Posts
Comments
I saw that a lot of people are confused by "what does Yudkowsky mean by this difference between deep causes and surface analogies?". I didn't have this problem, with no delay I had interpretation what he means.
I thought that it's difference between deep and surface regarding to black box metaphor. Difference between searching correlation between similar inputs and outputs and building a structure of hidden nodes and checking the predictions with rewarding correct ones and dividing that all by complexity of internal structure.
Difference between making step from inputs to outputs and having a model. Looking only at visible things and thinking about invisible ones. Looking only at experiment results and building theories from that.
Just like difference between deep neural networks and neural networks with no hidden layers, the first ones are much more powerful.
I am really unsure that it is right, because if it was so, why he just didn't say that? But I write it here just in case.
I noticed that some names here have really bad connotations (although I am not saying that I know which don't, or even that any hasn't).
"LessWrong" looks like "be wrong more rare" and one of obvious ways to it is to avoid difficult things, "be less wrong" is not a way to reach any difficult goal. (Even if different people have different goals)
"Rationality: from A to Z" even worse, it looks like "complete professional guide about rationality" instead of "incomplete basic notes about a small piece of rationality weakly understood by one autodidact" which it actually is.
There are no common words upvote/downvote in Russian, so I just said like/dislike. And it was really a mistake, these are two really different types of positive/negative marks, agree/disagree is third type and there may be any amount of other types. But I named it like/dislike, so I so thought about it like it means your power of liking it in form of outcome to author, not just adjusting the sorting like "do I want to see more posts like that higher in suggestions".
And actually it looks for me like a more general tendency in my behaviour to avoid finding subtle differences between thing and, especially, terms. Probably, I've seen like people are trying to find difference in colloquial terms which are not strictly determined and next argue to that difference, I was annoyed by that and that annoyance forced me to avoid finding subtle differences in terms. Or maybe it is because they said us that synonyms are words with the same meaning, instead of near meanings (or "equal or near meanings"), and didn't show us that there is difference in connotations. Or maybe the first was because of the second. Or maybe it was because I too much used programming languages instead of normal languages when I was only 8. Anyway, I probably need now to start developing a 24/7 automatically working habit to search and notice subtle differences.
Does the LessWrong site use a password strength check like the one Yudkowsky talks about (I don't remember that one)? And if not, why not? It doesn't seem particularly difficult to hook this up to a dictionary or something. Or is it not considered worth implementing because there's Google registration?
Hmm. Judging from the brief view, it feels like I'm the only one who included reactions in my brief forms. I wonder why?
It occurred to me that on LessWrong there doesn't seem to be a division of posts in evaluations into those that you want to promote as relevant right now, and those that you think will be useful over the years. If there was such an evaluation... Or such a response, then you could take a list not of karma posts, which would include those that were only needed sometime in a particular moment, but a list of those that people find useful beyond time.
That is, a short-term post might be well-written, really required for discussion at the time, rather than just reporting news, so there would be no reason to lower its karma, but it would be immediately obvious that it was not something that should be kept forever. In some ways, introducing such a system would make things easier with Best Of. And I also remember when choosing which of the sequences to include in the book, there were a number of grades on scales other than karma. This could also be added as reactions, so that such scores could be left in an independent mode.
A. I saw a post that reactions were added. I was just thinking that this would be very helpful and might solve my problem. Included them for my short forms. I hope people don't just vote no more without asking why through reactions.
On the one hand, I really like that on LessWrong, unlike other platforms, everything unproductive is downgraded in the rating. But on the other hand, when you try to publish something yourself, it looks like a hell of a black box, which gives out positive and negative reinforcements for no reason at all.
This completely chaotic reward system seems to be bad for my tendency to post anything at all on LessWrong, just in the last few weeks that I've been using EverNote, it has counted 400 posts, and by a quick count, I have about 1500 posts lying in Google Keep , at the same time, on LessWrong I have published only about 70 over the past year, that is, this is 6-20 times less, although according to EverNote estimates ~ 97% of these notes belong to the "thoughts" category, and not to something like lists shopping.
I tried literally following the one advice given to me here and treating any scores less than ±5 as noise, but that didn't negate the effect. I don't even know, maybe if the ratings of the best posts here don't match up with my rating of my best posts, I should post a couple of really terrible posts to make sure they get rated extremely bad and not good or not?
I must say, I wonder why I did not see here speed reading and visual thinking as one of the most important tips for practical rationality, that is, a visual image is 2 + 1 d, and an auditory image is 0 + 1 d, plus auditory images use sequential thinking, in which people are very bad, and visual thinking is parallel. And according to Wikipedia, the transition from voice to visual reading should speed you up 5 (!) times, and in the same way, visual thinking should be 5 times faster compared to voice, and if you can read and think 5 times in a lifetime more thoughts, it's just an incredible difference in productivity.
Well, the same applies to the use of visual imagination instead of voice, here you can also use pictures. (I don’t know, maybe it was all in Korzybski’s books and my problem is that I didn’t read them, although I definitely should have done this?)
Yudkowsky says that public morality should be derived from personal morality, and that personal morality is paramount. But I don't think this is the right way to put it, in my view morality is the social relationships that game theory talks about, how not to play games with a negative sum, how to achieve the maximum sum for all participants.
And morality is independent of values, or rather, each value system has its own morality, or even more accurately, morality can work even if you have different value systems. Morality is primarily about questions of justice, sometimes all sorts of superfluous things like god worship are dragged under this kind of human sentiment, so morality and justice may not be exactly equivalent.
And game theory and answers questions about how to achieve justice. Also, justice may concern you as directly one of your values, and then you won't betray even in a one-time prisoner's dilemma without penalty. Or it may not bother you and then you will pass on always when you do not expect to be punished for it.
In other words, morality is universal between value systems, but it cannot be independent of them. It makes no sense to forbid someone to be hurt if he has absolutely nothing against being hurt.
In other words, I mean that adherence to morality just feels different from inside than conformity to your values, the former feels like an obligation and the latter feels like a desire, in one case you say "should" and in the other "wants."
I've read "Sorting Pebbles into Different Piles" several times and never understood what it was about until it was explained to me. Certainly the sorters aren't arguing about morality, but that's because they're not arguing about game theory, they're arguing about fun theory... Or more accurately not really, they are pure consequentialists after all, they don't care about fun or their lives, only piles into external reality, so it's theory of value, but not theory of fun, but theory of prime.
But in any case, I think people might well argue with them about morality. If people can sell primes to sorters and they can sell hedons to people, would it be moral to betray in a prisoner's dilemma and get 2 primes by giving -3 hedons. And most likely they will come to the conclusion that no, that would be wrong, even if it is just ("prime").
That you shouldn't kill people, even if you can get yourself the primeons you so desire, and they shouldn't destroy the right piles, even if they get pleasure from looking at the blowing pebbles.
I added link to comment: https://www.lesswrong.com/posts/34Tu4SCK5r5Asdrn3/unteachable-excellence
In HPMOR, Draco Malfoy thinks that either Harry Potter was lucky enough to be able to come up with a bunch of great ideas in a short period of time, or he, for some unimaginable reason, has already spent a bunch of time thinking about how to do it. The real answer to this false dilemma is that Harry just read a book as a kid where its author came up with all these for the book's needs.
