the entire report was heavily AI-written
Huh, I knew they used AI to investigate, but did not get the impression that the report was AI-written. What signs led you to conclude that?
jasmine.ren
Karma: 336
Life and Meaning After Work
I don’t think Anthropic considers labor market disruption as a catastrophic for two reasons: First, as you mentioned, they largely left it out of this report and referred people to other studies. But more importantly, they stated so in the section on risks from automated R&D:
We are strongly incentivized to have our models substitute to the greatest extent possible for both the technical skills and general scientific/strategic judgment (or “research taste”) of all members of technical staff at Anthropic. Technical talent is scarce, and we invest heavily in this area to help improve our models.
Anthropic staff are among the most capable at what they do in the world. A model that’s capable of substituting Anthropic staff member at their jobs would be able to substitute human workers in many other jobs. Sections 3.4.2 and 3.4.3 of the Risk Report talk about how close their models are at being capable enough to doing so.
In section 3.5.1 they show that their model capabilities growth over time is superlinear, despite not meeting their RSP definition of “a doubling over pre-AI-assisted rates of progress.”
these risks do NOT depend on the abilities of unreleased models which Anthropic described in its report.
I agree that the Anthropic Risk Report does a lot of “narrowing of scope” and focus on risk that they define as “catastrophic.” For example, in section 3 where they discussed risk from automated AI R&D, they focus on risks like power shifts within or between nations or humans losing control, rather than labor market disruption (3.2 Threat model).
Is it a problem mostly solved by high-effort googling?
I agree that AI is far from the only source of misinformation in the world. That being said, I think there’s risk—not necessarily “catastrophic” as defined in the Anthropic risk report—if AI is involved in a significant fraction in human cognitive activities. I think this does inevitably happen, especially on jobs, where people have incentives to get jobs done quickly and it’s so much faster to offload things to AI.
Making sense of the misalignment risk model in the Anthropic Risk Report (August 2026)
Cultivating Gardens
New York City, NY – ACX Meetups Everywhere 2022
ACX Schelling Meetup NYC
War Mode vs. Peace Mode
If you have the time for it, please do. I tried to listen to an episode (namely, Your Time Might Be More Valuable Than You Think) and found it more difficult to parse than a regular podcast or audiobook because of the lack of intonation. Subtle cues like tone and pauses seem to make a larger difference in my understanding than I have previously realized. That being said, I haven’t experimented with the playback speed yet, maybe slowing down a bit would make the audio easier to parse.
Epistemic status: I have not been involved with Leverage Research in any way, and have no knowledge of what actually happened beyond what’s been discussed on LessWrong. This comment is an observation I have after reading the post.
I had just finished reading Pete Walker’s Complex PTSD before coming across this post. In the book, the author describes a list of calm, grounded thoughts to respond to inner critic attacks. A large part of healing is for the survivor to internalize these thoughts so they can psychologically defend themselves.
I see a stark contrast between what the CPTSD book tries to instill and the ideas Leverage Research tried to instill, per Zoe’s account. It’s as if some of the programs at Leverage Research were trying to unravel almost all of one’s sense of self.
A few examples:
Perfectionism
From the CPTSD book:
I do not have to be perfect to be safe or loved in the present. I am letting go of relationships that require perfection. I have a right to make mistakes. Mistakes do not make me a mistake.
From the post:
We might attain his level of self-efficacy, theoretical & logical precision, and strategic skill only once we were sufficiently transformed via the use of our debugging techniques. The overarching objective was to discover and “update” deep irrationalities and eventually become a sort of Musk-level super-person (“attain Mastery” of a world-saving-relevant domain).
All-or-None & Black-and-White Thinking
From the CPTSD book:
I reject extreme or overgeneralized descriptions, judgments or criticisms [...] Statements that describe me as “always” or “never” this or that, are typically grossly inaccurate.
From the post:
Another supervisor spoke wonderingly about Geoff’s presence in our lives, “It’s hard to make sense of the fact that this guy exists at all, and then on top of it, for some reason our lives have intersected with his, at this moment in history. It’s almost impossible to believe that we are the only people who have ever lived with access to the one actual theory of psychology.”
Micromanagement/Worrying/Obsessing/Looping/Over-Futurizing
From the CPTSD book:
I will not repetitively examine details over and over. I will not jump to negative conclusions. I will not endlessly second-guess myself. I cannot change the past. I forgive all my past mistakes. I cannot make the future perfectly safe. I will stop hunting for what could go wrong. I will not try to control the uncontrollable. I will not micromanage myself or others. I work in a way that is “good enough”, and I accept the existential fact that my efforts sometimes bring desired results and sometimes they do not.
From the post:
I sat in many meetings in which my progress as a “self-debugger” was analyzed or diagnosed, pictographs of my mental structure put on a whiteboard. What were my bottlenecks? Was I just not trying hard enough, did I need to be pushed out of the nest? Did I just need support in fixing that one psych issue? Or were there ten, and this wasn’t going to work out? If I was introspectively blocked for too long, I’d have to come up with something else I could do for the project, like operations or sociology, and quick. And if I wasn’t any good at those, I was out. I couldn’t be out, so I doubled down on trying to mold my mind in the “right” direction.”
Unfair/Devaluing Comparisons
From the CPTSD book:
I refuse to compare myself unfavorably to others. I will not compare “my insides to their outsides”. I will not judge myself for not being at peak performance all the time. In a society that pressure us into acting happy all the time, I will not get down on myself for feeling bad.
From the post:
No one else, with one exception being the head of sociology, was considered to have any theories that could possibly be good enough to significantly further the plan like Geoff’s could.
Overproductivity/Workaholism/Busyholism
From the CPTSD book:
I am a human being not a human doing. I will not choose to be perpetually productive. I am more productive in the long run, when I balance work with play and relaxation. I will not try to perform at 100% all the time. I subscribe to the normalcy of vacillating along a continuum of efficiency.
From the post:
I was regularly left with the feeling that I was low status, uncommitted, and kind of useless for wanting to socialize on the weekends or in the evenings. Geoff was known to sleep only about 5–6 hours. Multiple people in leadership had themselves booked from 7am until past midnight every single day of the week.
An Apprentice Experiment in Python Programming, Part 4
[Question] What are some good rationalist ice breaker questions?
I have disabled the RSVP function for now, because I realized that the LW RSVP system lists out people’s full names in public, without warning users that would be the case. Feel free to show up regardless of whether you’ve RSVP’d! We’ll have some other ways of tracking attendance, like sign-in sheets or counting the # of name tags given out :)
New York City, NY – ACX Meetups Everywhere 2021
what kind of thing do you want to do with Python?
Out of the things you listed, scientific computing & OSS libraries are things I want to explore more. I also don’t just want to learn Python—although I have chosen Python to be the language to try to get pretty good at—my goal is to get myself a proper CS education. I think it would be difficult to truly get good at a language without understanding how things work underneath.
Also, what gjm said.
I’m also a little confused about what you’re aiming to learn!
There are lots of gaps in my Python knowledge (this applies to my CS knowledge in general as well) and I’m trying to close those gaps. I asked gilch about decorators because I encountered them in
pytestand was very confused about how they worked.I didn’t have a project in mind, when I signed up for this apprenticeship, I just saw gilch offering to teach Python and thought I wanted to get better and learning from a mentor could be great.
I have attempted to contribute to open source in the past but have failed (ran into issues building things locally and didn’t know how to get help), would love to try again.
I just encountered this blog post by Dean Ball, and the conclusion shed some light on how he felt internally about AI existential risk culture over the years: