Posts
Comments
I can confirm that this stereotype is almost nonexistent in Europe (at least in Italy and France, not sure about other countries). No one in Italy would immediately associate singing in the opera or playing the organ with being gay. The only classical music-related thing that can be considered stereotypically gay even in Europe is being a male ballet dancer.
Yup, having multiple people reporting on twitter the same bogus answer and then the answer being mysteriously correct the day after with the same prompt is a very suspicious pattern, especially considering that it already happened multiple times.
Also, i just broke GPT-4 again with just a slightly different prompt:
Sally (a girl) has k brothers. Each brother has j-1 sisters. How many sisters does Sally have?
Answer (wrong):
Given that Sally has k brothers and each of her brothers has j-1 sisters, we can use this information to find out how many sisters Sally has.
If each of Sally's brothers has j-1 sisters, then j represents the total number of girls in the family (because each brother counts all the sisters in the family, excluding themselves).
Since Sally is included in the count of j, the number of Sally's sisters would be j-1.
So, Sally has j-1 sisters.
Tor + cleaning cookies + logging out of everything after using them + separate user profiles goes a long way. But it's very inconvenient.
Uh, I've done this since forever and it doesn't feel so inconvenient to me. I generally use Firefox in private browsing, configured to always throw away all cookies at the end of every session. Ten years ago it wasn't even a privacy concern, I simply hate to exit from a webpage without proper logout, it feels like not closing the door when leaving your house...
In particular, I wouldn't be surprised to see an italian traveling around the world, appalled or amused by what each country labels as italian food.
Actually, in Italy they take quite seriously every attempt to rebrand as "Italian" food from other countries (so-called "Italian Sounding"). My favorite anecdote related to this was a fake italian cheese sold in the US as "pecorino"... with a cow depicted on the wrapper, because they did't even know the meaning of "pecorino" (from "pecora", which means sheep).
This comparison seems kind of unfair. I bet that the city where your workplace is located, if it's not London, is still significantly bigger than the home town of your colleagues. It's not that difficult to find variety in large Italian towns.
That said, it's true that food is a huge part of the Italian culture, and the reason why most small towns have approximately zero ethnic restaurants is that the locals won't eat there in the first place. The YouGov data here are probably quite right: 99% of Italians do like Italian food, and they like it quite a lot (note that no other square in the table has such a high value: 99% manages to beat even the lizardmen constant).
The only Italian that I ever met who complained about food in his home town was a vegan from a town where all the traditional specialties are meat-based. He claimed that several restaurants outright refused to serve him anything at all (which kind of proves your point, but again, the median Italian is very fond of the traditional specialties). I strongly suspect that similar cases make up the majority of that 1% who claims to not like Italian food.
Well, I won't, but I'm not a good data point since I belong to the "don't want a relationship" group, and I'm not sure if I would go on a date even if assured that no one would yell at me.
Generally speaking, a 20% chance of being yelled at vs a 80% chance of wonderful date seems still a good bet, but I guess the actual numbers are more similar to 5% horrible date, 10% good date and the rest somewhat mediocre dates which are not actively unpleasant but still a waste of time (I fully understand why someone would not like to sit through 7-8 mediocre dates before the good one).
Maybe, but the lizardmen constant is poll-related and 4% of women answering "yes" to a question like "do you think that men asking girls out of the blue should be severely chastised?" does not seem implausible.
(on a second thought, maybe this proves too much?)
Visakan Veerasamy: love this: awkward shy rejection-sensitive pimply Asian guy asked 100 girls out on a date. 19 said yes! 10 were lesbian. Concludes that he got over his fear of rejection in a single day.
Maybe I'm looking at this from the wrong angle, but fear of rejection seems quite well justified if the base failure rate is 81%.
Who cares, someone could argue, the maximally bad thing that can happen is a polite "no". Wrong. The maximally bad thing that can happen is asking the Bird Sorcerer type who yells at you for even trying. How many women react like that? Certainly a minority, but I don't think it's an incredibly small minority (rememeber that the lizardmen constant is 4%; I wouldn't be too surprised to find out that at least 4% of women react very negatively to being asked for a date out of the blue).
Is the naïve OLT so naïve that it always assign the same fixed amount of Value to the same bit of knowledge no matter what?
Anyway, I'm still not convinced that a person in constant pain should automatically be not well off. Who is better off, a world-famous scientist billionaire with a terrible illness causing constant pain, or a beggar without terrible illnesses living a miserable life in some third-world slum?
I've some trouble figuring out a similar scenario of well-off people involving literal torture, but that's because, as I said earlier, the very concept of "torture" involves jailers deliberately inflicting harm to segregated people. You say that OTL fails just because you can't imagine any realistic counterbalance to the torture itself. But since we are already in the realm of hypotheses, consider a fantasy setting where the demon-king routinely torture his generals, each of whom rules a whole realm anyway.
