All of George's Comments + Replies

Boring machine learning is where it's at

Got Any examples of this being used? I'm always on the lookout for these kind of usecases.

Covid 10/21: Rogan vs. Gupta

Precisely, the mask thing keeps the tourturous forced herding and makes it even worst, it's inhumane.

Covid 10/21: Rogan vs. Gupta

how much protection would allowing children that have at-home parents to stay at home and only come to school optionally for recess to play with other kids give you? A lot more

Throw in being able to go home if they start sneezing, or feel like it.

Covid 10/21: Rogan vs. Gupta

What's the chance that the whole medical research apparatus in the US is corrupt and the vaccine numbers are off? That's the bit that might be underlying Rogan's model which he's not spelling out, that leads to "I don't trust statistics"

Is it 1% ? 10%? 20%? 90%? If so the unkown-unkowns of a vaccine become very high.

Given that it's widely accepted both Russia and China have faked their data, given that he knows people which had issues that were not reported, given previous failures of the health system during the pandemic.

It seems to me that Joe (and most ... (read more)

1tkpwaeub1moRight. It's hard to have a rational argument without an agreed upon set of facts.
Boring machine learning is where it's at

I don't disagree, as I said before, I'm focused on problem type not method.

The fact that human mimicking problems have loads of cheap training data and can lead to interesting architectures is something I didn't think of that makes them more worthwhile.

Boring machine learning is where it's at

Hmh, I didn't want to give the impression I'm discounting particular architectures, I just gave the boosting example to help outline the target class of problems.

Boring machine learning is where it's at

The main issue with surveillance is the same AI vs automation issue.

Once the population is fine with being tracked 24/7 and you have the infrastructure for it:

  1. The need to do so diminishes
  2. You can just make it illegal to not carry a wifi-enabled ID (there's the issue of having someone elses IDs, but so is there the issue of face-covering, and finding incentives to force people into else'scarrying only your own ID is easy)
Boring machine learning is where it's at

I think I was wrong in writing this, and I corrected it on my blog.

What I mean to say was closer to "human mimicking CV" (i.e. classification, segmentation, tracking and other images -> few numbers/concepts tasks). There's certainly a case to be made that image-as-input and/or output techniques as a whole have very large potential, even if not actualized

The LessWrong Team is now Lightcone Infrastructure, come work with us!

Looking forward to seeing projects that come out of this, the LW UX is certainly the most impressive thing about it and one of the few examples of "modern web design" done well that I've seen.

Not being remote seems really weird given the economics of it, but to each their own, I guess.

2ChristianKl2moOne of their projects seems to be to build an in-person campus. For that it's helpful if the whole team is in one place.
Contra Paul Christiano on Sex

In sexual reproduction there would be a large pool of correct copies out there and at some point these would be swapped back into this line. With cloning the information is lost for all descendants until random mutation recreates it. 


I think I get your point here, though I think this assumes a lot about how much cross-over mechanisms can actually "detect" genetic damage.

If this damage can mostly be detected only once the organism is mature enough to be selected for/against by "environment" then I think that kind of goes back into the "red queen"... (read more)

Contra Paul Christiano on Sex

How so? single-chromosome mutations can account for all variations one gets from the opposite sex, bad configurations can be selected against inside the germinal cells themselves or when the new organism is just a clump of a few thousand cells, which is how most "really bad" configurations get selected against in sexual organisms too.

2DanArmak2moMany genes and downstream effects are only expressed (and can be selected on) after birthing/hatching, or only in adult organisms. This can include whole organs, e.g. mammal fetuses don't use their lungs in the womb. A fetus could be deaf, blind, weak, slow, stupid - none of this would stop it from being carried to term. An individual could be terrible at hunting, socializing, mating, raising grandchildren - none of that would stop it from being born and raised to adulthood. There's no biological way to really test the effect of a gene ahead of time. So it's very valuable to get genes that have already been selected for beneficial effects outside of early development. That's in addition to p.b.'s point about losing information.
1p.b.2moLet's say there is a section in a chromosome with 10 genes. In one chromosome 8 of these have damaging mutations. In the other chromosome these 8 are good copies but the other two are damaged. Now crossover of that section could fix the first chromosome by replacing 8 bad copies with 8 good copies and only 2 good copies with 2 bad copies. But going forward the resulting organism only has bad copies of these two genes. In sexual reproduction there would be a large pool of correct copies out there and at some point these would be swapped back into this line. With cloning the information is lost for all descendants until random mutation recreates it. Positive mutations would have to achieve for each germline what in sexual reproduction they have to achieve for just a few members of the entire species.
Contra Paul Christiano on Sex