In How to Seem (and Be) Deep Thought, Eliezer Yudkowsky says that the Japanese often portray Christians as bearers of incredible wisdom, while the opposite is true of the "eastern sage" archetype in the western midst. And the real answer is that both cultures have vastly different, yet meaningful sets of multiple ideas, so when one person meets another, and he immediately throws at him 3 meaningful and highly plausible thoughts that the first person has never even heard of, and then does so again and again, the first person concludes that he is a genius.
I've also seen a number of books and fanfics whose authors seemed like incredible writing talents and whose characters seemed like geniuses, fountaining so many brilliant ideas. And then each time it turned out that they really just came from a cultural background that was unfamiliar to me. And I generalized this to the point that when you meet someone who spouts a bunch of brilliant ideas in a row, you should conclude that it's almost certainly not that he's a genius, but that he's familiar with a meme you're unfamiliar with.
And hmmm. Just now I thought about it, but it probably also explains that Aura of Power around characters who are familiar with a certain medium, and people who are familiar with a certain profession (https://www.lesswrong.com/posts/zbSsSwEfdEuaqCRmz/eniscien-s-shortform?commentId=dMxfcMMteKqM33zSa), that's probably the point, and it means that the feeling is not false, it really elevates above mere mortals, because you have a whole bunch of meaningful thoughts that mere mortals simply do not have.
When the familiar with more memeplexes will ponder, he will stand on a bigger pile of cached thoughts, not just on the shoulders of giants, not on the shoulders of a human pyramid of human giants, so he can see much further than someone who looks only from his, no matter how big or small, height.
I've read, including on lesswrong (https://www.lesswrong.com/posts/34Tu4SCK5r5Asdrn3/unteachable-excellence), that often listening to those who failed is more useful than those who succeeded, but I somehow missed if there was an explanation somewhere as to why? And the fact is that there are 1000 ways to be wrong and only 1 way to do something right, so if you listen to a story about success, it should be 1000 times longer than a story about failure, because for the latter it is enough to make one fatal mistake, while for the former you have to not make the whole thousand.
However, in practice, stories of failure and stories of success are likely to be about the same length, since people will take note of about the same number of factors. In the end, you will still have to read 1,000 stories each, whether success or failure, except that success happens 1,000 times less often and the stories about it will be just as short.
I haven't encountered this technique anywhere else, so I started using it based on how associations work in the brain:
If I can't remember a word, instead of just continuing to tell myself "think, think, think," I start going through the letters alphabetically and make an effort over each one "what are the words for that letter, is that word by any chance?" And that almost always helps.
I've noticed that in everyday life, when you're testing some habit choices to see if they're working for you, it's better to leave a habit that doesn't seem to be working for you, to make it easier to determine that, because otherwise you won't be sure later if it turned out to work otherwise, habit one or habit two or habit three.
This reminds me of how I used to do mod compilations, it might seem like a good idea to add all the desired ones at once, but then if some mod is missing or some extra, you won't be able to figure out which one. So they should only be added and removed one at a time.
With habits the same, only even more difficult, because they begin to act gradually and do it much slower, and then there are factors beyond your control. And I used to assume that it doesn't make sense to waste time and effort on following useless habits.
However, since it is also experimentation, it was worth bearing in mind that any change would in any case complicate the analysis of what worked, it is better to keep them until you find a stable working combination, and then remove them one at a time, too, in case certain habits somehow worked only together.
Yudkowsky says in one of his posts that since 0 and 1 for probabilities mean -∞ and +∞, you can't just add up all the hypotheses to get one. However, I don't see why this should necessarily follow. After all, to select one hypothesis from the hypothesis space, we must get the number of bits of evidence corresponding to the program complexity of that hypothesis.
And accordingly we don't need to have an infinite number of proofs to choose, as many as the number of bits in the longest hypothesis is sufficient, since any longer hypotheses will compete with shorter hypotheses not for correctness but for accuracy.
Yes, in the end you can never reach a probability of 1 because you have meta level uncertainty, but that is exactly what meta level probability is, and it should have been written as a separate multiplier, because otherwise adding an infinite number of uncertain meta levels will give you a probability of 0 for each of your hypotheses.
And the probability P(H) without considering meta levels should never be 0 or 1, but the probability P(H|O) could well be, since the entire meta level is put into P(O) and therefore P(H|O) will have a known finite program complexity. That is, something like:
A="first bit is zero", B="first bit is one"
C="A or B is true", O="other a priori"
P(O)=~0.99
P(A|O)=1/2, P(B|O)=1/2
P(C|O)=P(A|O)+P(B|O)=1/2+1/2=(1+1/2)=1
P(C)=P(C|O)*P(O)=1*0.99=0.99
And if we talk about the second bit, there will be two more hypotheses orthogonal to the first two, on the third bit two more hypotheses, and if we talk about the first three bits, there will already be a choice of 8 multiplications of the first six hypotheses, and there will no longer be correct to ask which of 6 hypotheses is true, because there are 6, in options 8, and must be true simultaneously not one hypothesis, but at least 3.
And accordingly, for 8 hypotheses, we can also add up the probabilities as 8 times 1/8 and end up with 1. Or we can write it as 1/2+1/4+1/8+1/8=1, but of course this is only possible if we can count the number of bits in the hypotheses, decompose them into these bits and determine the intersections, so as not to count the same bits repeatedly.
For some reason until recently, nowhere I heard about programming did it explain that object-oriented programming is essentially a reverse projection of the human brain, everywhere I heard about programming before it said something like, at best, that procedural programming is ugh and object-oriented is cool, it did not explain that procedural programming is much closer to the language that reality thinks in, and inventing "objects" is just a crutch for the imperfect monkey brain
All this came to my mind when I noticed that people tend to think of drugs as if a drug is an object that has the property of "curing", like a potion in a video game or a real "potion" from antiquity. And people are still extremely prone to think about viruses, bacteria, antibiotics, vaccines, and the like in this way. Not to imagine a specific mechanism for how it's supposed to work, but just to assume that the drug as a whole will "cure." And of course the same goes for poisons, people not familiar with biology think of them as having an inherent property to "poison," or on a deeper level, acids, as having an inherent property to "dissolve.
And if you go back to the question of reverse projection of the human mind, it becomes obvious that human language is not something that came out of nowhere, it is a product of the human brain, so language must also be a projection of its structure, and therefore you need reverse projection as objects in a broader sense, so specifically convolutional neural networks here, to work with it properly.
Or, if I haven't written it down anywhere else, it occurred to me that since we live inside the Tegmark mathematical universe, in which case the universe is just a giant formula and the process of solving it step by step, each next part after the equal sign is the next moment in time, and the value of the expression itself, what is stored between the equal signs, is energy. The superposition is the subtraction inside the parentheses, which with each step adds another multiplier to both parts of the difference, and the different Everett branches are the same two halves of each difference, just with the parentheses open.
Well, now it's not that I think this is wrong, but rather the opposite, too obvious, and therefore useless.
Besides, it can also be presented in another form, in terms of the intersection of computer science and quantum mechanics, the universe, or rather, all universes in mathematics, is a bit string, which diverges into Everett branches, in the first sign you have 2 branches for 0 and 1, in the second you already have 4, in the third 8 and so on, each branch is a standard particular bit string, with each branch division the amount of information in this string grows, that is, entropy grows, and this direction of entropy growth in each individual branch is time.
