I've been psychologically blocked for many years about contributing to alignment work, and recently unblocked. Not sure how generalizable this is, but wanted to share what was blocking me:
I had two modes of thinking that I label as [normal civic responsibility] and [heroic responsibility].
Roughly speaking, civic responsibility is act in such a way that if everyone acted this way, the problem would be solved. Going out to vote, emailing your senator, personally committing to not pushing the button.
Heroic responsibility is just solve the problem, all by yourself if necessary. Move p(doom) appreciably by yourself, maximize marginal impact, start a new moonshot org.
What I'm coming to realize is that there is space for a sliding scale, especially regards to alignment work, between civic and heroic responsibility. Starting with civic responsibility and going up, Responsibility Level n is act in such a way that if 1/n fraction of humanity acted this way, the problem would be solved. n = 1 is civic responsibility, n = 8 billion is heroic responsibility. Hill-climbing on n seems to be the right way for my brain to approach personal responsibility towards the world.
This is a helpful way to reframe things, especially from the perspective of "I don't know enough about [thing] to make meaningful progress on it by myself" and trying to not feel hopeless about it.
I imagine a lot of the intermediate range (between n = 1 and n = O(
It's not obvious to me how I can hill climb on n in a more direct way. I would guess that you would need somewhere between
I do agree that a lot of low-hanging fruit comes from "lean on social capital to recruit people more than you're normally comfortable with." That being said, it does feel like almost always recruitment strategies actually scale off raw power level in some fair way (not sure I have the right vocab for this), e.g. I found that I have pretty high success x-risk-pilling my friends and colleagues and they have relatively low success pilling theirs afterwards, because they've thought about alignment for like a month. So the higher-order network effects are much smaller than I expected given the initial success.
I've been trying to learn a bit about the broader game board for alignment, and notice I'm deeply confused about certain giant blanks in my map. I'm hoping someone can at least point me to any way of thinking about questions such as:
(1) Are we actually way more doomed if Chinese AI labs win the race?
I've heard takes ranging from "China has no safety work to speak of and this is an instant bad ending" to "Xi Jinping and the CCP are broadly competent and aware of the risks and we should trust them to manage risk better than the US, if anything" to "it's not even a consideration, Chinese labs are just copycats who rely entirely on distillation and their AI labs will never overtake the west."
(2) How do I think about the possibility of government takeover of AI labs?
I've heard takes ranging from "this is an inevitability once AI crosses certain thresholds" to "the administration is too incompetent and slow-acting to possibly carry this out."
(3) Are multipolar endgames realistic?
I've heard takes ranging from "the first lab across the finish line will dominate/absorb/neutralize the competition, we should only worry about alignment at the winning lab" to "the only way we survive is if lots of different ASI's appear nearly simultaneously and keep each other in check."
(4) Should we expect sufficiently coherent AGI to intentionally pause recursive self-improvement for a substantial period of time while it solves its own alignment problem?
I've heard takes ranging from "RSI is inevitable and capabilities will go to the moon" to "Yes AGI will pause for a microsecond but it'll be easy for it to solve its own alignment problem" to "alignment is so much harder than capabilities that barely supercritical AGI will take over and pause AI development for a long time, if not indefinitely."
The Bitter Lesson-pilled way to look at Nos. 1 and 3 is to realize that the bottleneck is compute not research:
1) In the medium term, the Chinese AI labs can't win the race because they lack the chips to do so because of sanctions, therefore the question is moot. Theoretically and in the long term, AI alignment is viewed in China as alignment to the party agenda not to some "universal human values" (in a sense how it's used in the West) which according to the Chinese state ideology do not exist.
3) As long as the top-2 frontier labs have similar capitalization and similar amount of funding available, they will have similar amount of compute and their RSI programs will be bottlenecked by that roughly equally. So as long as that holds, they will be in similar position, and note there's no concrete "finish line" because the AI capabilities are jagged and will remain so.
I have not thought through how long will it hold though, presumably if one the AI labs folds financially (AI bubble hypothesis etc.) they might decidedly lose the race. But even then their alignment research might get published and used in the leading lab so it won't be useless
2:
"the administration is too incompetent and slow-acting to possibly carry this out."
I think overall competence shouldn't be the most important factor here. As I see it the Trump administration is faster acting and more willing to take flashy actions despite social and legal precedents (or even legality), and I think this is something people across the spectrum should notice (for example, I have a quite negative view of the administration, in large part because Trump will quickly take abnormal actions that I don't like. People who have a positive view often like him for fast abnormal action in favor of stuff they like!). In some scenarios this makes a government takeover more likely (e.g. if the AI cos get on his bad side for some reason, and if there's some ballpark of pretense, then other factors matter less).
4: I am not sure AGI can solve its own alignment problem, as it's a chicken and egg situation. It would have to have an understanding of what human values we aim to instill in models to solve it's own alignment. If it did have this understanding, it would be already by definition aligned.
Instrumental Rationality Mini-Retrospective
I promised several years ago to write a retrospective on Hammertime a year after it was released. I broke that promise but I wanted to take some time to do the work now, and to summarize my current beliefs about how much rationalist self-improvement affected my personal growth. I'd also like to estimate how it compares to other schools of self-improvement I've dabbled in.
First, I should mention that epistemic rationality has been directly useful in my career, although this is highly unlikely to generalize. At least three of the best papers I wrote in graduate school featured tricky applications of Bayes' Theorem. I was able to do from muscle memory certain calculations about conditional probability and expectation that might have taken weeks otherwise (if we figured them out at all). I attribute this ability in large part to reading the Sequences. YMMV if you don't work in or adjacent to probability theory. The rest of this review will ignore this relatively idiosyncratic improvement and focus on my beliefs about what is usually categorized as "instrumental rationality."
Over the last three years, I've seriously attempted the following "self-help schools of thought": CFAR applied rationality, therapy, Jordan Peterson, Tony Robbins. Insofar as numbers mean anything, in terms of personal progress in my life: CFAR is responsible for a ~100% improvement, therapy ~15%, Jordan Peterson ~40%, and Tony Robbins ~200%. Tony Robbins I encountered quite recently so the error bars are the biggest, and the effects have not yet had time to play out. As weak but verifiable evidence that the other three actually made an impact on my life, my publishing rate at least doubled since CFAR, the theorems have improved despite this. Also, I believe the quality of my writing has radically improved in these three years.
For a few more gears into these numbers:
CFAR is unique in that it (a) synthesizes the best bits of many other schools and couches it in language that is palatable to its target audience, (b) immerses you in practice in a way that any other immersive workshop does approximately equally well, but is nevertheless strong in a way regular practice and habit isn't, and (c) is astronomically memetically safe compared to the alternatives, due to deliberate choices like noting when models are "fake but useful but definitely fake."
Therapy can be hit-or-miss, the average-case scenario seems to be roughly that you pay a savvy friend armed with one or two CFAR techniques to listen and consciously apply that technique once a week. Still probably worth it.
Jordan Peterson was very useful as a charismatic psychology professor. His Youtube lectures seem to be the best available resources on the subject, and I don't believe there is a single other source that will get you the same level of understanding of the academic literature at the same cost in effort. What you get out of that abstract understanding will depend primarily on your general problem-solving ability. I don't believe there's anything remarkable about his self-improvement philosophy, although doing it will help (just because doing anything like this helps).
Tony Robbins is the best-in-class person at simultaneously blasting every part of your brain with charisma and motivation, and works for a general audience of homo sapiens (unlike CFAR). He is also extremely memetically dangerous - he knows all the buttons of the human brain that have been discovered and will push whichever ones it takes to effect the change he envisions. (For example, much of his technique rests on attaching positive meaning to suffering.) If it's any comfort, his intentions seem to be about as good as one can possibly expect for someone armed from adolescence with this superweapon. He is currently my leading candidate for "human best at getting out of the AI box." I do not believe I could have safely jumped into watching a bunch of his videos, learned a lot, and jumped back out, without much of the preparation and growth I got from the other sources listed above.
Interesting perspective! As someone who came to CFAR from the perspective of "ooh, another self-help bootcamp to try" I'm highly interested in these sorts of comparisons.
I'm curious in what medium you tried Tony Robbins. I've never gone to a workshop of his, but my own experience with his books and audio programs is maybe only a 10% improvement. I haven't found neuro-associative conditioning to be that effective, and found the motivation from his programs to be fleeting. It's possible that having grown up on NLP, I'd internalized a lot of his best ideas already.
I'm also surprised that you call CFAR astronomically memetically safe. In my experience, there's a big memetic danger of CFAR, which is that it tends to make people overly cautious of Jordan Peterson/Tony Robbins style "just do it" advice. It emphasizes gentle yin style congruency advice to such an extent that it innoculates people against other effective forms of self-improvement. Much of this is the CFAR advice combined with the general CFAR culture of course, and not just the lessons.
The way I tried Tony Robbins was "spend two full days watching all of his youtube videos at 2x speed." I don't believe that much of his power will transfer through his books/audio programs. In fact I don't believe much of his power can be acquired by listening to his ideas at all - they are helpful and were somewhat novel to me but mostly forgettable.
To see what his power is, I think it's worth watching some of his relationship intervention videos. As far as I can tell, one of his core strategies is "solve an irreversibly damaged relationship in an hour by making both parties fall deeply in love with him (TR) and then transfer that love to each other." (Of course, the examples on youtube are only the cases where TR succeeded most totally, so his actual success rate doing this is hard to estimate. If I had to ballpark, it works 50% of the time, so TR can make half of all humans fall in love with him in under an hour, under a weak precondition of the other person being receptive to a normal conversation.)
Unfortunately (or fortunately for him) this has the uncanny side effect of turning a consistent fraction of the attendees (in particular exactly the fraction who get the most out of his workshops) into a zealous army of unpaid volunteers who follow him all over the world. I may be using the words "memetically safe" incorrectly, but this is the danger I'm pointing to that I don't feel from CFAR. I didn't consider the opposite danger of CFAR immunizing against other forms of self-improvement, it seems like it wouldn't be too strong of an effect but much less bad in any case?
To see what his power is, I think it's worth watching some of his relationship intervention videos. As far as I can tell, one of his core strategies is "solve an irreversibly damaged relationship in an hour by making both parties fall deeply in love with him (TR) and then transfer that love to each other."
I don't think I've watched anything of his on Youtube except his TED talk, but I have watched some of the Robbins/Madanes trainings and seen some episodes of Breakthrough, not to mention the Netflix special, "I Am Not Your Guru". In none of those places do I recall see him doing anything that looks like making people fall in love with him and transferring it, though I do occasionally recall him directing couples to do things that, in the moment, I would expect to create a sense of intimacy or love for each other. I've also been to one of his live events (almost twenty years ago), and didn't see anything like that.
My impression from that live event, and from the recordings I've seen, is that his events are very attractive to extroverts (or people who need excuses to act in a more extroverted way), as they're rather like a rock concert slash religious revival slash sporting event. It's not that he doesn't also have very sensible and intelligent things embedded in that context, but without the extrovert juice I doubt they'd sell out so well. Despite the fact that he despises being known as a "motivational" speaker, a huge amount of his success is likely due to just him being The Most Enthusiastic Man In The World.
While his books read in a fairly overblown way, and some things are just repackaged versions of other things, I think there are a lot of actually new or useful insights in his work, taken as a whole. But his big-audience stuff is relatively dumbed-down because, when dealing with a big audience, you have about five words, and he clearly knows it. And for him, the five words to use in the context of relationship repair appear to be: "Get your partner's needs met."
The phenomenon I'm pointing to via "making couples fall in love with him" (which might be the wrong words) is that in relationship interventions he uses a combination of explicit models, personal charisma, and hard-to-transfer people-reading skills to make each person feel understood at a level that causes them to trust him deeply. This level of trust seems pretty extraordinary and hard to distinguish from love. After that he proceeds to use exercises to transfer this trust to between the couple in a way that requires very little agency on the couples' part. They sort of just go along with/paraphrase TR's statements to each other and then get this massive intimacy boost. I would guess that they come out of the experience feeling as positive or more so about TR than about each other.
I would love to hear which pieces of his written work you think of as "actually new or useful insights," the only thing that fits that description for me from his youtube videos is the Six Human Needs, which is a useful template for goal-factoring for me.
The trust stuff you're talking about is generally referred to in NLP as creating rapport, though Robbins is certainly really good at it.
The purpose of creating rapport in a coaching or therapeutic context is to get someone to actually do something that will help them, and especially to do that thing sincerely (vs. "trying" to do it, pretending to do it half-heartedly or ironically, etc.). They need to trust that it's safe to do the thing, and know that you're not going to pull something on them or use it against them somehow. Otherwise, they may not commit to it.
This trust is extremely contextual, and to some extent conditional on the shared belief that the thing you're telling them to do will ultimately be good for them. I don't think "love" is a particularly good name for the state, though. If the intervention is successful, it's likely that you will retain respect or a certain fondness for the mentor that helped you with something important. There's a personal connection, yes, but it's the connection of a student to a good teacher, or someone who set you on a good path.
At most, one might compare it to the sensation that happens when trusting someone to belay you in a ropes exercise or catch you in a trust fall. Successfully completing the exercise doesn't magically make you trust and love the person forever.
I would love to hear which pieces of his written work you think of as "actually new or useful insights," the only thing that fits that description for me from his youtube videos is the Six Human Needs, which is a useful template for goal-factoring for me.
Ironically, that's one of the models that he didn't create; it's a refinement on an earlier model by Hyrum W. Smith. (The founder of what became later became Franklin-Covey.)
The written work I had in mind was Awaken The Giant Within, especially his chapters on Questions and Rules. While much of the book isn't terribly original, and many pieces are just a popularization of similar concepts from NLP, his integration of some of those pieces into his "master system" was an advance of the field.
(Remember, though, that ATGW was written about 30 years ago (1992 publication), so many of the ideas that were brand new or barely heard of at the time of its writing, have since become part of the self-help zeitgeist... in large part because of said book.)
This trust is extremely contextual, and to some extent conditional on the shared belief that the thing you're telling them to do will ultimately be good for them. I don't think "love" is a particularly good name for the state, though.
Transference is a relatively common phenomenon in any therapeutic change setting. If a person is working with feelings of love, those often get transferred to the therapist.
Many of the techniques Tony uses I expect to enhance this effect, such as rapid rapport building using anchoring, pattern interrupts, state-pumping, using himself as a model for change, and the general rock concert vibe with himself as the Rockstar.
It's not that these techniques are ineffective, but you have to look at if the change worker is being responsible with these easy to predict effects and has policies in place such as:
Since it seems that Tony in practice does tend to take advantage of Transference and altered states rather than using them responsibly, I would agree with alkjash that there is a tail risk with Tony Robbins seminars.
Very interesting! This thread is the first time I've heard of NLP (might have seen the acronym before but I thought it was ML people referring to Natural Language Processing), I will definitely check it out. I guess I just rounded off my observations to the nearest things I recognized. I'm not surprised that Robbins stuff is embedded in a larger technique but am kind of surprised that I've been ignorant of it for so long.
Is there a book or resource that you would most recommend to learn NLP?
NLP stands for Neurolinguistic Programming -- a spur-of-the-moment name given by Richard Bandler after glancing at the titles of the books in his car when he was stopped by police for speeding, and was asked his occupation. Before that point, it was just a group of students and academics doing weird psychology experiments, after Bandler noticed some common language patterns between certain therapists whose books he was transcribing and editing (one a Gestalt therapist, the other a family therapist), and went to ask his linguistics professor about it.
Bandler later settled on a definition of NLP as, "an attitude which is an insatiable curiosity about human beings with a methodology that leaves behind it a trail of techniques." Which, one might argue, is just another way of saying "Science!"... but the more philosophically-oriented works of the NLP creators spend a lot of time talking about how so much of psychological science at the time (60's and 70's) was "how do we define how fucked-up somebody is", not "what can we do to help".
In contrast, the philosophy of NLP presupposes that people are not broken: whatever it is they're doing, they're doing perfectly according to their programming: a programming that can be understood in terms of internal processing steps (represented in sensory terms), and in terms of people's internal models, or maps of the territory. Behavior that may seem crazy or stupid can thus be understood as straightforward, even rational, when considering both a person's map and the processing steps they are using to think and respond to what they observe.
The Structure of Magic (the first book on NLP, which IIUC was also Bandler's masters thesis) was written to capture something that it appeared that more-effective therapists were doing to change people: specifically, noticing map-territory gaps and getting people to confront those gaps.
Bandler noticed the verbal patterns because he was typing the same kinds of questions and statements over and over, so he consulted the linguistics professor at his college to ask about them, and got help to describe the patterns in linguistic terms (like "lost performatives", "modal operators", and "complex equivalence").
Together, they concluded that the distortions of model-making -- that is, the distinctions between map and territory -- had specific, observable linguistic markers for the information that was being generalized, distorted, or deleted in the process of mental model-making. And that some effective therapists were people who had learned to pick up on these markers and respond to them with certain types of questions, if they perceived that the modeling distortions were relevant to the problem at hand. (Since we all distort things constantly, forcing everything to be specific just grinds all communication to a halt.)
Many of their original classifications and inventions from back in the 70's would be recognizable as LessWrong-style rationality moves. For example, one of Bandler's favorite techniques was to effectively Taboo people's problem descriptions, by telling them he was going to be hired to "have their problem" for them, so that they could have a day off from it, so he needed to know all the details. "How will I know when to start panicking?," he might ask, getting a person to literally coach him on the details of whatever the problem was, in the process eliciting all sorts of falsifiable information about what's going on in the person's model, cues, and behavior, rather than listening to a person's ideas about the problem.
This also illustrates the "methodology" of NLP: empirically observing behavior and taking people's statements literally and seriously, to an autism-like degree. (In fact, when I first read Animals In Translation by Temple Grandin, one of my first thoughts was that its author was basically applying the NLP philosophy to animal husbandry: seeing the details of what was happening from their perceptual and world-modeling point of view, instead of projecting expectations.)
Is there a book or resource that you would most recommend to learn NLP?
It sort of depends on what it is you want to learn. The more academically-oriented materials are things like The Structure of Magic (volume 1 is the more useful one of its two volumes) and Neurolinguistic Programming, Volume 1 (I don't know if a volume two was ever written.) Those are the books with the most formal structure and attempts at making falsifiable claims and coherent theories; most other books by the creators are essentially workshop transcripts of them teaching therapists to do interesting things.
(But then, while Using Your Brain For A Change is a workshop transcript and is mostly about techniques, there is also a fair amount of offhand commentary that describes the philosophy of NLP, with regard to things like empiricism, testing, "going first", doing different things until you find something that works, etc.)
If you have an interest in hypnosis, and the links between it and NLP, then Patterns of the Hypnotic Techniques of Milton H. Erickson, M.D. is an interesting contrast to The Structure of Magic, as it is all about the language of deliberate vagueness (which can be very useful in hypnosis and persuasion), while Magic is about the language of specificity. NLPers refer to the models presented in these books as the "meta model" (for specificity and "chunking down") and the "Milton model" (for vagueness and "chunking up").
The meta model should be of particular interest to LessWrongers, since it is a catalogue of patterns that help identify where a person's map (expressed in language) may differ from the territory, and a set of questions that can be used to increase specific understanding. (For example, the move described in The Power To Demolish Arguments is what NLP called "chunking down" (going from abstracts to instances or examples) as opposed to "chunking up" (going from examples to abstracts).
Most other books tend to be catalogues of techniques developed using the theoretical models, philosopy, and methodology described in those books. A lot of these techniques are considered "NLP", in part because Bandler and Grinder did workshops to teach a bunch of the interesting things their experimentation club came up with, and called it an NLP practitioner certification. (Tony Robbins took an early course in NLP and then wrote his book Unlimited Power based on that and some other stuff.)
In general, though, popular works on NLP tend to have little connection or relevance to the core ideas, philosophies, or methods, and instead focus on specific ideas or techniques in a particular area of application. (Which is a bit like going around writing books on calculation tricks for how to target artillery shells effectively, and calling what you're writing about "Science". It might even be technically correct, but it's terribly misleading!)
In addition, they often promote concepts the NLP creators themselves discarded or updated decades ago, or discard essential information for applying the technique comprehensively. (The main missing ingredient usually being what the CFAR handbook describes as "Polaris", i.e. aiming at a goal and doing different things until you get it. For example, the NLP "swish" technique is often described in shorthand with a very specific set of visualization parameters, but for best results you actually need to both tune those parameters to the individual and test the results to make sure you didn't miss anything -- steps that are typically omitted in popular descriptions.)
In general, I view NLP as a kind of proto-rationality that was aimed at practical individual improvement, wherein many important tools we use today, were discovered, named, or experimented with. Its history also offers lessons regarding how difficult it is to take something that's rationality-oriented and bring it to a larger audience without losing the very thing that made it useful in the first place, and why it's important to focus more on what predictions a theory makes vs. how ludicrous the theory itself is. (Paraphrasing something the NLP developers said once, "Everything we're going to tell you is a lie. Some of these lies will be more useful than others.")
Fwiw I haven't found his list of human needs more insightful than other lists of needs like Maslow, Self-determination theory, Daniel Pink's, etc.
I agree that most of his written work isn't particularly insightful, and is just a mash of NLP and Behaviorism. New stuff like money, master the game seems moderately useful but too dumbed down.
I didn't consider the opposite danger of CFAR immunizing against other forms of self-improvement, it seems like it wouldn't be too strong of an effect but much less bad in any case?
Yeah I guess it depends on what you're trying to measure. From an individual perspective getting sucked into a cult for a few years is much worse, but from a collective perspective a smart EA or rationalist operating at only 50% of their counterfactual impact is probably much worse.
Hey! Could you say more about a causal link between Sequences and writing these papers, please:
I was able to do from muscle memory certain calculations about conditional probability and expectation that might have taken weeks otherwise (if we figured them out at all). I attribute this ability in large part to reading the Sequences.
I think my confusion comes from (a) having enough math background (read some chapters of The Probabilistic Method yers ago); (b) while reading Sequences and more so AF discussions added to my understanding of formal epistemology, I am surprised that your emphasis how Sequences affected your muscle memory and ability to do calculations.
To be clear, the papers would almost certainly have gone through anyway, the helpful thing was being very comfortable with Bayes rule and immediately noticing, for example, that conditioning on an event with probability 1-o(1) doesn't influence anything by very much.
Another trick I derived from this comfort is to almost never actually condition on small-probability events. Instead, the better thing to do is to modify the random variables you care about to fail catastrophically in the small probability scenario.
For example, in graph theory I might care about controlling a random variable X which is the number of times a certain substructure appears in the random graph G(n,p), but to do so I need to condition away some tail event E like the appearance of a vertex of extremely high degree. Instead of working with conditional probability for the rest of the argument (which might go on to condition away 3 or 4 other tail events), the nicer thing to do is to modify X into a variable X' which is defined to be 0 when E occurs, and reason about X' instead. This is better for multiple reasons; the most important one being that the edge appearances in G(n,p) are no longer independent when you condition on E complement.
I think mostly what I got out of the Sequences was removing an air of mystery around Bayes rule. Here by mystery I mean "System 1 mystery," i.e. that before I read the Sequences, to figure out a conditional probability I would have to sit down and carefully multiply and divide. This post also helped.
I'm seeing a lot of great snappy responses to "How exactly would AI kill us all?" but not as many on "If they actually believed in slowdown they'd just slow down." Here's some attempts:
Mark Cuban has said he specifically does not use a variety of tax avoidance techniques for this reason.
The US government unilaterally dismantled (well, threw them into the process for dismantling...) its' tactical nuclear stockpile in the early 90s. They did the same with their bioweapons in the 1970s.
A lot of people don't use social media for this reason, myself included.
US has used unilateral ceasefires during ukraine and iran wars. Colombian rebels (farc and eln) have used unilateral ceasefires. There are a lot of examples of unilateral ceasefires.