having pairs of chromosomes and crossover are sufficient to resolve it

1p.b.2moDo you mean cloning instead of sexual reproduction but with two chromosomes and crossover? That wouldn't be enough to avoid mutational meltdown.
Contra Paul Christiano on Sex

One is genes your (adult) partner has that are different from yours. The other is additional de novo mutations in your partner's gametes.


Neither of which are guaranteed to yield viable offspring, the latter won't carry all or maybe any of the benefits when mixed with your genes. Indeed, chances are most won't.

On the other hand just getting random mutation on a constant set of genes seems like it has a much higher chance of still yielding a viable combination.

The former has already undergone strong selection, because it was part of one (and usually ma

... (read more)
1p.b.2moI think Dan is correct. Sex is necessary to avoid Müller's ratchet if you can't have a gazillion offspring. Müller's ratchet is not avoided by introducing more variance (at least that's a very weird way to look at it). It is avoided by getting "undamaged" genes for your "damaged" ones and "fixing" your chromosome by recombination. (Of course this only happens randomly, but then there's selection on top.) The strategy of trying many mutations in many offspring just doesn't work for very complex organisms. And higher mutation rates just speed up Müller's ratchet. Without sex and recombination you'd need an insane amount of selection to counteract mutation.
Contra Paul Christiano on Sex

Sex being stable i.e. lack of gender switching or hermaphrodites in most animals (with exceptions ala snails).

Also, the recombinant benefits seems like they could also come by via increased varrisnce from e.g. mutation and 'survival of the fittest' style mechanism.

[Personal Experiment] One Year without Junk Media: Six-Month Update

Hmm, I did this by default for a while (gave up TV 10 years ago, vidyogames 5, news and YouTube ~3).

What I discovered last year is that there's certainly a thing I'm "missing out on" by doing this, in that (dumb) movies/cartoons/yt-videos can help with neurosis in particularly bad periods, and I tentatively reintroduced them for short period every few months, in all cases I did this when I was really sick (as in, physical injuries)

So I do think there's a point where renouncing all media can backfire.

Also I'm torn if e.g. less wrong, reddit, Twitter, hacker... (read more)

I wanted to interview Eliezer Yudkowsky but he's busy so I simulated him instead

Given the alarmist and uninformed nature of LW's audience, it might be wise to demand the source code (in this case none, pressumably and API was used) and methodology used to generate any such content.

In this case it seems kind of obvious the author either wrote both sides and/or cherry picked a lot. All fun and games, but you have an AGI death cult going here and this kind of thing can be fodder for false beliefs that stochastic parrots are infinitely more powerful than what's experimentally proven thus far.

This comment did not deserve the downvotes; I agree with asking for disclosure.

It does deserve criticism for tone. "Alarmist and uninformed" and "AGI death cult" are distractingly offensive.

The same argument for disclosure could could have been made by "given that LW's audience has outsized expectations of AI performance" and "it costs little, and could avoid an embarrasing misunderstanding".

I upvoted you because you caused this response to be generated, which was informing to read, and I like informative things, and whatever generates informative things can't be all bad <3

Thank you for that! :-)


However, I strongly disagree with your claim that LW's audience is "uninformed" except in the generalized sense that nearly all humans are ignorant about nearly all detailed topics, and: yes, nearly all of the contributors to Lesswrong are humans and thus ignorant in general by default.  

Based on my personal experiences, however, most people... (read more)

I strongly agree that the methodology should have presented up front. lsusr's response is illuminative and gives invaluable context.

But my first reaction to your comment was to note the aggressive tone and what feels like borderline name-calling. This made me want to downvote and ignore it at first, before I thought for a minute and realized that yes, on the object level this is a very important point. It made it difficult for me to engage with it.