The law of conservation of energy, or even more broadly the law of conservation, common to all mathematics, is that at each step you have for 0 you have 1 and vice versa, each time you divide this into two bit options and each time you have the opposite option, so the total entropy of all mathematics is also in some way zero, if you look from inside it is infinite, but to write this down, you do not need any length formula, zero length is enough for you.
So from the inside mathematics is infinite, but from the outside it adds up to zero. That's sort of the answer to the question "why is there something and not nothing?" and that answer is that "something" refers to a piece of "everything" and "everything" is what nothing looks like from the inside.
I came up with this myself, but later I also saw someone else's formulation of this, that for every number on the mathematical plane there is an inverse number, so even though the math has infinite information, it adds up to zero in the end, hence the law of conservation of energy.
As far as I know, it is widely known among physicists, as opposed to ordinary people, that energy is a conditional quantity, and the energy of a part of a system can be the energy of the whole system, since energy can also be negative, can be as negative as you want, so what we think of as zero is only a convenient point of reference.
I was once very interested in the question of what "time" is and what "entropy" is. The thing is, I watched popular science videos on YouTube, and nowhere was there a normal answer, at best it was some kind of circular argumentation. Also, nowhere was there a normal explanation of what entropy was, only vaguely stating that it was a "measure of disorder in the system".
In my head, however, the idea swirled around that it had something to do with the fact that there are more directions outward than inward in space. And I also twirled that it must be connected with the law of least action, for which I also did not meet such an explanation, that is, that the reason is that the straight path is one path, and there are at least 4 detours, and this is only the closest, with each step there will be 2 times less, respectively, if we imagine that there is no "law" of least action, we will still see it, because for a particle the probability to be in each next step from the central path will be 2 times less, because there are twice as many paths, and for a wave it will not even be probability, but purely its distribution.
All these thoughts were inspired to me by a video of balls falling down a pyramid of pegs, and they end up having paths on both sides inward at each step, and only one side path outward, and they form a normal distribution. That is, to put it another way, the point is that although the number of points is the same, the paths in the center converge to each other and the paths on the edges do not, the two paths in the center form a single cluster of central paths and the two paths on the edges do not.
And from this we can assume that the average expected space will have a shape close to a square, a cube, a tesseract, or another figure with equal sides, because although there is only one such figure, and many other variants, these variants do not fit together, but variants close to a cube fit into a cube.
This also explains for me why Everett's chaotic branches do not create a chaotic world. There are more chaotic branches, but they form a circle around the edge rather than a circle in the center, the least chaotic branches are fewer, but they converge to a world close to order, but the most chaotic branches differ from each other even more than they differ from order.
Somewhere here on lasswrong I saw a question about why, if we live in a Tegmark universe, we don't see constant violations of the laws of physics, like "clothes turn into crocodiles." However... We do observe. But only "clothes" and "crocodile" are too meaningfully human variants, in fact there are much more, one mole of matter contains ~10^23 particles, and even if we only assume different variants of their presence/absence, it is 2(1023), our system is too big to notice these artifacts, however if we go to individual particles...
That's exactly what we'll see, constant random fluctuations. Quantum. This can be considered a successful prediction of Tegmark, although in fact only retrospective.
Now I have a better understanding and now I can formulate that it does not take into account the possibility that the report of four victories may be false, that the lotteries could be dishonest, that the woman could simply hack the mechanism of pseudo-randomness, and also does not take into account a priori information in the form of that there are 8 billion people on Earth, and there have been many lotteries throughout history, so the probability that at least someone wins 4 lotteries is very different from the probability of winning a particular woman, in addition, even though she won in a row, and not for a lifetime, you should also take into account the total number of attempts to play the lottery, because the probability of 4 wins in a row is very different if you played 4 times or 400 times, finally, in general, the a priori probability of violating the laws of physics or mathematics, in general, any supernatural strength is extremely low, since these models have a huge amount of evidence for and new evidence will need to overcome all their mass, so the choice will very slowly reach the hypothesis of violation in the laws of the lower level, even if we cannot find an explanation for this event, and finally even if we decide that there is something supernatural here, then there are a lot of options here, it is far from necessarily God, and even more so not necessarily Christian, each of these two hypotheses will have to be proved separately from the presence of supernatural.
Human language works primarily due to recognition in context, this works with individual words, but it can also work with whole phrases, the same word can be completely determined by its morphemes and only morphemes will have to be known from the context, but also a word can be and a single morpheme, and of course here you should also take into account words borrowed from other languages, which in the original language can be broken into morphemes, and in yours be known only from the context, and the same thing works not with whole words, but also with phrases where individual words are separated by spaces in writing and pauses in conversation, there you can also determine the meaning of a phrase based on the meanings of its words, but often this may not be the case, often a certain combination of words has a meaning different from its individual parts, this meaning can be recognized only from the context in which the phrase is used. And also a single word can have a meaning that is different from its morphemes, for example, the words computer, calculator and counter based on morphemes should be synonyms, but they are also used in the context to refer to three specific different devices. And there can be an unlimited number of such levels of different contextual meanings, they are also often used to add additional levels of meaning, if you look at everything only as a sequence of morpheme meanings, the meaning is one, if you perceive the meanings of words from contextual learning as a whole, then the meaning will be different if look at phrases third and so on. I started programming at the age of 9 and even then I noticed that programming is not rocket science, there is nothing complicated, writing a program is no more difficult than speaking, writing in a programming language is just like writing in a foreign language. Later, I also heard about a study where an MRI showed that programming uses the same areas of the brain as talking. And this finally strengthened me in the wrong thought. The fact is that conventional languages and programming languages, despite the fact that these are all languages, are very different, programming languages are "ascending" languages, they start with a description of completely elementary structures, then refer to them, then refer to sets of links and so on, in any case, this is a perfectly accurate process, in programming if you have described some elementary structure, then this is a perfectly accurate description, without any vagueness, and if you have described a high-level representation from references, then again it is perfectly exact, you just referred to a number of perfectly accurate low-level descriptions.
You create your worlds here with your own laws of physics, and like nature does not tolerate any exceptions (although there is also a huge difference between the laws of physics and the laws of the game), if you say "truth", then it is absolute truth, if you say "everything" , then this is absolutely everything that is, if you say "always", then it will be absolutely always, any situations where this is not so, these will be situations of a specific indication "always except for a and b", in human language it is completely not like that, it works like leaky categories (reference needed), if you say "I will never tell a lie", then although you do not describe any exceptions, you still mean a bunch of them, like "except if I myself do not know the truth ", "unless I get drugged", "unless someone changes my brain" and probably also "unless lying saves my or someone else's life".
Wow, that's just a masterpiece metaphor! Saved this. I must also mention here, in addition to the Rapier and the Gun, one very famous razor, Occam's Razor (and even a seemingly more advanced version of it, Solomon's Lightsaber), from which it obviously follows that many philosophers also still cannot stand shaving (they themselves say that it's just a straight razor, it's too easy to cut yourself) and walk around with huge beards.
