faul_sname
Fuck around once, find out once. Do it again, now it's science.
No bio...
User ID: 884
faul_sname
Fuck around once, find out once. Do it again, now it's science.
3d ago
Same. As is likely true for most people living in large metro areas in the US.
faul_sname
Fuck around once, find out once. Do it again, now it's science.
3d ago
Whereas if I see a mountain lion in my US neighborhood, I have something mildly unusual to post about on nextdoor. Occasionally a mountain lion even makes its way into the heart of a major city, though that makes headlines when it happens.
faul_sname
Fuck around once, find out once. Do it again, now it's science.
6d ago
More than zero rationalists are extroverted. Not gonna dispute the NT part of that though.
faul_sname
Fuck around once, find out once. Do it again, now it's science.
8d ago
Can a robot turn a canvas into a beautiful masterpiece? (Yes)
Can an orangutan? (No)
[...] I'm also going to send out a bat signal for @faul_sname to chime in and correct me if I'm wrong.
This is actually an area of active debate in the field.
Shitpost aside this seems reasonable to me, aside from a few quibbles
- RLVR is absolutely not only a year old -- you can trace back the core idea REINFORCE paper from 1992. RL from non-verifiable rewards (e.g. human feedback) is actually the more recent innovation. But the necessary base model capabilities and training loop optimizations and just general know-how and tooling for training a model that speaks English and writes good lean proofs was just not there until quite recently.
- How important the static model problem is is very much a subject of active debate, but I come down quite strongly on the side of "it's real and AI agents are going to be badly hobbled until it's solved". An analogy I've found compelling but lost the source on is that current "agentic" AI approaches are like trying to take a kid who has never touched a violin before and give them sufficiently good instructions before they touch the violin that they can play Paganini flawlessly on their first try, and then if they don't succeed on the first try kicking the kid out, refining your instructions, and then bringing in a new kid.
Intelligence is the general-purpose cognitive ability to build accurate models of the world and then use those models to effectively achieve one's goals
I basically endorse this definition, and also I claim current LLM systems have a surprising lack of this particular ability, which they can largely but not entirely compensate for through the use of tools, scaffolding, and a familiarity with the entirety of written human knowledge.
To your point about the analogy of the bird that is "unintelligent" by the good swimmer definition of intelligence, LLMs are not very well adapted to environments that humans navigate effortlessly. I personally think that will remain the case for the foreseeable future, which sounds like good news except that I expect that we will build environments that LLMs are well adapted to, and humans won't be well adapted to those environments, and the math on relative costs does not look super great for the human-favoring environments. Probably. Depends a bit on how hard to replicate hands are.
faul_sname
Fuck around once, find out once. Do it again, now it's science.
10d ago
Yet AI skeptics tend to make moving the goalposts into the entire sport. I will grant that their objections exist in a range of reasonableness, from genuine dissatisfaction with current approaches to AI, to Gary Marcus's not even wrong nonsense.
I may or may not be an AI skeptic by your definition - I think it's quite likely that 2030 is a real year, and think it's plausible that even 2050 is a real year. But I think there genuinely is something missing from today's LLMs such that current LLMs generally fail to exhibit even the level of fluid intelligence exhibited by the average toddler (but can compensate to a surprising degree by leveraging encyclopedic knowledge).
My sneaking suspicion is that the "missing something" from today's LLMs is just "scale" - we're trying to match the capability of humans with 200M interconnected cortical microcolumns with transformers that only have 30k attention heads (not perfectly isomorphic, you could make the case that the correct analogy is microcolumn : attn head at a particular position, except the microcolumns can each have their own "weights" whereas the same attn head will have the same weights at every position), and we're trying to draw an equivalence between one LLM token and one human word. If you have an LLM agent that forks a new process in every situation in which a human would notice a new thing to track in the back of their mind, and allow each of those forked agents to define some test data and fine-tune / RL on it, I bet that'd look much more impressive (but also cost OOMs more than the current stuff you pay $200/mo for).