So I'd like to ask you what exactly you meant (because it's easy to mistake tone on the internet) and why. Cal... (read more)

Source code: None. I used OpenAI's GPT-3 playground.

"Wrote both sides": No. I only wrote my side.

"Cherry picked": Yes. I frequently generated several responses and then kept the best ones.

Long Covid Is Not Necessarily Your Biggest Problem

Tangential but: sorry to hear about the nerve endings damage thing :/ I got something like that because I made the (in hindsight potentially dumb) move to pull out a wisdom tooth and adjecent tooth.

Dentist was not the best and the tug fucked up a nerve.

Not as bad as your case, but occasionally it gives a very "creepy" feel and a bit of pain.

What I can stay is that being mindful to the pain always seems to drive it away, this doesn't work with other types of pain. So unlikely but if you can't find anything else, it might be worth a shot to see if close examination can "cure" it.

Humanity is Winning the Fight Against Infectious Disease

Humanity is misdiagnosing infectious diseases. Statistically speaking you have a few dozen HPVs and say 0.7 HSVs infecting you. Like, permanently, your immune system can't destroy them, they lay dormant in immunno-privileged areas.

How many other viruses do we misdiagnose as "not there" when really they just act in an undetectable way? hundreds? thousands? millions? I don't think there's a good estimate out there

Bacteria and fungi that cause superficial infections enough to an immune reaction to leaving you with localized genetic and structural damage?

Like ... (read more)

7orthonormal3mo1. Overall mortality mortality and morbidity rates don't lie. You can't do enough creative accounting to hide vast amounts of infectious disease mortality within a much longer healthy lifespan. (And yes, it's healthier on average, even counting obesity. Painful disability was more common in the past.) 2. The nice thing about sequencing is that eventually it'll be feasible to take a slice of your tissue and identify everything that's not you. Easier to make progress at that point. 3. Replacing your cells with nanotech cells that bacteria/viruses/prions/etc can't crack, and which have very solid checksum/error-correcting codes to prevent things like nanotech cancer... you're safe against any non-intentionally-designed attack. (To say nothing of the abstraction layers possible with uploading.) "Normal" transhumanist technologies aren't perfect, but they are barely epsilon-susceptible to natural infectious diseases.
3DanielFilan5mopost now has correct timestamps for the non-ad version for what I mean.
2DanielFilan5mooops, forgot that I didn't have to deal with ads
[Link] Musk's non-missing mood

I'll make an analogy here as to get around the AI-worship induced gut reactions:

I think most people are fairly convinced there isn't a moral imperative beyond their own life, as in, even if behaving as if your own life is the ultimate driver of moral value is wrong and ineffective, from a logical standpoint it is, once your conscious experience ends everything ends. 

I'm not saying this is certain, it may be that the line between conscious states is so blurry that continuity between sleep and awakenes is basically 0, or as much as that between you can ... (read more)

Reframing the evolutionary benefit of sex

Oh, I though they had to be hand picked to some extent to get there, in that case I guess it makes sense.

I still hold my original cirticism of the ideas presented but that's unrelated.

Reframing the evolutionary benefit of sex

Mutations of germline cells come with huge fitness penalties. Taking 0.01% of your genes introduces an extremely small amount of variance.  And unilaterally replacing 50% of my genes with yours is equivalent to a 50% drop in fitness

99.9% of those genes will be identical, hence why I used the 0.01% number, if you want to induce more mutations you can (see bacteria) and if you want to introduce more mutations in a controlled way (i.e. not break anything important) you also can, and humans actually already do this, as do most multi-celular eukaryotes (se... (read more)

Reframing the evolutionary benefit of sex

As in, given that the information here is seemingly false, I'm unsure why it appeared under the "Recommended" section of the website.

4habryka5moAh, that section is a random sample of posts that you haven't read, weighted slightly exponentially by their karma. So any frontpage post can show up in that section.
Reframing the evolutionary benefit of sex

I'm not sure why this is pinned. This seem like a wrong explanation.

You can get varriance with mutations of germinal cells, and indeed this does happen. Not to mention gene transfer (e.g. I'll keep 99.99% of mine and take 0.01% of yours), which over the last decade has been observed in hundred of eukaryotes.