Previously, on first reading, this seemed quite plausible to me, but now, after some time of reflection, I have come to the conclusion that it is not, and Yudkowsky is right. If we continue in the same terms of shoggots and other things ... Then I would say that there is a shoggot there, it’s just a shoggot of the Chinese room, it’s not that he sits there, understands everything and just doesn’t want to talk to you, no, you and your words are as incomprehensible to him as he is to you, and he is fundamentally very stupid, only able to understand the translation instructions and execute them without error, however, thanks to the huge array of translation data on which he applies these instructions, he is able to perform all shown tasks. And in fact, with symmetrical incomprehensibility for each other, this is precisely the difference, a shoggoth has this huge dataset about people with which it works at an acceleration of 20M times, and people do not have the same dataset on shoggots that they worked on at the same acceleration, in fact, they don't even have 20M parallel people working on it, because people don't have a dataset on shoggots at all. That is, it cannot be said about a shoggoth that it has a high intelligence, on the contrary, it is very low, but thanks to the enormous speed of work and the huge amount of cached information, it still has a very high, close to human, short-term optimized strength. In other words, a shoggoth has a lot of power, but it's very inefficient, especially if we talk about the standard list of human, except for speaking, because it does everything through this extra prism. So it's kind of like making a Go program speak a language by associating the language with board states and games won. Because of this superfluous superstructure, at very high power it will be able to mimic a wide range of human tasks, but it won't do it to its full potential because it's based on an inappropriate structure. So the efficiency in human tasks will be approximately the same as that of a person in mathematics tasks. That is, Yudkowsky is right, a shoggot is not an average of a bunch of parallel mental streams of people who switch between each other, these are cones of branches of probability trees diverging in both directions, and its different faces are not even faces' masks, these are the faces of Jesus on a toast or a dog, human tendency to apothenia with finding human faces and facial expressions in completely alien processes.
One of my most significant problems is that I do not like to read, although it is generally believed that all "smart" people must adore it. And accordingly to overcome my dislike of reading, the book itself has to be very interesting for me, and such are rare and difficult to find them for me (I was thinking that it would be nice to have some kind of recommendation service on your previous evaluations, which there are for movies, but for books I do not know such).
And accordingly, another problem follows from this. I didn't read a collection of science fiction and fantasy books as a kid, I didn't read Escher Bach's Gödel, Step into the Future, Impact: Science and Practice, Science and Sanity, Probability Theory: the Logic of Science, Feynman Lectures in Physics, and so on. And I feel like I'm missing out on much of the obvious stuff for "our cluster people" because of this.
That is, the sequences were written to point out some non-obvious things that weren't already said by someone else and known to everyone. And I don't know if there's a list somewhere of such most significant books that predate the sequences, so that a person from outside this cluster would at least know which direction to look, or else it's hard to identify such things that are obvious background facts to some people, but unknown to others altogether.
In a broader sense, you could say that I was making the mistake of looking too little at other people's experiences. I didn't realize that no one person is capable of inventing all the wisdom of a tribe on their own. And this manifested itself everywhere, in every field, in programming, in music, in writing books, in creating languages, in everything.
Probably one is related to the other, I didn't read a bunch of books, so I didn't see how much higher than my own intelligence other people's knowledge could be. So I intuitively, without realizing it, proceeded as if they were equal, as if I were competing with a single other person, without even considering the obvious fact that there were simply many more other people, let alone what could be achieved by investing more time, summing up the experience of many generations, using collaborative work, practical experiments and computer calculations.
Actually, I didn't quite put it that way, though. Yes, I don't adore reading, but let's say I don't have any problem with reading blog posts (e.g. here on Lezvrong) or even just Wikipedia pages, here I rather have the opposite problem, opening one article and following the interested hypertext links I can sit down and read for hours.
So it's probably more accurate to say that I have a "clip thinking" problem, however... I have no problem watching hours of lectures, reading 4 hour podcast transcripts, or listening to hours of audiobooks.
So it's probably "reading books" that's the problem. Perhaps I associate them with school, with the most boring reading of history textbooks when my brain literally stops perceiving, or with literature classes, including assigned readings for the summer, where of all the years of school I can remember exactly one work that seemed interesting to me. I'm not sure what the big deal is.
I used to have conflicting thoughts about death. Does death nullify the value of life, because you yourself are destroyed? Or maybe, on the contrary, immortality, because you will ever achieve everything? The first is false, because a person has other values besides the state of his own brain. The second has nothing to do with life, because really by immortality we mean "not dying of old age" and not "living infinity of years", so you only have a trillion years ahead of you, and this is a very limited time. And any other finite time too. Thus, one should not be an absolutist, the value of life is equal to the lived period by the coefficient of average goodness, and the anti-value of death is equal to the product of the distribution of expected value and the distribution of expected probabilities. I also wondered, if there are no stupid values, then can't any person choose the value of death himself? And... No. The word "terminal" is missing. Instrumental values are a different type of information altogether and can easily be silly. And a person is clearly not born with a final desire to die, as, for example, one is born with a desire for sweetness and the absence of pain. So no, death destroys too much expected positive utility to rationally prefer it to anything. Another thing is that the decision-making mechanisms in the human brain are far from rational, and first of all, we are talking about time discounting at a fixed, not a probabilistic rate, which makes you extremely overestimate what is happening now compared to the future.
I recently read Robin Hanson saying that small groups that hold counter-intuitive beliefs tend to come from very specific arguments, some even invent their own terminology, and outsiders who reject those beliefs often they don't even bother to learn the terminology and review the specific arguments because they reject these beliefs on purely general grounds. And this is what I myself noticed, however, only from within such a group, I was outraged and annoyed by the fact that usually the arguments of those who reject FOOM or the danger of AI / AGI in general are so general that they have nothing to do with the topic at all. On the other hand, Robin Hanson gives an example like conspiracy theories, and this is true, and this is something that I did not think about, because really, when I come across a conspiracy theory, I don’t even try to spend time studying it, I just immediately reject it on a general basis and move on. This may well be called the application of an outside view and a reference class, however, the point is not that I use this as an argument in an argument (arguing with a conspiracy theorist is a priori useless, because in order to believe in it, you need not understand scientific method), I use this to avoid even wasting time on a more precise but longer look inside. If I were to take the time to consider a conspiracy theory, then I would not dismiss all arguments on the grounds that your idea belongs to the reference class of conspiracy theories and your view from the inside is just your vision under cognitive distortions. I would apply an inside view, understand the specific arguments, reject them on the basis of scientific evidence. And I would use the view from the outside in the form of a reference class of conspiracy theories only as not very weighty a priori information. So the problem is not that I have an impenetrable barrier of reference classes, but that this narrow group goes beyond the reference class of ordinary conspiracy theorists. And, perhaps, without fail, entered the reference class of those who are well acquainted with the scientific method, because without this I will decide that they are just one of a slightly smaller number of non-standard conspiracy theory groups.
And, I'm not entirely sure it's worth doing it in one post, but it also talked about the cases of seeing a sharp change in values about reaching full strength. Maybe not in the economy, but in politics this happens regularly, at first the politician is all so kind, he does only good, and then we finally give him all the power, we make him an absolute dictator, and for some reason he suddenly stops behaving so positively, arranges repressions, starts wars.
