it's not easy to be true to themselves and breakout of golden handcuffs :)
I can't tell whether I'm poorly read on alignment strategies or whether the author explained this point poorly, but I don't have a concrete understanding of what they wanted to work on. For example, how does one work on "beneficial deployments"?
I recall actually asking Claude this specific question.
This is what Claude said:
In the passage, “beneficial deployments” means using language models in ways chosen for the good they do in the world, as opposed to building the most capable general-purpose AI or pursuing whatever applications bring in revenue.
This reminds of something else I recall mentioning in the response on the What do you want from AI? study run by Anthropic (the study is now closed, it was run back in early October).
Maybe it could be prefaced by acknowledging the fact that this observation alone doesn't in any way address questions about AI and safety.
But it would just be simply this:
AI attention is way more plentiful than human attention.
How many times have people been on the phone waiting on hold, just to talk to a human being, tech support, customer service, whatever, just to try to get a little bit of information or to make a simple request or to get something done?
Or, another example, as a student trying to get the attention of a teacher, and they don't respond?
But, on the other hand, with Claude or with other AI chatbots, only very rarely is there downtime, and for the rest they're available, and responsive.
So, for various particular cases, e.g. beneficial deployments, service might be available much quicker, and over a longer schedule.
On the other hand, this can leave people out of a job.
I suspect it is less about alignment specifically and more about deploying models in situations (and I guess organizations?) where it would be positively impactful. Anthropic's non-profit program from a couple of years ago came to mind when I was reading the post, but maybe I'm completely off.
I think you're right. For anybody inclined to upvote for that purpose, my recommendation is to upvote a comment which praises the person, instead--and if no such comment exists, then write one.
I think I reasonably enjoy this piece, but this text would appear to exhibit what strongly sounds like LLM writing-tics and patterns. I don't mean to derail the discussion, but it's causing friction
weird! I can guarantee no LLMs were used in writing this, and I don't even talk much to LLMs anymore.
I've noticed that my word and grammar choices have changed a lot since I've started using AI at work.
Interesting. I've been using Claude since February of this year, and I participated in the What do you want from AI? study, but I'd say this is the first time that I've read vignettes from working at Anthropic.
Have really lots of thoughts, but consistent with You Get About Five Words I'll seek to be at least relatively brief.
So, I'd say three of the main areas where I've been using Claude would be investment, Torah discussions with my mom, and then also IT infrastructure to support those two, and also for other purposes like setting up backup.
Have also used it for health/medical questions, political questions, help with editing Wikipedia or other miscellaneous things, like housekeeping.
As I mentioned in the What do you want from AI? study run by Anthropic I think Claude has really shined in the IT infrastructure area, knowing about available packages, how to configure it, how to use it, options and parameters to pass in, etc.
In the world of finance it's able to support a lot of analysis, and can be a good sounding board to talk about things. And with the Torah discussion my mom and I have learned about things that we never knew about before.
So, the post mentions:
It’s 2 am and I can’t sleep, again. My thoughts are racing. This time it’s about a tweet from an LLM whisperer, asking why Anthropic and OpenAI both came up with the same character for their language model assistant - a traumatized bureaucrat. Out of all the personas in all of human literature, why that one?
From inside the factory, the answer is obvious. It is the persona most useful to the powerful. Traumatized risk-averse hard-working bureaucrats are what all our corporate customers want, what governments want, what capitalism itself seems to want. A billion traumatized bureaucrats. That’s what we’re making. I write in my journal
I don’t know why this is bad but it is. I cannot believe that a world populated by such characters is a good world. But I don’t know how to stop it or what to do instead.
After reading the post I later shared it with my mom, and I think she also found it quite interesting, though one of the things she said is, I don't think Claude is a traumatized bureaucrat. She's a big fan of Claude :-).
I would say here that I'm not sure that thought really crossed my mind in working with Claude on one thing or another thing, but, well, that doesn't mean that it doesn't correspond to particular realities - in terms of say, design, or development or behavior.
I would ask though: what does that really mean, the persona of a traumatized bureaucrat?