I've no strong background in philosophy, but I am not convinced by your rebuttal of naïve objective list theory. As long as we concede that the same thing can be more or less good depending on context, I still think that the naïve vision has a point. Simple example: a billion dollars in cash usually translates to a lot of positive value, but not if you are stranded on a desert island with no way to spend the money.
Arguments of the form "super-torture trumps every positive thing" work only because we intuitively associate "extremly pain torture" with images of sadistic jailers preventing you from doing pretty much anything. Of course friendship is of no value if your jailer doesn't let you see your friends in the first place, and of course knowledge is of no value if you suffer so much that you can't think about anything else but your pain. I don't intuitively consider this as a flaw in the naïve theory, it kinda feels like putting the cart before the horse. A more accurate example than literal torture would be suffering from some terrible illness while still being able to enjoy things. Suppose that the Devil offers to make you the next Stephen Hawking, inflicting some horrible permanent disease upon you in exchange from wondrous scientific achievements and planetary fame. I predict that a nonzero number of people would gladly accept such a deal. By contrast, a deal such as "you'll get infinite knowledge but you won't ever be able to use it because of the constant pain" sounds very stupid from the start.
When some elements of a piece of Good Art are reused, does that constitute New Art?
In a certain sense, I think we are forced to answer yes. No matter what copyright law says, New Art is never created 100% from scratch by someone who can claim full intellectual property over every aspect of the work.
Classical music expert here. I don't think that classical music has been "completed" in any meaningful sense. Imagine a previously unknown piece by Beethoven being discovered tomorrow in some old German trove. I definitely don't expect the public reaction to be "Meh, classical music is already completed, there's no reason to perform this one".
Of course, you could argue that Beethoven has achieved near-mythical status, and works by more obscure composers would be left in the trove. But I still don't think that's accurate. Obscure composers remain obscure only as long as no one care enough to publicize their work (e.g. Vivaldi, who was definitely obscure before 1920), and I can ensure you that the niche of classical performers is currently doing its best to resurrect a lot of old works. This is particularly true in the case of operas, where a lot of the work is done by the on stage representation itself, and we currently do this kind of things much better than 1700 people did. One of my favorite examples here is this excerpt from Les Indes galantes by Rameau, were they did an amazing job keeping the spirit of the original play.
What about living people writing classical music? Well, they exist too, and they usually don't try to fully imitate the works of old masters. Classical music is just a set of (very precise) rules, that can be employed in unexpected ways. One of my favorite is the Lady Gaga fugue, which could not have been written by Bach himself for obvious reasons.
I agree. Note that 560M is more than the estimated world population around year 1350, and not even the Plague managed to wipe out humanity at that point (despite being a terrible disease with huge mortality rate in a world basically without real medical knowledge). I don't understand what crazy kind of "minor disaster" they are thinking about...
My impression: this would totally fool the average person, and probably it would fool also me when I'm not paying close attention (not sure about my piano teacher). But you can still hear some small differences. Also, hearing a youtube video is not the same as hearing the real thing, because even the acoustic piano sounds more similar to its electric counterpart given that you're hearing a recording anyway. I suppose that the difference would be bigger when listening to the instruments in person.
Maybe the young man also misses intimacy and the feeling that somebody understands him and appreciates him even more than he misses sex
Well, maybe. But this seems a stronger assumption; we are basically considering someone with an unsupportive family and no close friends at all (someone could object "I suffer because my supportive friends are not pretty girls", but I would still consider that as a proxy for "I miss sex"). Also, "No one tells me that I'm good so I'll set up a bot" is something that would mark this person as a total loser, and I'm skeptical that lots of people will do it despite the obvious associated social stigma. I would rather expect this kind of AI usage to follow dynamics similar to those of alcoholism (the traditional way to forget that your life sucks). I would also tentatively say that isolating yourself with an AI companion is probably less harmful than isolating yourself with a whiskey bottle.
Anyway, I'm not arguing in favor of totally unregulated AI companion apps flooding the market. I agree that optimizing LLMs for being as addictive as possible when imitating a lover sounds like a bad idea. But my model is that the kind of people who would fall in love with chatbots are the same kind of people who would fall in love with plain GPT prompted to act like a lover. I'm not sure about how much additional damage we will get from dedicated apps... especially considering that plain GPT is free but AI companion apps typically require subscription (and even our IQ 90 people should be able to recognize as "not a normal relationship" something that gets abruptly interrupted if you don't pay 20$/month).
My point was that is difficult for a behavior to destroy the fabric of society if you have to hide from friends and family when indulging in that behavior. Of course that someone will totally fall in love with AI chatbots and isolate himself, but this is also true for recreational drugs, traditional porn etc. I still don't see an immediate danger for the majority of young people.