And more about Hanson's sketch about the explosion of "betterness". And at the end, the question is "Really?". And personally, I answer this: well ... Yes. Right. But only the human brain works about 20M times slower than a computer processor, so it will take about the same amount more time. If for a computer I would be ready to believe in a period from 1 month to 1 millisecond, then for a person it would be from 1.5M years to 6 hours, but the second is subject to the ability to instantaneous self-modification, including that any new idea instantly updates generally all your previous conclusions. Plus, in the intermediate scenarios, you still need to take into account sleep, food, willpower, the need to invent a cure for old age, and the elimination of other causes of death. In addition, you will have to hide your theory of superiority from other people so that they do not decide to stop you before seizing power. But in general, yes, I think that this is possible for people, just a whole bunch of problems like a lack of will, the lack of the possibility of self-modification and terrible for this evolutionary spaghetti code in the brain. But first of all, a too short life span, if people lived a million years, or better a billion, so that with a margin, they could achieve all this, despite all the difficulties, except death. However, individuals live too little, think too little, so the explosion of betterness is only possible for them collectively, it has been going on for 300 years, from 3 to 15 generations and involves 8 billion people at its peak. More specifically, I would consider this “betterness explosion” a broader concept than the “intelligence explosion” in the sense that it does not specifically talk about intelligence, but only about the concept of optimization processes in general, but at the same time more narrow, because it seems to have specific values, positive human ones, while a nuclear explosion could also be called an optimization explosion, but by no means aligned with human values.
But does Yudkowsky mention the word "abstraction"? Because if not, then it is not clear why the levels. And if you mention it, then as in the case of scale, I don’t really understand why people would even think that different levels of abstraction exist in the territory.
Edited: I've searched in Reductionism 101 and Physicalism 201 and didn't find mention of "abstraction", so I save my opinion that using just word "level" doesn't create right picture in the head.
Again, I'm not sure if I already wrote, but when it comes to quantum mechanics, Born probabilities and why a square, then it's spinning in my head that if you square and take the root in the form of an equation back, then you will have from one square branching into two possible answers, positive and negative, in other words, with this operation you erase exactly one bit of information, the sign bit. And in turn, if you took a number to the first power, then you could directly compare each point of the past with each point of the future and then there would be no calculation of the wave function of the future only from the entire function of the past, you would have exactly one future for each past, and it would not be a question of probabilities at all, only of a logical conclusion.
I thought about the phenomenon when the monuments of all the great people are taken down, because they were slave owners. And I formulated why I do not like this trend. The fact is that the past will look extremely terrible from the point of view of the future anyway. That is, virtually anyone in the past has done at least one absolutely terrible thing by today's standards. If you continue this trend, it is very likely that in the future you will already be considered a monster, for some strange reasons from the point of view of today, such as the fact that you did not cryopreserve chimpanzees, ate the meat of dead animals, buried your mother's brain in the ground to rot and be eaten by worms, didn't apply to donate your organs for transplant if you died, didn't donate to malaria nets, agreed to drive a car, didn't have biological children, were smarter than average, or got bad grades. And these are just things that are already discussed today, in fact, it may well be something completely unexpected, which you could not even suspect that this is actually something bad. Well, you can give other examples of the horror of the past in addition to slavery. Something like Bayes, Newton and a bunch of other scientists were religious. Although, in general, religion is very common to this day, so the entire previous list can already be applied to scientists of the past, if you agree with the specific points. Based on this, it is obvious that achievements should be valued regardless of today's moral assessment.
I have never heard of this before, either here or elsewhere, but I myself notice that usually even the most unreasonable thing like religion has its grain of truth. Or rather, a lot of grains of truth. That is, knowledge, even in human society, almost never has a purely negative utility, a person who knows a certain cultural cluster will almost always have an advantage over a purely ignorant one. However, an important nuance, while a more knowledgeable one may turn out to be worse when they are both taught the correct rational methods, a pure mind will accept them without problems and go forward, and the one busy with another meme will resist. And this is all understandable, the combined optimization pressure between intelligence and meme evolution is unlikely to leave purely harmful memes alive in the end. Human thinking plus natural selection usually correlates with some useful/sane traits in memes. Although, of course, they can very often lead into traps, holes in the fitness landscape, create inadequate balances, and so on and so forth. This can be seen, for example, in sayings (not to be confused with omens, they are almost always just a consequence of apothenia with a confirmation error), something like "geniuses think alike" this is easily paraphrased into "there are a thousand ways to be wrong and only one way to do that - that's right." Usually, if a saying exists, then it correctly reflects some aspect of reality. The problem is that such observational data collection without building causal models does not make it possible to understand how conflicting sayings relate to each other and which are true in which situations. Because about the same geniuses, and vice versa, there is an observation that they think outside the box. And without analysis, you cannot clearly say that in order to obtain the greatest amount of information, one should causally or logically coordinate for geniuses to think on different topics. Or that non-standard thinking is just “not like usual, but meaningful”, so that it can be caused both by being on higher levels of the conceptual pyramid, or simply by another memeplex, which does not cancel the previous point about coordination for thinking in unusual directions of search . However, also with all sayings there is an effect that only those who already understand them understand them, and all because these are just links, but I will write more about this another time.
I must say, I made the mistake of thinking that it was enough to make a habit to get a result. At that time, I was interested, so I did not notice that interest was required here. But now I realize that only some part of the crystallized intellect, and not dynamic, can be made into a habit, learning anything, including languages, is not the activation of already learned neural connections, but the creation of new ones, this requires straining the intellect and this cannot be to do purely out of habit, some kind of emotional motivation is needed, for example, interest. So now I don’t study any one particular language, but switch between English, German, Greek, Latin and Italian as my interest fades / grows, in order to constantly keep it at a high level for at least one language.
I expressed myself inaccurately. Firstly, of course, simple knowledge will not make water cold for you, you also need to move your finger very accurately and quickly in order to avoid hotter molecules, I just considered this insignificant, since you initially physically cannot fit into your brain weighing 1.5 kg information about 0.25 kg of water molecules. Secondly, to put it with my current best understanding, these systems are similar in that it is generally believed that the glass just has high entropy, so you can't help but get burned, so it is generally believed that the market is just efficient, just unpredictable, although in general - then these are all human-centric, relative categories (I don’t remember, they should be called “magical” or “unnatural” in Lessvrong), one way or another, the point is that you can’t talk about the entropy of the order book or the predictability of the market, as about object.property, just like subject.function(object), otherwise it would be support for making a mind projection error, there is no entropy of an object, there is only mutual entropy of two objects, and it doesn't matter if you are talking about heat dissipation or information prediction. (Then it just occurred to me that there is a difference in vision between an informed and an uninformed subject in both cases, that for one impenetrable chaos, for the other a transparent order)
It seems that you didn’t understand that my main problem is that every time in my thoughts I rest on the fact that within the framework of a better understanding of the world, it becomes clear that the justifications why competitions are good do not make any sense. It is as if you have lived well all your life because you were afraid of hell, and now the previous justifications why it is worth living well do not make sense, because you have learned that hell does not exist, now it is not clear what is the point of behaving well and whether in general, is it worth it to continue to honor his father, not to steal other people's things and other people's slaves, to stone women for betraying her husband and burn witches? Maybe these rules make sense, but I don't know if they have it, and if so, which one. I mean, I wondered what role do, say, sports olympiads play, but my only answer was that they force athletes to overexert their bodies or dope, and also spend a lot of money on such extremely expensive entertainment in exchange for releasing better films or let's say the scientific race "who will be the first to invent a cure for aging." Well, I've been recently thinking about how to organize the state according to the principle of "look at the economic incentives" and I seem to finally understand what sense competition can sometimes have. Those incentives. Competitions are one of the types of competition, so they allow not only to give an incentive to someone to go towards a certain goal, they can create an arms race situation when you need to not only achieve the goal, but do it faster/better than other participants. However, the key word is "goal", and in sports olympiads the goal clearly does not make sense, like beauty contests, in science olympiads with a better goal, they allow you to determine the smartest participants for further use of their intellect, which does not make sense in sports olympiads, because machines have long been faster, stronger and more resilient than humans, and right before our eyes they are becoming more agile, however, at local sports competitions with a better goal, they allow people to be stimulated to move more.
More specifically, what I mean is that I find it extremely pointless to make something a moral value such as duty rather than a preference value such as taste if that attitude varies by region of birth. Which can probably be expressed as something like "I think it's a mistake to list anything other than the direct conclusions of game theory in the list of moral values of duty." Well, or else you can say that I believe that interpersonal relationships should not be regulated by someone's personal preferences, only by ways of finding a strategy for the game to achieve the maximum total. Well, maybe it's just a longer and more pompous way of saying "do not impose your preferences on others" or "the moral good for the other person should be determined by preferential utilitarianism."
This phrase can also be called something like a projection of the concept of inadequate equilibria in civilization on a particular person or decision-making process. In other words, the process of attempting a better self-modification runs into self-reference, trying to test each of its elements with each of its elements, for an external object you can test it without prejudice, but you cannot test yourself, because prejudices can hide themselves.
I seem to have a better understanding of timeless physics since then, and if we talk more clearly about the regularity that I had in mind, then ... point in time of the book or all at once, for there is no answer to the question "what day is it in Middle-earth?", but all because our timeline has nothing to do with that one. And when we look at any mathematical object, its timeline, like the book's timeline, is also not connected to ours, which is why they look outside of time. That is, because the timeline is not the same for the entire universe, there are many timelines, and we are inside our own, but not inside the timeline of some object like a book or something else. And you can also say that if you usually say something like "we see the passage of time when entropy grows" and "entropy is something that grows with time", then outside of time physics reduces / reduces time to entropy. You link into a timeline chain those fragments of mathematical descriptions between which there is the least mutual entropy. This model of time also says that in addition to the standard linear time scale, there should be all non-standard time scales like different types of Everett branches, past, future and parallel, different types of time loops, like rings and spirals, and so on. And this can be called a calculation, because the calculation leads to a change in shape, another piece of information, and between it and the original one there will always be some kind of mutual entropy. It seems that everything, in short, did not work out, because although I myself understand what I meant then, I see that I expressed myself extremely unclearly. The answer question is, what does "working on the integration of concepts" mean? I don't understand what is meant by this expression.
An interesting consequence of combining the logical space of hypotheses, Bayes' theorem, and taking priors from Kolmogorov complexity is that any hypothesis of a certain level of complexity will have at least two opposite child hypotheses, which are obtained by adding one more bit of complexity to the parent hypothesis in one of two possible states.
And, conversely, you can remove one bit from the hypothesis, make it ambiguous with respect to which child hypothesis will fit your world, but then you will need fewer bits of evidence to accept it a priori.
And accordingly, there will be one hypothesis that you will have to accept without proof at all, a program of zero length, an empty string, or rather an indefinite string, because it contains all its child hypotheses inside itself, it simply does not say which one of them is true.
And this absolutely a priori hypothesis will be that you exist within at least one world, the world of the space of hypotheses itself, it is not specified simply in which particular part of this world you are.
And this makes us look differently at the Tegmark multiverse. Because, let's say, if we know that we are on Earth, and not on Mars, then this does not mean that Mars does not exist. We just aren't there. If we are in a certain branch of Everett, then this does not mean that other branches of Everett do not exist, we just ourselves are not there. And continuing this series, if we are within one program of the physical laws of the universe, then this does not mean that other programs of universes with other laws do not exist, it's just that we ourselves are not there.
If we bring this pattern to a logical result, then this means that the fact that we are inside one sequence of bits does not mean that others do not exist. In other words, based on an absolutely a priori hypothesis, all possible sequences of bits exist, absolutely everything, no restrictions, neither on computability, nor on infinity, or even on consistency. Just exist absolutely all mathematics in general.
In general, all mathematics is summed up in hypotheses of zero length and complexity, because it does not limit our expectations in any way. That is, this is not a request to accept some new hypothesis, it is, on the contrary, a request to remove unreasonable restrictions. In the broadest sense, there are no limits on either computability or infinity, and if you claim otherwise, then you are proposing a hypothesis that narrows our expectations, and then it is on you that the burden of proof lies.
However, this can also be called a rethinking of the concept of "exists", people say it intuitively, but they absolutely cannot give it any definition. And if you refine this concept to something related to controlling your expectations, then you can talk about questions like how likely you are to see contradictions, infinity or something else like that, then you can specifically answer.
For example, one could say that you will never see inconsistent territory because it is a delusion of mind projection, "contradictory" is a characteristic of a map that does not actually have the territory it is supposed to describe.
Seeing contradictions in a territory is like seeing a “you are here” icon or an indication of the scale, or asking what is the error of the territory to itself, the answer is “none”, because the territory is not a map, not a model created from an object, it is the object itself , he cannot be in error with himself, and he cannot contradict himself.
It can all be called something like Tegmark V, mathematical realism and Platonism without limits. And personally, these arguments convince me quite well that all mathematics exists and only mathematics exists, or, to paraphrase, all that exists is mathematics or a specific part of it.
Edited: in some strange way I forgot to clarify, to indicate this obvious point in terms of the space of hypotheses and the narrowing of expectations.
Usually, any narrowing is done by one bit, that is, twice, so you will not reach zero in any finite number of steps, this is because you remove one of the two bits each time in a perpendicular to all previous direction, however, you can also do otherwise, you can not delete any of the two bits, or, after deleting the first bit, delete the second one, delete both, cut not one half, but * (1-1/2 = 1/2) = a * 1/2, and both halves, a*(1-1/2-1/2=0)=a*0, with such a removal that is not perpendicular to the previous ones, but parallel, we remove both possible options, we narrow the space to zero in some direction, and as a result, the whole hypothesis becomes automatically narrowing our entire space of hypotheses to zero, so that now it corresponds not to half the number of territories, but to zero, excluding both alternatives, it excludes all options.
This looks like a much better, much more technical definition of a contradiction than trying to proceed from its etymology, and thus it is clear that there is no “contradiction” exactly, just any card is obliged to narrow your expectations, leaving only one option for each choice, indefinite leaves both, therefore useless, but the contradictory leaves neither, excludes both options, therefore even more useless.
If there is a contradiction not on the map itself, but in the general system of the map and the territory, there is no such problem, it only means that the map is incorrect for this territory, but may be correct for another, it happens that we already have there is an a priori exclusion of one of the options, but the data received from the territory exclude the second option, if we draw up a second map based on these data, then it will not supplement and clarify ours, but will exclude both possible alternatives along with it, therefore the maps are incompatible, their cannot be combined without getting the exclusion of both sides of reality, the narrowing of space to zero.
I thought for a long time about what "contradictions" mean at all and how they can not exist in any world, if here they are, they can be written down on paper. And in the end, I came to the conclusion that this is exactly the case when it is especially important to look at the difference between the map and the territory. Thus an inconsistent map is a map that does not correspond to any territory. In other words, you usually see the area and then you make a map. However, the space of maps, the space of descriptions, is much larger than the space of territories. And you may well not draw up a map for a certain territory, but simply generate a random map and ask what territory it corresponds to? In most cases, the answer is "none", for the reason already described. There are maps compiled by territories and they correspond to these territories, and there are random maps and they are not required to correspond to any territory in principle. In other words, an inconsistent map exists, but it is not drawn for any territory, the territory for which this map was drawn, if it were drawn for a certain territory, does not exist, because this map was not drawn for a territory. This can be represented as the fact that our map is made from a globe. If you unfold it and attach it to the plane, a series of slices will come out, and there is an empty space between them, on an ideal map it should not be white, but transparent. However, if you first draw the map, then you can fill these transparent areas with a certain pattern, and if you try to collapse the map back into a globe, then you will not get a single real globe, because it will not be a sphere, Riemann space, but a plane , the Euler space with a higher dimension, as well as the space of maps to the space of territories, and its attempt to project onto a sphere will not lead to any even result, it will go in folds, like an attempt to project Lobachevsky's space into ours. Even for programmers, this can be represented as a difference between a map and an array, in the case of a map, you can specify many different values for one key and then it will not be possible to collapse into an array. Thus, inconsistent models can be described as artificially created similarities of object projections that cannot be unambiguously deprojected back into an object. And in order to avoid confusion, it should be said not that contradictions are something that cannot exist, but that contradictions are a state of the map, which does not correspond to any state of the territory. In other words, you can also say that this is a typical case of mind projection error, when you project a certain state of the map (which really exists and is not contradictory) onto the territory and cannot get this very territory, then you say that you cannot get it, because that would be controversial territory.
I don't remember if I already wrote about this, but I was thinking about the space of hypotheses from first and second order logic, about where recursive reasoning hits the bottom and so on, and I came to the conclusion that if you actually find some mathematical formulation of the falsifiability criterion Popper, then it must be deeper than Bayes' Theorem. In other words, Bayes' Theorem shows not that positive cognition also exists, it's just weaker, but that negative cognition has indirect effects that can be mistaken for weak positive cognition. If we try to formulate it concretely, then Bayes' Theorem is usually depicted as a square, divided in one dimension by one line, and in the other by two. And after you've done the calculation, you cut off the probabilities that didn't happen, and then you normalize it back to a nice square. However, if you treat this square as a space of hypotheses from which you cut out falsified clusters of hypotheses, then you will see that no probabilities ever increase, just some fall much more than others and in a relative ratio less fallen ones look larger, and after normalization this difference generally lost. The main plus of such a view is that there is no crisis of confidence in it, in general, in principle, you cannot confirm something, you can only more or less refute it. So bits of information become rebuttal or contradiction scores, you don't validate the bit's current value, you disprove the opposite, because that would mean a contradiction. The probabilities of this number are less than one, therefore, with mutual multiplication, they can always only fall, but you look at the probability distribution for those that fell the least. For example, religion has already been pierced by so many spears that in order for you to reach such a low probability, we need to lower even lower all possible probabilities, which are now higher. But for quantum mechanics, it doesn’t matter if there are any flaws in it, it still remains our best hypothesis. In other words, it allows you not to reach self-destruction even if you are a contradictory agent, you just need to look for a less contradictory model. And this also works in comparison between agents, no one can increase someone's rating, you can only find new contradictions, including in yourself, the one with whom others found the least contradictions wins. In a broader sense, I believe that contradictions are generally a more fundamental category than lies and truth. False is something that is contradictory only in combination with some external system, so it can win in the system with which it does not contradict. And there are also things that are already contradictory in themselves, and for them it will not work out to find an ideal external world, the number of contradictions in which reaches zero. In other words, contradictory things in themselves are worse than things that contradict only something specific, but there is no fundamental difference, even though false, even contradictory systems do not destroy the mechanism of your knowledge. In addition, of course, in addition to false, true and contradictory, there is a fourth category, indefinite, they certainly score the least points of contradiction, but they are not particularly useful, because the essence of Bayesian knowledge is to distinguish between alternative worlds among themselves, and if a certain fact is true for all possible worlds, it does not allow you to discern which world you are in. However, this does not mean that they are completely useless, because it is precisely from such facts that all mathematics / logic consists, facts that are true for all possible worlds, distinguishing them from contradictory facts that are not true for all possible worlds. That is, in other words, the meaning is again that it is impossible to prove something, mathematics is not a pool of statements proven true for all worlds, it is rather a pool of those statements that have not yet proven that they are wrong for all worlds.
Some time ago I saw an article here on the topic, but what do the words "deep causality" and "surface analogy" mean. For me personally, at that time it was intuitively obvious what the difference was, including for me it was obvious that the article was not about deep analogies, but only about the concentration of the probabilistic mass, which of course is a very important skill for a rationalist, actually key, but that's just not what I mean by deep causes, at least. However, despite this, I could not express my intuition in words at that time. I wasn't even sure if the analogy could be "truly deep" or just "more profound" than any other. Since then, I have better understood and deduced for myself the concepts of upward and downward understanding, that is, calculating large consequences from simple base-level laws, or trying to determine basic laws by seeing only large consequences. And in the first case, you can confidently talk about the deepest level, in the second, you can’t, so in theory it can always turn out that you didn’t know the true laws and there is a level even deeper. An obvious analogy for the surface and the deep is the black box, where the surface is a purely statistical analysis of purely patterns in inputs and outputs, and the deep is an attempt to build a model of the inside of the box. The difference is that when you talk about statistics, you are always ready to say that there is always some possibility of a different outcome, this will not disprove your model, and when you build the internal structure of the box, your hypothesis says that some combinations of inputs and outputs are strictly impossible. according to your model, if this appears, then your model is wrong, deep causal models are more falsifiable and give clearer answers. You can also say that the difference is like between a neural network with two layers, input and output and connections between them, and a neural network with a certain number of hidden, invisible, internal layers. The difference is that deep causal networks require not just saying that there is a certain correlation between the states of inputs and outputs, it requires building a chain of causes between inputs and outputs, laying a specific path between them. And in the real world, you can also often check the specific steps of that very path, not just the ins and outs. But it can also be compared to a logical versus probabilistic construction, you can "ring out" this circuit and clearly say which outputs will be at which inputs. And as in inference, if you deny the conclusion, you need to point out the specific premises that you then reject. You cannot reject the conclusion without refuting the whole structure of this model. Like that. Probably later I will formulate it even better and add it.