Sex in primitive species is not a varriance inducing mechanism it's a hidden-trait preserving mechanism.

The chromosomal setup that allows for sex means we can have hidden traits that only show up in a small % of offsprings (e.g. simple story is you need... (read more)

2paulfchristiano5moMutations of germline cells come with huge fitness penalties. Taking 0.01% of your genes introduces an extremely small amount of variance. And unilaterally replacing 50% of my genes with yours is equivalent to a 50% drop in fitness (!). The bacterial method itself also doesn't seem to work in humans (because you need to have the genes during development, and less importantly you need to spread them throughout your body). So it seems to me like sex adds very significant benefits over these alternatives. Is your view here coming from some quantitative estimate or further reasoning you didn't include in the comment, or is it reflecting the consensus view in some field? In either case, it would be great to see a pointer. If this is just your guess based on the reasoning in the comment, that's fine and I'm happy to leave the argument here (or with your rebuttal).
2habryka5moNot sure what you mean by this?
Can group identity be a force for good?

Yeah, but that's not to say it's not good.

The world would be better off with a lot less violence. But in a world where other people are violent, being so is sometimes good.

Even if no humans were violent we'd still need a minimal level of violence to defend against pests and such.

But the world would still be better off without violence.

I don't know, it seems to me like you're making a category error, but maybe I'm missing something.

2Slider5moPossibility strickler in me notices a claim of "It is impossible to deal with pests without resorting to violence". While doing poisoning and outrigth killlings for pest control is rather easy ethic bar to clear I don't see the inevitability of it. You could have things like plant surfaces being engireered to be repulsive to pests, you could do things like allowing pests to only grow outside of industiralised farming. For a lot of these options the effort extended would overshadow the gains in "ethical" operation. For example in a very simple view of law enforcement the police just straight up murder bad guys. But for a more nuanced and complex system, use of force is more detailed and actual application of lethal force would be rarely the prescription. There is a important line between "policing involves use of legitimised state violence" vs "policing will always involve force".
1UnexpectedValues5moI think we're disagreeing on semantics. But I'd endorse both the statements "violence is bad" and "violence is sometimes good".
Can group identity be a force for good?

Is anyone arguing that we shouldn't affiliate with any tribe?

3UnexpectedValues5moI'm not sure. The strongest claim I make in that direction is that "Many in the rationalist sphere look down on tribalism and group identity." I think this is true -- I bet each of the people I named would endorse the statement "The world would be better off with a lot less tribalism."
Can group identity be a force for good?

Tribalism can be good for the individual and for society as a whole.

But it's bad for reasoning about tribe affiliated subjects.

I think you're kinda talking past the point eliazer was making there.

1UnexpectedValues5moTo be clear, I'm agreeing with Eliezer; I say so in the second paragraph. But for the most part my post doesn't directly address Eliezer's essay except in passing. Instead I point out: "Yeah, the 'bad for reasoning about tribe affiliated subjects' is a drawback, but here's a benefit, at least for me."
The Apprentice Thread


  • Book writing - I'm publishing my first technical book this year (no name publisher, they gave me a good deal) and I want to publish a fiction book next year (this one paid out of pocket) | I'd love help from anyone that's gone through the process and published a few books, especially if your starting point was blogging style articles.

  • Trading - I eer on the side of Taleb here in thinking that the chances someone sharing true knowledge on the subject is low. But if anyone would like to help me with automated trading I'd be very happy to lis

... (read more)
2Yoav Ravid5mo[APPRENTICE] Interested in your offer about getting in tech without college. I'm 20, I've learned python on my own, and I'm currently looking for a job without much success. FYI, I am from Israel, so I don't know if it's as relevant.
What are good resources for gears models of joint health?

Oh, not at all, chronic inflammation could be a thing.

I'm just saying that tracking how your exercise routine affects chronic inflammation is a very min-maxy type of things.

Chronic inflammation could very much be a macro problem that leads to joint pain.

Did you already test basic stuff? Like immune cell counts, homocysteine, uric acid, CRP, fibrinogen, ASLO blah. Basically, a standard blood panel an obsessed GP would give you? If not I'd certainly start with that.

I mean, for all you know this thing could be caused by eating too much meat, or gluten intoler... (read more)

What are good resources for gears models of joint health?

I did a little quick searching for "knee stability training protocol" and.. found a few things that looked pretty obvious. Quads, hams, calves, etc. More or less what I'd expect. I don't suppose you have any secret sauce beyond that?


No, it's really very much about your individual body and where you have lacks, you need an in-person trainer to be able to see this and over time as you move more you'll become more aware of your body and be able to say "Ah, it's x area that's too stiff, or activating too much, or that should be working but isn't or whate... (read more)

1Randomized, Controlled5moAre you implying I shouldn't [overly] worry about chronic inflammation? I'm pretty confident there is an inflammatory component here -- icing and voltaran do seem to help. It's been long enough (3 months) that I've started to think about trying a curcumin supplement and/or just cooking with tumeric+black pepper a bunch.
1Randomized, Controlled5moThank you for the info! Do you have any thoughts on how to evaluate a trainer's ability to discern this sort of thing? I'm happy to work with trainers and pay for expertise, but my general sense is that standards of research in sports medicine aren't great.
What are good resources for gears models of joint health?

If your problem is personal, i.e you're dealing with joint issues, unless you're suffering from a muscle-wasting disease or are over the age of 50, reading about stuff will be low yield.

Long term joint pain is solved by:

  • strengthening muscles in order to not put a strain on "weak" joints [evidence: solid]
  • Hormetic effects joint usage [evidence: weak clinical, but look at e.g. people doing yoga, I'd say this is an issue of people not studying the correct demographics]
  • Zone 2 training, aka cardio, allowing you to more efficiently partition fuel to muscles and t
... (read more)
1Randomized, Controlled6moI did a little quick searching for "knee stability training protocol" and.. found a few things that looked pretty obvious. Quads, hams, calves, etc. More or less what I'd expect. I don't suppose you have any secret sauce beyond that? Ie, "train to failure"? If so, I was under the impression that training to failure is now considered less effective/useful. I'm not an athlete, but what would the proxies be?
Shall we count the living or the dead?

It is ultimately about interpretation.

This paradigm doesn't matter if the physician has in mind a cost/benefit matrix for the treatment, in which it would be fairly easy to plug in raw experimental data no matter how the researchers chose to analyze it.

More broadly, see the comment by ChristianKl.

2ChristianKl6moHaving cost/benefit in mind is not enough. If you don't use a heuristic like the one Anders writes about, you need either causal models or something like prediction-based medicine which gives you a way to decide which of two algorithms for decision making is better by looking at the Briers score (or a similar statistic).
2Anders_H6moI very emphatically disagree with this. You are right that once you have a prediction for risk if untreated, and a prediction risk if treated, you just need a cost/benefit analysis. However, you won't get to that stage without a paradigm for extrapolation, whether implicit or explicit. I prefer making that paradigm explicit. If you want to plug in raw experimental data, you are going to need data from people who are exactly like the patient in every way. Then, you will be relying on a paradigm for extrapolation which claims that the conditional counterfactual risks (rather than the magnitude of the effect) can be extrapolated from the study to the patient. It is a different paradigm, and one that can only be justified if the conditioning set includes every cause of the outcome. In my view, this is completely unrealistic. I prefer a paradigm for extrapolation that aims to extrapolate the scale-specific magnitude of the effect. If this is the goal, our conditioning set only needs to include those covariates that predict the magnitude of the effect of treatment, which is a small subset of all covariates that cause the outcome. On this specific point, my view is consistent with almost all thinking in medical statistics, with the exception of some very recent work in causal modeling (who prefer the approach based on counterfactual risks). My disagreement with this work in causal modeling is at the core of my last discussion about this on Less Wrong. See for example "Effect Heterogeneity and External Validity in Medicine" [] and the European Journal of Epidemiology paper that it links to
Shall we count the living or the dead?

This to me seems like a non-issue.

The core problem here is "doctors don't know how to interpret basic probabilities", the solution to this is deregulation in order to hoist the work of decision trees from men.

Discussions like this one are akin to figuring out how to get paedophiles to wear condom more often, in principle they could be justified if the benefits/cost were proportionally immense, but they are a step in a tangent direction and moving away focus from the core issue (which is, again, why are your symptoms, traits and preferences not weighted by ... (read more)

No. This is not about interpretation of probabilities. It is about choosing what aspect of reality to rely on for extrapolation. You will get different extrapolations depending on whether you rely on a risk ratio, a risk difference or an odds ratio. This will lead to real differences in predictions for what happens under intervention.

Even if clinical decisions are entirely left to an algorithm, the algorithm will need to select a mathematical object to rely on for extrapolation. The person who writes the algorithm needs to tell the algorithm what to use, a... (read more)

Against intelligence

On a species level though, the specific niche of human intelligence arose and filled an evolutionary niche, but that is not proof the same strategy will be better.

Bears fill an evolutionary niche of being able to last long times without food, having a wide diet and being very powerful, but that's not a conclusion that a bear that's 3x bigger, can eat even more things and can survive even longer without food would fare any better.

Indeed, quite the opposite, if a "better" version of a trait doesn't exist that likely means the trait is optimized to an extreme... (read more)

Against intelligence

Roughly speaking, yes, I'd grant some % error, and I assume most would be cofounders, or one of the first researchers or engineers.

Back then people literally made 1-niche image recognition startups that work.

I mean, even now there are so many niches for ML where a team of rather mediocre thinkers (compared to, say, the guys at deep mind) can get millions in seed funding with basically 0 revenue and very agressive burn, by just proving very abstractly they can solve one problem or another nobody else is solving.

I'm not sure what the deluge of investment and... (read more)

Against intelligence

I think "very" is much too strong, and insofar as this is true in the human world, that wouldn't necessarily make it true for an out-of-distribution superintelligence, and I think it very much wouldn't be. For example, all you need is superintelligence and an internet connection to find a bunch of zero-day exploits, hack into whatever you like, use it for your own purposes (and/or make tons of money), etc. All you need is superintelligence and an internet connection to carry on millions of personalized charismatic phone conversations simultaneously with pe

... (read more)
4Mitchell_Porter6moSuch an interesting statement. Do you mean this literally? You believe that everyone on Earth who "understood AI" ten years ago, became a highly successful founder?
2ChristianKl6moA person who runs a company worth a few hundred millions is mainly spending his time managing people. There are plenty of cases where it makes more sense to listen to scientists who spend their time studying the subject then to managers when it comes to predicting future technology.
Against intelligence

I think the usual rejoinder on the "AI go foom" side is that we are likely to overestimate x by underestimating what really effective thinking can do

Well, yeah, and on the whole, it's the kind of assumption that one can't scientifically prove or disprove. It's something that can't be observed yet and that we'll see play out (hopefully) this century.

I guess the main issue I see with this stance is not that it's unfounded, but that its likely cause is something like <childhood indoctrination to hold having good grade, analytical thinking, etc as the highe... (read more)

Against intelligence
  • I'm pretty surprised by the position that "intelligence is [not] incredibly useful for, well, anything". This seems much more extreme than the position that "intelligence won't solve literally everything", and like it requires an alternative explanation of the success of homo sapiens.


I guess it depends on how many "intelligence-driven issues" are yet to solve and how important they are, my intuition is that the answer is "not many" but I have very low trust in that intuition. It might also be just the fact that "useful" is fuzzy and my "not super useful" might be your "very useful", and quantifying useful gets into the thorny issue of quantifying intuitions about progress.

Against intelligence

The question you should be asking is not if IQ is correlated with success, but if it's correlated with success in spite of other traits. I.e. being taller than your siblings, facial symmetry and having few coloured spots on your skin are also correlated with success... but they are not direct markers, they simply point to some underlying "causes" ("good" embryonal env, which correlates with being born into wealth/safety/etc | lack of cellular damage and/or ability to repair said damage | proper nutrition growing up... etc).

Also, my claim is not that humans don't fetishize or value intelligence, my claim is that this fetish specifically pretains to "intelligence of people that are similar enough to me".

5ChristianKl6moI think Gates/Bernake being a standard deviation higher in IQ then Kasparov does suggest that IQ is more important for what they did then it is for chess. The ability to spend a significant amount of time training for Chess also comes with a privileged background. Your model suggests that Bill Gates and Bernake should be taller then average. Ben Bernanke is 5-foot 8-inch while Gates is 5′ 10″. As far as Yanet goes she's called "small lady with a large IQ" for a reason. Rumors have it that she wasn't reappointed by Trump because Trump thought that she was too small. We don't have testscores for Bezos was smarter than his childhood enviroment given that he was valedictorian and with 5′ 7″ he's smaller then his brother. How do you think your model that predict those people to be talled should be updated based on the prediction of them being tall not representating reality? How do you explain that being born into wealth correlates with IQ? I think the best explanation is that high IQ parents are more likely to be economically successful. All the points to being taller better facial symmetry don't explain Jewish wealth which is best explained by intelligence being economically useful. If you don't believe that explanation what's your alternative? I think the thing you call fetishing intelligence is fetishing things like being good at chess or go where IQ isn't very central while at the same time ignoring that it's central for other domains. Thinking of Go/Chess as high intelligence activities is part of a certain stereotype and that stereotype is not helpful but it's distinct from IQ being useful.
Against intelligence

I guess I should have worded it as "while most people", I certainly agree some people can "think the pain away" and hypnosurgery is a thing and has been for over 100 years, so yeah.

Re: Fierce Nerds

I think the thing missing here is "fierce about what".

Being fierce about spacecrafts, osk therapy or ecological materials is basically good.

Being fierce about Unix, ml, rust or fpgas is morally neutral, but can be good or bad depending on the trends in society and your industry.

Being fierce about my little pony, debating people online, arguing for extremist political views, reading up on past wars, being a ""PUA"" and playing starcraft is bad, bad of society, but more so for the individual which is slowly consumed by it.

Elon musk is annoying because he thi... (read more)

Death by Red Tape

"Progress" can be a terminal goal, and many people might be much happier if they treated it as such. I love the fact that there are fields I can work in that are both practical and unregulated, but if I had to choose between e.g. medical researcher and video-game pro, I'm close to certain I'd be happier as the latter. I know many people which basically ruined their lives by choosing the wrong answer and going into dead-end fields that superficially seem open to progress (or to non-political work).

Furthermore, fields bleed into each other. Machine learning ... (read more)

5ChristianKl7moNo, the FDA would very likely have shut them down. You can't simply evade the FDA by saying that you are selling a supplement.
George's Shortform

I've been thinking a lot about replacing statistics with machine learning and how one could go about that. I previously tried arguing that the "roots" of a lot of classical statistical approaches are flawed, i.e. they make too many assumptions about the world and thus lead to faulty conclusions and overly complex models with no real insight.

I kind of abandoned that avenue once I realized people back in the late 60s and early 70s were making that point and proposing what are now considered machine learning techniques as a replacement.

So instead I've decided... (read more)

Predictive Coding has been Unified with Backpropagation

There is no relationship between neurons and the "neurons" of an ANN. It's just a naming mishap at this point.

2samshap8moIncorrect. Perceptrons are a low fidelity (but still incredibly useful!) rate-encoded model of individual neurons.
1peak.singularity8moWell, when reading this : [] Less Wrong & SSC => ACX clearly seem to me to be much closer to the empiricist side than the rationalist one ?
MIRI comments on Cotra's "Case for Aligning Narrowly Superhuman Models"

Sometimes, those tokens represent words and sometimes they represent single characters.

Hmh, ok, quick update to my knowledge that I should have done before:

Seems to indicate that GPT-2 uses a byte-level BPE, though maybe the impl here is wrong, where I'd have expected it to use something closer to a word-by-wrod tokenizer with exceptions for rare words (i.e. a sub-word tokenizer that's basically acting as a word tokenizer 90% of the time). And maybe GPT-3 uses the same?

Also it seems that sub-word t... (read more)

5Quintin Pope9moGPT-3 (and most pretrained transformers) generate tokens, not words or characters. Sometimes, those tokens represent words and sometimes they represent single characters. More common words receive their own token, and less common words are broken into two or more tokens. The vocab is tuned to minimize avg. text length.
Load More