This is an interesting concern, and I mean that seriously. Fortunately, it doesn't seem to be empirically borne out. LLMs are increasingly better at solving all bugs, not just obvious-to-human ones.
LLMs are increasingly better at solving a particular subset of bugs, which does not perfectly intersect the subset of bugs which humans are good at solving. Concretely, LLMs are much better at solving bugs that require them to know or shallowly infer some particular fact about the way a piece of code is supposed to be written, and fix it in an obvious way, and much much worse at solving bugs that require the solver to build up an internal model of what the code is supposed to be doing and an internal model of what the code actually does and spot (and fix) the difference. A particularly tough category of bug is "user reports this weird behavior" - the usual way a human would try to solve this is to try to figure out how to reproduce the issue in a controlled environment, and then to iteratively validate their expectations once they have figured out how to reproduce the bug. LLMs struggle at both the "figure out a repro case" step and the "iteratively validate assumptions" step.
I don't see this as a major impediment, why can't LLMs come up with new words if needed, assuming there's a need for words at all?
In principle there is no reason LLMs can't come up with new words. There is precedence for the straight-up invention of language among groups of RL agents that start with no communication abilities and are incentivized to develop such abilities. So it's not some secret sauce that only humans have - but it is a secret sauce that LLMs don't seem to have all of yet.
LLMs do have some ingredients of the secret sauce: if you have some nebulous concept and you want to put a name to it, you can usually ask your LLM of choice and it will do a better job than 90% of professional humans who would be making that naming decision. Still, LLMs have a tendency not to actually coin new terms, and to fail to use the newly coined terms fluently in the rare cases that they do coin such a term (which is probably why they don't do it - if coining a new term was effective for problem solving, it would have been chiseled into their cognition by the RLVR process).
In terms of why this happens, Nostalgebraist has an excellent post on how LLMs process text, and how that processing is very different from how humans process text.
With a human, it simply takes a lot longer to read a 400-page book than to read a street sign. And all of that time can be used to think about what one is reading, ask oneself questions about it, flip back to earlier pages to check something, etc. etc. [...] However, if you're a long-context transformer LLM, thinking-time and reading-time are not coupled together like this.
To be more precise, there are 3 different things that one could analogize to "thinking-time" for a transformer, but the claim I just made is true for all of them [...] [It] is true that transformers do more computation in their attention layers when given longer inputs. But all of this extra computation has to be the kind of computation that's parallelizable, meaning it can't be leveraged for stuff like "check earlier pages for mentions of this character name, and then if I find it, do X, whereas if I don't, then think about Y," or whatever. Everything that has that structure, where you have to finish having some thought before having the next (because the latter depends on the result of the former), has to happen across multiple layers (#1), you can't use the extra computation in long-context attention to do it.
So there's a sense in which an LLM can coin a new term, but there's a sense in which it can't "practice" using that new term, and so can't really benefit from developing a cognitive shorthand. You can see the same thing with humans who try to learn all the jargon for a new field at once, before they've really grokked how it all fits together. I've seen it in programming, and I'm positive you've seen it in medicine.
BTW regarding the original point about LLM code introducing bugs - absolutely it does, the bugginess situation has gotten quite a bit worse as everyone tries to please investors by shoving AI this and AI that into every available workflow whether it makes sense to or not. We've developed tools to mitigate human fallibility, and we will develop tools to mitigate AI fallibility, so I am not particularly concerned with that problem over the long term.
faul_sname
Fuck around once, find out once. Do it again, now it's science.
12d ago
faul_sname
Fuck around once, find out once. Do it again, now it's science.
1mo ago
Don't container ships usually travel much, much faster than barges?
faul_sname
Fuck around once, find out once. Do it again, now it's science.
1mo ago
... and that's bad, right? You see how the "think of the children" argument shut down critical thinking in this case?
The happens every time. Saying "think of the children" to mandate certain medical treatments does not go well. Saying "think of the children" to ban those same medical treatments will also probably not go well. Most puberty blockers don't go to trans kids, and most trans kids don't use puberty blockers, so the second-order harms of a bad policy here are likely to be larger than the primary benefits.
That leaves the door open for a good policy of course. Ha ha ha.
faul_sname
Fuck around once, find out once. Do it again, now it's science.
2mo ago
Sorry, I might be missing something, but I honestly cannot grasp how I should shrug at the extremely poor quality of evidence for prescribing puberty blockers, because it hasn't been prescribed to that many kids, but shriek in horror at the suggestion of banning them, because the number is growing
It has not been prescribed to that many kids for purposes other than delaying extremely precocious puberty. Puberty blockers are in fact useful at blocking precocious puberty, which most people agree should be blocked, and that is the primary thing they're used for. The rate of precocious puberty happening is growing rapidly, and so banning the only effective means of mitigating the problem we have would be a moderately large problem today and a much larger problem in the future.
What is incoherent about [the social contagion hypothesis]? We do have other references for phenomena that are social contagion for sure, because no one has been abducted by aliens. Are these hypotheses incoherent too?
What observations does the social contagion hypothesis exclude? If there is some evidence that would lead you to think that "social contagion" is more likely to be the correct explanation, there must be some other evidence which would lead you to decrease your credence in that hypothesis. What specific evidence would that be, in your case?
The internet is a thing these days. People can read, watch Netflix shows with capital "D" diversity up the wazoo, etc. There's parasocial effects stemming from following influencers. Subreddits, Discords. Sorry, but this is pure cope.
The specific observations I make here about MTF people (I have many more MTF than FTM in my social circle, can't speak to FTM)
- Are interested in functional programming and also Rust
- Watch particular types of anime
- Are interested in mathematics, particularly category theory
- Listen to very particular obnoxious anime music (sped up electronic remixes with high pitched voices)
- Play the game touhou (specifically)
I suspect you can make a quite accurate FTM risk score with a linear score over how many of these descriptions apply to a particular person. Now it is possible that this just indicates social spread through this particular demographic, but it very much feels like there's a "type of person" who is into all of these things. Specifically, it feels like the type of person who was at risk of being a furry in the early 2000s or a ham radio operator in the 1980s.
If mere exposure to trans was the primary explanation, I would expect the normal FTM demographic to instead look like normies who happen to like stuff which portrays a lot of trans people, e.g. I'd expect them to
- Listen to lots of Kim Petras
- Watch Orange is the New Black and Euphoria
- Attend drag shows and musical theater
- etc
As such I don't think "this is a purely social phenomenon, and it is only by chance that it spread through this particular group of people" is a parsimonious hypothesis, at least for the MTF demographic (again, I don't really know much about the FTM demographic, maybe they do look like normies who were convinced to become trans by positive portrayals in mass media. I doubt it, but that doubt isn't really informed by anything).
I resent overriding my instincts for a lie, which is why I'm so invested in pointing out that the liars have, in fact, been lying.
That is fair and valid and also not a very good basis for making policy about what medical treatments should be forbidden. It is a good basis for deciding who to listen to in the future for general policy stuff - my objection is narrowly scoped to having policy people make uninformed broad sweeping decisions about medical treatment, because that does not have a history of going great.
faul_sname
Fuck around once, find out once. Do it again, now it's science.
2mo ago
Also, if blockers aren't such a big deal, then let's just ban them. After all very few people would be affected by the ban.
Argh no this is exactly the reaction I am worried about from the people who want to ban puberty blockers because trans. About 20k kids a year enter puberty extremely early (before 8 for girls, before 9 for boys, sometimes much before). This number is going up extremely rapidly over time. You could perhaps ban puberty blockers for kids over a certain age but I am not confident there wouldn't be substantial negative effects from that, and I bet you aren't either, because this is not our field of expertise. If I expected that, conditional on legislation existing here, the legislators would consult with pediatricians and write the legislation to actually be sane and minimally scoped, I wouldn't raise this objection, but I don't expect that legislation in this area (or any area really) would be sane and minimally scoped.
Just look at some of the other contagions - the Satanic Panic, alien abductions, recovered memories, anorexia - the incidence of these, and many other things, increase with coverage from the media, and there's no evidence they increase with the actual phenomenon increasing in frequency, or at least I'm yet to see evidence of an actual alien abduction. So I see no reason to assume that gender dysphoria is any different.
Calling both alien abductions and recovered memories "social contagion" is exactly the sort of thing I'm talking about when I say that "social contagion" is a non-explanation. Those had very different causes from each other. Effective interventions aimed at reducing the incidence of false recovered memories (e.g. "the APA stops endorsing dream interpretation therapy") would probably not help very much for the alien abduction craze.
With gender dysphoria we see that there is a particular personality (two personalities, actually: mtf and ftm seem like two pretty much non-overlapping personality groups) that is much, much more prone to it than others. My understanding is that, to the extent that "social contagion" is a coherent hypothesis at all, it predicts that trans hit these groups hardest just because it reached those social groups first and then spread within them, and if it had started in a different social group we would see a different distribution over trans frequencies by personality type. This strikes me as unlikely, especially since "antisocial and lonely" is a risk factor for the mtf group.
Social contagion is exactly one of the explanations offered for this. It says that what leads to discomfort with being embodied is talking about discomfort with being embodied, especially when you glamorize it during the conversation.
I agree it's not related to any discussion on gender affirming care. I'm quite puzzled why you'd think otherwise, actually.
Base rate. Lots of people care start caring about things when they become personally affected by them, and orders of magnitudes more people were personally affected by the pronoun craze and the corresponding threats to livelihood and job security than even know someone personally affected by puberty blockers.
faul_sname
Fuck around once, find out once. Do it again, now it's science.
2mo ago
Most of the 1.5 million trans people in the US are not minors, and the fraction of the trans minors who go on puberty blockers is about 3%, not "only 10%". Unless you're talking about what they do once they reach adulthood, but if you want to forbid adults from doing things they want to do with their bodies, trying to add regulations around what kids can do probably won't help.
faul_sname
Fuck around once, find out once. Do it again, now it's science.
2mo ago
I would say "more attention than puberty blockers", because the number of affected kids is much higher. Something caused a massive uptick in either the experience of dysphoria, the reaction to dysphoria, or some combination thereof. I think "social contagion" is a thought-terminating non-explanation here. To reduce the rate of trans identification, I think it would be worth looking into what generally leads to discomfort with being embodied (as that seems to correlate extremely strongly, and also seems to be much more common than it used to be).
Of course, if you don't actually care about that and your main objection is to "point deer say horse", that is perfectly valid. But in that event I also don't take statements of concern about puberty blockers at face value, and will discount your policy suggestions in that area accordingly.
faul_sname
Fuck around once, find out once. Do it again, now it's science.
2mo ago
... I don't think puberty blockers are as damaging as lobotomies, and also puberty blockers in the context of gender affirming care are like 10x less frequent than lobotomies were at the peak of that craze. I think people 50 years from now (assuming the world of 50 years from now substantially resembles the world of today) will probably think of it similar to how we think about high schoolers smoking or using tanning beds (i.e "basically not at all").
faul_sname
Fuck around once, find out once. Do it again, now it's science.
2mo ago
Or loses a malpractice suit when they do malpractice, yeah. Again, 40k kids a year start "gender affirming care", only 1k of those 40k start puberty blockers. I really don't think puberty blockers warrant special attention here.
faul_sname
Fuck around once, find out once. Do it again, now it's science.
2mo ago
I don't think the once-ascendant ideology is particularly entrenched anymore. The anti-woke/anti-trans movement at this point feels very similar to the way the atheism community felt in the 2012 era, as they ran out of defensible causes and started to turn to indefensible causes and on each other. Had they packed up and gone home once their original goals were met (e.g. no prayer in schools) I think the world would be a happier place.
faul_sname
Fuck around once, find out once. Do it again, now it's science.
2mo ago
Lots of societies have had to deal with some folly of youth causing some number of kids to ruin their lives in one way or another in their quest for status and acceptance. In ancient Rome, kids seeking social status joined gladiatorial schools, and many of those kids ended up crippled or dying. In Victorian England, girls wore incredibly tight corsets which caused reduced lung capacity, skeletal deformations, and abdominal muscle weakness, which led to lots of health problems (including much higher chances of miscarriage or death in childbirth).
Just because something is a problem doesn't mean a political solution exists. The politician's fallacy ("We must do something. This is something. Therefore we must do this") is frequently cited as a fallacy due to the third line, but the first line is often also wrong - we don't actually have to try to solve every problem.
I think the use of puberty blockers is a problem of small enough scale and low enough severity that it's probably better to just let it ride.
faul_sname
Fuck around once, find out once. Do it again, now it's science.
2mo ago
Me personally? No. If the US government had given a single known sociopath a license to kill 5 people, though, trying to get them to change that decision would not be a very high priority for me.
faul_sname
Fuck around once, find out once. Do it again, now it's science.
2mo ago
faul_sname
Fuck around once, find out once. Do it again, now it's science.
2mo ago
Rephrasing - is it a big enough problem that the disease of having ~1k kids/year go on puberty blockers is worse than the "cure" that would be implemented by the political apparatus would be? Being realistic about what historical political "solutions" have looked like.
faul_sname
Fuck around once, find out once. Do it again, now it's science.
2mo ago
I do see the difference, but moral panics over "think of the children" have a history of having the reactions be cures that are worse than the disease, and I see no particular reason to think that this time is different. Do you have a reason to think that this time is different?
faul_sname
Fuck around once, find out once. Do it again, now it's science.
2mo ago
Look like puberty blockers were prescribed for trans reasons to about 1400 kids in 2021, with that number increasing by about 200 kids / year. Puberty blockers were additionally prescribed to about 20,000 kids in 2021 for central precocious puberty (puberty starting before age 8 for girls or age 9 for boys).
As a point of comparison, about 3100 teens between the ages of 12 and 19 died in car crashes in 2021.
Is there a reason you think that puberty blockers, specifically, are a big problem?
faul_sname
Fuck around once, find out once. Do it again, now it's science.
2mo ago
At all. If you cared about corruption by anyone as much as you claim, you should already have investigated the claims against the previous administration, and you would have had no choice but to conclude that it at least looks fishy, and therefore you would have investigated it and you would now have bulletproof arguments that it wasn't corruption
Which specific claims are you talking about here? Hunter Biden? Stolen election? Biden's "fuck all y'all I'm pardoning everyone" end-of-term pardons? The congressional insider trading thing? Or is there some other specific, credible, and concrete accusation of corruption that you are referring to?
faul_sname
Fuck around once, find out once. Do it again, now it's science.
2mo ago
I'd say not as good as 2014 era commentariat but better than 2019 era commentariat.
faul_sname
Fuck around once, find out once. Do it again, now it's science.
2mo ago
To be "hundreds" there would have to be at least ~25 writers you like as much as Taibbi. Which is definitely plausible, especially if they're less prolific, but I bet that means you have good recommendations. Would you be up for sharing a list of 10 or so writers you like as much as Taibbi, with like a sentence about why you like them?
Let's say you currently live in Botswana. You could move to Lesotho, where you'd be 95th percentile, or to America, where you'd be 40th percentile. I think most people in that situation would still choose America, even though there's no substantial difference in switching cost between the two options.
More options
Context Copy link