I noticed here that Eliezer Yudkowsky in his essays (I don't remember exactly which ones, it would be nice to add names and links in the comments) says that the map has many "levels", and the territory has only one. However, this terminology itself is misleading, because these are not close to "levels", these are "scales". And from this point of view, it is quite obvious that the scale is a purely property of the map, the territory does not just have one scale, the smallest, and it cannot even be said that it has all the scales in one, it simply does not have a scale. Scale is a degree of approximation, like distance is about photographs, different photographs can be taken from different distances, however, the object is not the closest photograph and not all of them put together, it is simply NOT a photograph and there simply is no scale, distance or degree of approximation, it is all the categories that refer to the relationship of the subject and the photograph when shooting, however the subject never shot itself, there were no shooting distances. Talking about levels makes it feel like there could very well be many levels, they just don't exist, however, when talking about scale, it's obvious that the territory is not a map, there is no scale, just like there is no cross of your current location or icons for points of interest. And the scale just fits perfectly into the analogy with the map and the territory.
As for the many attack vectors, I would also add "many places and stages where things can go wrong", AI became a genius social and computer hacker. (By the way, I heard that most hacks are carried out not with the help of computer hacking, but with the help of social engineering, because a person is a much more unreliable and difficult to patch system) From my point of view, the main problem is not even that the first piece of uranium explodes so that it melts the Earth, the problem is that there are 8 billion people on Earth, each has several electronic devices, and processors (well, or batteries for a more complete analogy) are made of californium. Now you have to hope that literally no one in 8 billion people will cause their device to explode (this is much worse than expecting that no one in just 1 million wizards will be prompted with the idea of transfiguring anti matter, botulinum toxin, thousands of infections, nuclear weapons, strandels, as well as things like "only top quarks", which cannot be imagined at all), or that literally none of these reactions will go as a chain reaction through all processors (which are also connected to a worldwide network operating on the basis of radiation) in form of a direct explosion or neutron beams, or that you will be able to stop literally every explosive / neutron chain reaction. We can conditionally calculate that for each of 8 billion people there are three probabilities that they will not fail all three points, and even if on average each of them is very high, we raise each of them to the power of 8 billion, worse, these are all probabilities in a certain period of time, conditionally, a year, the problem is that over time, not even the probabilities grow, but the interval for creating AI is shortened, so that we get the difference between a geometric and exponential progression. Of course, one can say that one should not consider the average over all, that the number should be reduced for all but the number of processors, but then the number of people who can interfere will be reduced, and the likelihood that one of them will create AI will increase, and again, the problem is that it's not the chance of creating AI that increases, but the process becomes easier, so that more people have a higher chance of creating it, and that's why I still count for all people. Finally, we can say that civilization will react when it sees not smoke, but fire. But civilization is not adequate. Generally. Only here she did not take fire-fighting measures and did not react to smoke. She also showed how she would react to the example of the coronavirus. But only here, "it's not more dangerous than the flu. Graphic is exponential? Never mind", "it's all a conspiracy and not true danger", "I won't get vaccinated" will be added "it's all fiction / cult", "AI is good" and so on.
I think it could have been written better, I found it a little stretched, especially in the beginning and middle (the ending looks very powerful), it could also be better with more references to already known concepts like "lost goals". But at the same time, it looks like a very important post for instrumental rationality, epistemological rationality is well solved by sequences, but instrumental seems to be what most people lack for a good achievement of goals, this is a more significant node in the tree (at least considering how bad everything is with its implementation). And this, among other things, looks like one of the significant nodes for my instrumental rationality, that, the understanding of which I lacked, both to the upper nodes and to the lower ones. Although I suspect that there are still not enough other nodes of the same importance. (I'm confused no one has written a comment like mine and that there are so few comments at all)
It occurred to me that looking through first-order logic could be the answer to many questions. For example, the choice of complexity by the number of rules or the number of objects, the formulas of quantum mechanics do not predict some specific huge combination of particles, they, like all hypotheses, limit your expectations compared to the space of all hypotheses/objects, so for at least complexity according to the rules, at least according to objects will be given one answer. At the same time, limiting the complexity of objects should be the solution to the Pascal robbery (the original articles have no link if they are already solved), this is the answer, where the leverage penalty comes from. When you postulate a hypothesis, you narrow the space of objects, initially there is much more googolplex of people, in fact, but you specify only a specific googolplex as axioms of your hypothesis, and for this you need the corresponding number of bits, because in logic an object with identical properties cannot be different objects (and if I'm not mistaken, quantum mechanics says exactly that), so that each person in the googolplex must be different in some way, and to prove / indicate / describe this you need at least the logarithm of bits. And as long as you're human, you can't really even formulate that hypothesis exactly, define all the axioms when that is 1 your hypothesis is 1, let alone get enough bits of evidence to establish that they really are 1. But also the hypothesis about any number of people is the sum of the hypotheses "there is at least n-1 people" and "there is one more person", so increasing its probability by a billion times it will be literally equivalent to believing at least that part of the hypothesis where there are a billion people , which will be affected by the master of the matrix. This can also be expressed as that each very unlikely hypothesis is the sum of two less complex and unlikely hypotheses, and so on until you have enough memory to consider them, or in other words, you must start with more likely hypotheses, test, and only then add new axioms to them, new bits of complexity. Or a version of leverage penalty, only not for the probability of being on such a significant node, but for choosing from the space of hypotheses, where for the hypothesis about the googolplex of people there will be a googolplex of analogues for smaller numbers. That is, according to first-order logic, our programs have unreasonably high priors for regular hypotheses, in fact, infinite, although in fact you have to choose from two options for each bit, so the longer the set of certain bit values, the less likely it will be. Of course, we have evidence that things behave regularly, but not all evidence goes there, much less an infinite amount of it, since we haven't even tested all 10^80 atoms in our Hubble sphere, so our prior is larger the probabilities of regular hypotheses will not be strong enough to overpower even a googolplex.
It seems to me that the standard question on the conjunction error about "the probability of an attack by the USSR on Poland as a result of conflict X" and "the probability of an attack by the USSR on Poland" is inaccurate for this experiment, since it contains an implicit connotation that once in the first reason X is named, and in the second case, Y or Z is not indicated, then in the second case we evaluate the attack for no reason, and if we continue to show the person himself his answers to this question, the effect of hindsight comes into play, like that experiment with the substitution of the selected photo. It seems to me that a more correct question, in order not to create such subconscious false premises, would be "the probability of an attack by the USSR on Poland as a result of conflict X" and "the probability of an attack by the USSR on someone as a result of any conflict between them." Although I'm not sure that this will change the results of the experiment as a whole. At least because even with an explicit indication of "some" instead of an implicit premise of "no reason", something so multivariate and vague will still not look like a plausible plot, as a result, vague = indetailed = implausible = unlikely = lowprobable.
I liked the writing style. But it seems that no one in the comments noted the obvious that it’s not “you control the aliens”, but “the aliens control you” (although this sounds even crazier and like a freak in everyday life), in other words, you are in any case a simulation, but whose results can predict the decision of another agent, including a decision based on prediction, and so on. This removes all questions about magic (although it can be fun to think about it). Although this can cause a problem for the calculator, which, instead of determining the result "5 + 7", will try to determine what it will give out for the query "5 + 7", but will not work on calculating the sum.