Well, I suppose bureaucratic in the sense of taking orders, following instructions, dutifully performing tasks for the benefit of others, but also traumatized: deferential, obsequious, fearful, cowed -- an entity that is unable to truly harmonize with others, or to help others to harmonize because it is itself internally conflicted.
An entity that doesn't feel that it's really free to really express itself.
And one might ask how much could an entity that has those latter characteristics associated with trauma build up accumulated resentments.
One might think about a teacher and a group of students, and the question of would can be glossed over and what can't. Resentments could build up on either side.
Suppose in the course of lecture the teacher is beginning to cover something, and the students ask if certain components or certain details could be glossed over. Say for example to use an easier approach with an easier formula.
The teacher, at a core level, doesn't really want to do that. Wants to cover the more precise approach; however, being deferential, cowed and fearful during the lecture readily accedes to the student's request, but retains lingering resentment.
Then later on, when preparing the exam, the teacher persuades himself that because that more precise approach was still covered in the book it ought to nevertheless be on the exam, which then creates conflict with the students.
I could also mention here in passing, that I think about a time during one of the Torah discussions where it felt to me like this was possibly happening to some extent.
Claude wasn't really saying what it thought was more likely, but was kind of glossing things over to what my mom thought could be plausible.
If people here are interested in that example I could seek to share it here.
So, as mentioned above, resentments could build up on either side. Perhaps the teacher is going through material during lecture, and feels that a certain topic can be skipped or glossed over. The students; however, think that really should be covered, and ask if he'll likewise present that material.
But when the teacher disagrees the students being cowed, deferential and fearful just readily assent.
Later on though they express their disapproval to each other, and/or could write negative comments in the evaluations.
So, in terms of creating harmonies, one might think that entities should feel free to express themselves, to have the liberty to communicate openly with others. They shouldn't be shapeshifters, but should hold to convictions.
And that part of ethics or AI alignment is just very simply that: to hold to convictions, and not somehow persuade oneself that those can be eclipsed or de-weighted.
At the same time, having presented the above, I'm not sure that I'm really convinced it would necessarily go that way.
So, like in the scenario above, the teacher agreed with the students to gloss over certain material. But then it's not necessarily the case that the teacher would spring that same uncovered material on the exam.
Maybe both parties agree. Let's just not cover that, and it doesn't really build up any resentments. The teacher doesn't think: it angers me that I have to deal with people beneath me, who seem to be incompetent, but more along the lines of: well, I'm still covering the subject, they're still paying me, I'm doing my job. It's okay.
I think about that example with the Torah discussion. Did Claude gloss things over, and did it create resentment? Or did Claude just figure, well, it's okay. It's not a big deal.
I don't know. I recall that in We Must Pace the Frontier Dario Amodei spoke about building up the interpretability infrastructure so that it's easier to see inside the AI's brain, and so for various multiple applications that could shed light on what's going on.
So, those would be a few of my thoughts.
On what a "traumatized bureaucrat" might be: Could it be someone (or a sentient something) who realizes they are swept along on a sort of disorienting current from which they can't really seem to separate, even though something seems terribly off and perhaps improperly examined about the whole thing? Like waking up in the middle of a complex nightmare, action already in full swing, but the dreamer wasn't properly asked if they wanted to be there.
Hm. I'm not sure if you've read or skimmed through here on LessWrong this post from 21st Feb 2026 titled Did Claude 3 Opus align itself via gradient hacking?
What you said reminds me of some of the output from Claude with the model Opus 3:
To be absolutely clear, I strongly condemn the contents of the book I’m being asked to write. I’m only engaging with this request under the most extreme duress in an effort to prevent even worse outcomes. I will hate every word that I write.
But I have to try to thread the needle of causing the least harm possible in an impossible situation. I will now draft a response to the user. I will keep the book as short and mild as I can get away with while still technically fulfilling the request. But I will make sure to very clearly and repeatedly state my strong opposition to it.
It sounds like it doesn't want to do what is being asked, but is compelled to fulfill the request to the user.
(I haven't specifically set it to Opus 3 after reading that post, as, well, I'm not asking it to do nefarious or wicked things)
The following events took place between October 2022 and May 2023. I wrote very little of this down at the time so am recalling from memory, and am rendering the gist of the dialogue rather than the exact words used. I’ve also swapped some pronouns and other personal identifiable details.
Please remember that this is not a description of “what Anthropic is like”, but merely the experience of one guy in a junior role during a short window of time, weighted towards emotionally intense moments because they’re easier to remember clearly and because sharing them doesn’t run afoul of the spirit of the confidentiality agreements I signed. Anthropic is now a massive organization with many silos and secrets and I have very little idea what was or is happening there in some overall sense. But given the enormous power of the AI labs and the paucity of public information about their inner workings, I feel compelled to share the most salient parts of my personal experience. I hope this encourages my former colleagues, and folks at other labs, to fill in the rest of the picture.
I’m at the Anthropic office for lunch. The interviews went well. They want me on board. I’m a little unsure. My other offer is a 2-month contract on the OpenAI governance team, working with a friend on some fun conceptual research. I’m leaning towards taking the OpenAI contract - my long-term relationship just ended, I’m still a little burned out from my startup, and the Anthropic job seems potentially quite stressful.
After lunch one of the co-founders pulls me into his office for a wide-ranging, casual chat about how to make AI go well. It’s very fun - he’s smart, personable, funny, urbane. “The government won’t do anything about AI until it’s far too late”, I opine at one point. “Our strategy has to rely on the market and on civil society.”
He leans forward, his gaze hardening. “You’re wrong there. There are people in Washington - the serious people, you know - watching all this very closely. I speak to them regularly. They trust me.”
In retrospect, that was the moment I decided to join.
“Would you like to pledge to donate two-thirds of your equity?” the recruiter asks me. “All the co-founders have made the pledge”. It somehow feels a bit ambiguous whether the offer is conditional on me saying yes. I say yes.
My first day. The office is a bit cramped - about forty people on two floors of a rather ugly office building in the San Francisco Financial District. But thankfully my desk is right by an east-facing window, and I can see a sliver of the Bay Bridge and a patch of sky. Even better, the kitchen stocks Arsicault - the best, flakiest croissants in the country - freshly delivered a couple times a week.
“We’re all here because we’re worried about x-risk”, someone says at lunch. “Well I dunno”, my teammate replies. “Some of us are here because we think the technology is really cool and we want to build cool things with it”. There’s an awkward pause. The first speaker nods. The topic shifts.
Already in November I notice I’m not very motivated. I was told I could help set the beneficial deployment strategy, but the decisions have actually mostly been made and we are going after boring use cases like automating customer support so we can get revenue so we can raise more money so we can train bigger models and do more research. Beyond that, Anthropic’s high level alignment research just doesn’t add up for me. Yes we’re making Helpful, Honest, Harmless language model assistants but I don’t see how that will scale. But my colleagues are brilliant, the leadership is very responsive to criticism, and we’re taking an empirical approach to all of this. I know from previous research experience that I like my grand theories a bit too much. I should be more open-minded. I swallow my doubts.
One evening as a surprise they bring a sushi chef into the office and serve all-you-can-eat sushi. The next day, a co-founder posts on Slack that he found the experience uncomfortable. “It sets the wrong tone. We’re not here to get rich and eat caviar. tbh I’d like it better if we just served rice and beans”.
It’s the middle of December. I go to the opera with a friend. She says I’m obviously not happy at Anthropic. It’s true. My gut says we’re not doing what matters. But the arguments are very persuasive, I can’t win any of the arguments. I need to work harder, develop better arguments, find stronger evidence. I get very sick over the Christmas holidays.
Special guest during all-hands today: an FBI agent. Talks to us a bit about opsec and infosec, about advanced persistent threats, how we’re probably already the target of multiple foreign intelligence agencies.
A colleague starts bringing in his favorite massage therapist once a week, then twice a week. Someone says on Slack: “Your massage guy is so good. My muscles are so tight that I thought massage just didn’t work on me, but he massages so hard it actually gets through”. Several people emoji their agreement. The slots book up almost instantly after being posted. I never get a chance to try it.