Oh wow! I am actually noticeably confused by this and the other comment!
Some thoughts, not necessarily disagreements:
I consider LW as a noncentral element of "social media" (Fable agrees), so, well akshually you do use social media.
After thinking about it, I agree that the ceasefire one is just wrong.
I have not even considered that one would pay extra voluntary tax. I am confused about your conflation of "not exploiting tax loopholes" with "paying extra voluntary tax." Separately, I would guess that the vast majority of people who genuinely believe in a higher marginal tax rate do not unilaterally pay voluntary tax, and I do not consider them to be hypocrites.
I draw a distinction between viral platforms that seek users then shovel them algorithmic content; and issue specific forum communities that attract humans who want to engage on the topic, like lesswrong. Though this arguing about the definition of words, so I'll accept your frame and admit to some amount of hypocrisy.
For taxes, I assert that there is a continuum between 'sending a check to the treasury for fun' and 'tax protestor who goes to prison before paying'. I think 'declines to take advantage of all available tax strategies' is somewhere on that continuum.
It's rare that someone would just send a check to the Treasury, but maybe a better example than Mark Cuban would be Timothy Mellon's public attempt to send 130m to the US government to pay military salaries during the shutdown. Was it a loan or a donation? I don't know, but it probably supports my thesis.
As someone who needs convincing; I agree with 3/4 of those. Nuclear is the only one that's clearly wrong because then your adversary can just bomb you but yes "If you actually believed in higher marginal tax rate, you'd just pay extra voluntary tax." 100%
Obviously you disagree, I'm just saying that they don't convince me (a person who needs convincing)
It's human intelligence that's jagged.
I'm getting kind of tired of the notion that AI capabilities are somehow intrinsically jagged, where jaggedness is always drawn against the perfect circle of human intelligence. Actually it feels to me that AI capabilities are jagged, but much less jagged than human capabilities, according to any reasonable heuristic like "base computational difficulty of problem."
Obvious examples of human jaggedness:
*sometimes it's a pile of clothes in the closet but hey it's a hard problem
My last two posts seem to be the most unpopular things I've ever written. Are they actually unusually bad somehow or is it just the LLM blocks turning people off?
I use LLMs for some fact-checking and line-editing at the end of each writing session; both posts detect as 100% human-written in pangram. Would it be dishonest to just drop the LLM content blocks?
For https://www.lesswrong.com/posts/JrmLzrCqtpXSvjeDj/why-focus-on-extinction, I didn't downvote, but I glanced at it, saw that it was a link post and that it looked like a claude prompt and response, and went on with my life.
There may be something interesting there, but a statement of your POV and conclusion, and a bit more structure to the exploration/argument is necessary to make me suspect it's worth the effort to read it.
Sorry, just to reiterate, the LLM generated few of the words but did help with some minor editing.
My AI safety essay was flagged for LLM content (I think because I had Claude generate the references and links), so I started putting everything I write in LLM content blocks to not deal with the headache.
I think this was an overreaction.
Hmm. I didn't look that closely at the formatting/generation - my reaction wasn't "this is AI, and I hate it", it was "this is not worth reading, and it looks like AI is the cause", which may mean I'm the wrong audience and you should ignore my comments. Or it may mean that the content actually isn't compelling, separate from how much AI was used for it.
I see you've edited it, so it's clearer that it's your voice. That's good, but it still doesn't match my interests for the site. It doesn't clearly state a position or exposition of facts, or even a non-trivial model for thinking about something. Or maybe it's about "how to convince a norm" rather than about "what is Altman's strategy" or "what are my beliefs" or "how should you form beliefs here". I am less interested in the messaging/advocacy topic, and it's not clear without reading closely that this is what it is. And my caveat above applies even more: I'm only one reader, others have different criteria.
A better title and thesis statement (or even summary at the bottom) would perhaps make it more useful for people who are interested, and easier to ignore for those who aren't.
For the first post, the LLM comment block made me assume it was the pure unedited output of Claude and stop reading.
I assumed the blocks were for substantively LLM-written prose (as would be flagged by Pangram), not "was an LLM consulted in any way whatsoever during the production process."