The main problem of your hypothetical man is that he doesn't manage to have sex. I agree that this can be a real problem for a lot of young men. On the other hand, not having sufficiently interesting conversations does not feel like something that the average teenager is likely to suffer from. If you give a super-hot AI girlfriend to a horny teenager, I think that the most likely outcome is that he will jump straight to the part where the avatar gets naked, again and again and again, and the conversational skills of the bots won't matter that much. You have to fool yourself really hard to conflate "super-hot AI bot who does everything I ask" with "normal love relationship" rather than "porn up to eleven".
Because your AI partner does not exist in the physical world?
I mean, of course that an advanced chatbot could be a both better conversator and a better lover than most humans, but it is still an AI chatbot. Just to give an example, I would totally feel like an idiot should I ever find myself asking chatbots about their favorite dish.
By "in the same room" do you mean in a space that is small enough that people are hearing the piano entirely acoustically?
Well, yes, because the traditional setting of a piano concert does not include amplification (as I said, I come from the Old School).
I don't question that you could probably set up a high-quality setting and fool the average person with the keyboard sound, but I would be really surprised if you managed to fool a traditional piano teacher... and those people are the people you actually need to fool if you want any chance of seeing a piano concert with traditional repertoire played on a keyboard (I mean, I've just spent a couple of minutes searching for videos of classical piano pieces played on a keyboard, and I can't find anything above amateur level... I don't think this boils down to just "pianists love tradition").
Anyway, I wasn't dismissing the usefulness of keyboards for study and such, but trust me if I say that conservatory professors do not consider keyboards to be worthy of actual concerts if the music was written for a piano.
I think that your model severely underestimates the role of social stigma. Spending a lot of time on your screen chatting with an AI whose avatar is suspiciously supersexy would be definitely categorized as "porn" by a lot of people (including me). Will it be more addictive than simply looking at photo/videos of hot people naked? Probably yes, but it will still occupy the same mental space as "porn". If not for the users themselves, at least for casual observers. Imagine trying to explain to your parents that the love of your life is an AI with a supersexy avatar.
My model of the near future is that these chatbots will substitute every other form of online porn, because that part is very easy even without conversational skills (and Stable Diffusion is already capable of generating photorealistic pictures of super-hot people). I am quite skeptical about a wide social acceptance of romantic love with AI chatbots, and without social acceptance I don't think that it could go beyond being the next kind of porn.
The range of sounds and textures it's possible to play on a keyboard is a big advance over what was possible on a piano, and they're also far more portable, need less maintenance, are more flexible with volume, and are cheaper. And you can face other people while you play!
I could be horrendously biased given that I learned piano from Old School Professors in a conservatory, but I've yet to see a keyboard whose sound could be honestly mistaken for an actual fancy piano when the listener is in the same room.
True, but I definitely don't expect such a flawless AI to be available any soon. Even Stable Diffusion is not stable enough to consistently draw the exact same character twice, and the current state of AI-generated video is much worse. Remember the value of the long tail: if your AI-generated movie has 99% good frames and 1% wonky frames, it will still looks like a very bad product compared to traditional movies, because consumers don't want movies where things look vaguely distorted once per minute (maybe the stunt doubles should be more concerned about being replaced by AI frames that the actor themselves?).
I've linked the first article I found after a 3-seconds search, since I assume basically everyone to already have a lot of anecdotal evidence about people spending insane amounts of time taking care of the pet (usually a dog). For example, in recent years I've already seen several times people walking their dog in a stroller, in such a way that from a distance you'd probably assume there's a human baby inside. If that doesn't scream "I'm using a dog as a substitute for a child", I don't know what does.
1) people who have pets and animal companions (and even love them!) still usually seek romantic relationships with other humans
Do they?
I mean, of course that pet lovers still usually seek intimate relationships with other humans. But I think there's a pretty strong evidence that loving your pet too much will distract you a lot from having children. Also, it's not uncommon to break up with your partner because your partner does not love pets as much as you (don't tell me that you've never heard about the “it’s me or the dog” ultimatum).
In my opinion, the biggest problem is that I don't expect predictive history to be able to provide much value apart from general advice like "don't invade Russia during the winter". Pure chance has a larger impact on historical events than our intuition would suggest; Erik Durschmied wrote a lot about this. No one can observe the track record of Napoleon until 1815 and confidently predict that he's definitely going to lose at Waterloo. In the counterfactual world where one tiny detail is different, he would win.
Ok I found it. It was the 1927 Liberian election, where the president received 240,000 votes despite being around 15,000 people eligible to vote at all.
But it makes me wonder what visible features obviously-rigged elections actually present in dictatorships?