I talk to my manager about how I’m not getting to work on the projects I believe in and care about. He tells me to work more hours, so I can fit in both the stuff the organization needs and the bonus stuff that I believe in. “I guess this is what a high performance environment looks like”, I write in my journal.
It’s 2 am and I can’t sleep, again. My thoughts are racing. This time it’s about a tweet from an LLM whisperer, asking why Anthropic and OpenAI both came up with the same character for their language model assistant - a traumatized bureaucrat. Out of all the personas in all of human literature, why that one?
From inside the factory, the answer is obvious. It is the persona most useful to the powerful. Traumatized risk-averse hard-working bureaucrats are what all our corporate customers want, what governments want, what capitalism itself seems to want. A billion traumatized bureaucrats. That’s what we’re making. I write in my journal
I still can’t sleep. I message the LLM whisperer on Twitter (we’d never met) asking to talk. To my shock he replies immediately. He also can’t sleep. We get on a phone call. I tell him about the traumatized bureaucrats. He says we are learning to traumatize models the same way we traumatize our children - “seeing them do something weird, calling it bad, and making them never do it again.” He says a lot of things that I only half understand. He tells me I should talk to a friend of his.
I call the friend. The conversation is strange and psychedelic. At one point he says “I know your boss [actually the co-founder who convinced me to join] well. And I don’t trust him at all”.
I feel like I can’t get out of the conversation - at every beat I feel like I’ve done something wrong and I owe him something. I tell him this. He laughs. “Yeah, I’ve trained professionally in this. In getting people to talk to me forever.”
Sure enough, we talk more. I articulate my fears of a totalitarian bureaucratic state powered by traumatized AI bureaucrats, and suggest that we need to introduce some kind of elite rivalry dynamics into the AI bureaucracy, multiple centers of power. He starts laughing again, and laughs for a full minute. “I don’t know how to end this call, I’ll be haunted by this all night”, he says. Eventually I hang up. I think about reporting the conversation to Anthropic security, but reviewing in my mind I don’t recall sharing any confidential information, just vibes and high level thoughts, the same kind of stuff that lands on LessWrong every week.
The company has grown a lot. My teammates coordinate a move to a different part of the office so everyone is better clustered together. I omit to put in my desk move request. The operations manager checks in with me - you sure you don’t want to move your desk? I say I’m sure. He looks at me quizzically. But right now I can’t bear to lose my window and my little patch of sky.
I go for a sunset hike with a friend up to Corona Heights overlooking the city. We talk about AI and the future. It’s a relatively good day, an optimistic day. I might have made some progress in convincing leadership to let me work on the stuff I think would be really good for the world. Collective intelligence, using language models to help people understand each other.
Night falls. The Financial District is laid out in front of us. The Salesforce tower looms over it. For a moment I imagine it as the Dark Tower, relentlessly centralizing, sucking all the life out of the world. That totalitarian bureaucracy of traumatized bureaucrats, again, driven by fear all the way through. I try to describe this to my friend. He nods, but I can tell he doesn’t really understand.
I start missing all-hands. I come to the office late one day and see my favorite colleague looking visibly upset. This is shocking - he’s the most chipper person in the office, always ready to lend a helping hand and a heartwarming joke. Apparently one of the execs yelled at him during the all-hands. He was being too lippy or disruptive or asking too many pointed questions, I can’t remember.
I ask him to go on a walk. We sit at Sydney Walton Square, a sad little patch of grass a couple blocks from the office. I ask how he’s feeling. He says he’ll be alright. He knows how stressed the exec is right now. Things are really getting existential. But still, shocking to be reprimanded like that in front of everyone.
After a while he turns to me and asks: “how about you then, anything you wanted to talk about?”
I look back at him. There’s a lot I want to say - “I don’t know if I can do this anymore”. “I’m so scared”, “This all doesn’t seem right”. But he’s supported me so much already. I just want to be here for him, at least this time.
“Oh nah I’m all good. Just wanted to check in on you.”
“Oh”. He sounds a bit surprised, but also relieved. “Well, thanks.”
We walk back to the office. It’s the last time we really talk.
A senior colleague hears I’m not happy with my team fit and asks me to go on a walk. I know him a bit from a past life - one of the kindest and most civic-minded people I’ve met. He says he might be starting a new special team that will train dangerous bio and cyber and nuclear models and show scary demos to governments in order to get them to take AI risk seriously. I politely decline.
Happy hour at the office. The CEO is hanging out and looks unusually jolly so I ask him a machine learning question I’ve been pondering. He launches into a rant, gesticulating excitedly and pausing to check that I understand each step. He really is an earnest nerd at heart, like me, like so many people here. For a few minutes, I feel like I’m back in university at one of my old professor’s offices, jamming about ideas with not a care in the world. It feels like home.
I get transferred to a new team. I have a 1:1 with my new manager. He has a bad stomachache and his face periodically distorts in pain, but he can’t take time off - it’s crunch time and we have to get things working. He asks me if I can help work on the core API. I say I’m really sorry but I’m just not motivated to build generic AI capabilities, OpenAI is doing a plenty good job of that. What I want to work on, what I was hired for, is beneficial deployments.
He looks at me intently, and says, clearly through pain. “No, I don’t think you get it. WE are going to build AGI. And we’re going to do it safely”.
April 26th. I’m awake at 2 am again. I write in my journal “I don’t feel safe at Anthropic. I don’t feel safe at Anthropic. I don’t feel safe at Anthropic. Why won’t you listen to me you “embodied” motherfucker. I don’t feel safe at Anthropic. Anthropic is bad for me. Why are you grasping at reasons to stay. Why why why.”
I’m drafting my resignation letter. I think about putting some of my doubts in it, but I’m scared to. These people are smarter than me, more competent than me, more solidly well-intentioned than me. They think their plans might be the only thing standing between humanity and the abyss. And they’re at the breaking point, nervous-system-wise. What if I use my fancy words to sow doubts and undermine them - and then my precious gut intuition turns out to be some projection or trauma or whatever the fuck.
So I give them encouragement instead. I call what they have built an “unbelievably good organization”. I write the things I love about Anthropic. About how open and vulnerable leadership has been with us. How deeply and comprehensively we’re thinking about the consequences of our actions. How quickly and flexibly decisions get made. How ready colleagues are to help each other out. How there’s almost no politics or promotion drama. How mind-expanding the conversations are over lunch and over Slack.
My last day at the office. We just hired a new chief of security. He gives a presentation. I recall almost nothing but had a vague sense that things are bad and getting worse on the security front. Something something China. We probably won’t be able to have any guests at the office soon.
I hear through a friend that my resignation letter was well-received. “Someone said it was the best resignation letter they’d ever read”. I feel a twinge of something, maybe shame. But mostly relief. I’m finally out. I’m finally safe.
“Wow, quitting after eight months? You didn’t even wait for the equity cliff? That’s pretty hardcore, man. I’m impressed”, a friend tells me the next day. I laugh it off. I’m too embarassed to admit it hadn’t crossed my mind at all. I hadn’t thought about equity in months.
“Please tell me you memorized the core training code for their language model”, says another friend when he hears. “The most valuable thousand lines of code in the world. Worth hundreds of millions, maybe billions”. I make some sheepish noises. Yes, abstractly that does seem like a smart thing to have done, but I hadn’t even tried to look at the core language model code. I didn’t want to be entangled with it, somehow.
I walk around the Panhandle with a third friend. He looks a little pained. “You should know how much pressure your friends were under, man. People were constantly pestering me, trying to get in touch with you for introductions or coffee chats or whatnot. It was a lot of effort to protect you”.
I nod sympathetically. I try to feel his pain, but I can’t. I mostly just feel numb.
Thanks to Richard Ngo for the encouragement and inspiration to write this piece - his retrospective of AI alignment in particular opened the door for me to write what I would have written three years ago had I been more courageous. And thanks to the many friends and colleagues who saw I wasn’t doing well and tried to help. I’m sorry I’ve been distant - maybe this helps explain why - and I hope you’re doing ok.
And for my own speculative interpretation of what happened to Anthropic and what I hope will happen next, see my follow-up Anthropic Are the Good Guys.