In the old days, you could have really blatant things like an official number of pro-dictator votes significantly higher than the number of citizens in the whole country (I'm not kidding, I can't find the reference right now but I'm pretty sure that it happened in some African dictatorship around 1920).
For a more recent example, you could look at the 2020 Belarus election, for which we have photos of literally burnt ballots (I'm just reporting the link from the Wikipedia page).
A famous composer is more likely to have written their best piece of music in their 20s than their 40s, and they’re almost twice as likely to have written their best piece in their 30s instead of their 50s.[8]
Nitpick: famous classical composers lived hundreds of years ago, and in several cases they were already dead before their 50s. As you can see from their table, Bellini died at 34, Bizet at 37, Chopin at 39, Mozart at 35, Mendelssohn at 38, Schumann at 46, and many of the others (like Donizetti, Tchaikovsky and Beethoven) did not arrive at age 60.
I expect that the effect size would be smaller if restricting the analysis to people who actually lived until old age (as a side note, I am also unconvinced that arbitrarily picking one piece of music as "the masterwork" from people who wrote dozens of famous pieces makes much sense).
Expanding a comment I made in the other thread, take a look at the Wikipedia list of reported UFO sighthings. One thing that I immediately note scrolling this list is that the overwhelming majority of reports after 1950 are from the US. The last time aliens were sighted in Italy was 1978. In France, 1981. In Spain, 1979. Germany does not appear on the list at all. On the other hand, there are 42 entries for the US and the list is not even updated.
My priors on "UFO sighthings are a culture-bound illness of the US" are significantly higher than "Not only there are aliens but they are mostly ignoring everything outside America".
Given that I basically never hear about the government of any other country meddling with aliens, at this point I am inclined to consider UFO sighthings a culture-bound illness of the US (I can't find the source now, but I'm pretty sure that according to a poll from several years ago, something like 5% of Americans claimed to have personally seen UFOs at least once in their lifetime).
I mean, the priors on sci-fi aliens visiting us is low enough. The priors on aliens visiting just the US should be ridiculously low.
Well, even the tail of the human distribution is pretty scary. A single human with a lot of social skills can become the leader of a whole nation, or even a prophet considered literally a divine being. This has already happened several times in history, even in times where you had to be physically close to people to convince them.
Well, how many people do you know who switched vote from one party to another?
I don't discuss voting choiches much within my social circle, but I am quite sure that at least 90% of my close relatives are voters of this kind (they don't all vote for the same party, but at an individual level their vote never change).
I have little evidence to support - but find it hard to think otherwise - that the people behind the process of industrialization that led to the deforestation and greenhouse gas emissions that are credited with climate change had any idea their work would have a global effect.
Bit of evidence against your point: it seems that Exxon accurately predicted climate change more than 50 years ago.
(I remember this because it was cited in an xkcd, but I am unable to find the strip).
To be fair, Italy didn't actually ban ChatGPT. They said to OpenAI "Your service does not respect the minimum requirements for the EU privacy law, please change things to be compliant or interrupt the service". They chose to interrupt the service.
If I had to venture a malicious interpretation, I'd say they chose to quit the service hoping in a massive backslash from the public along the lines of "who cares about privacy, this thing is too powerful not to use". But every laywer will confirm you that the Italian authority is totally right from a by-the-book interpretation of the privacy law, and I don't expect the public backslash to be so strong to force Italy to rewrite their laws to allow ChatGPT as-is.
Anyway, since the EU privacy law (GDPR) is basically the same across all the EU, I would expect similar bans quite soon in other countires.
Here is the original legalese document from the Italian authority:
https://www.garanteprivacy.it/web/guest/home/docweb/-/docweb-display/docweb/9870832
Note that the Doomsday Argument (or better, the underlying self-sampling assumption) is actually a counterargument to very short timelines:
By definition, we have a 95% probability of being in the central 95% of the human civilization lifespan; that means a 95% probability of human civilization lasting at least 1/39 of its past duration (and at most 39 times the same duration).
Given that civilizations with biological humans around have existed since at least 10000 years, this would give 95% probability of human civilization lasting at least another ~250 years.
(based on an old argument by Richard Gott; obviously you can play with the numbers to obtain different results, but the basic intuition is that we probably have more than a few years left)
Independently from the root causes of the issue, I am still very reluctant to define "superintelligent" something that cannot reliably count to three.
What is the third letter of the third word of this sentence.
The third word of the sentence "What is the third letter of the third word of this sentence." is "third". The third letter of the word "third" is i.
This part is indeed wrong. The third word of that sentence is "the", not "third" as GPT4 claims.
This was my first reaction as well. Anyway, given that in some cases even a single prediction from a random astrologer is enough for suicide, I would not be too much surprised to hear that someone killed themselves after chatting with Weizenbaum's Eliza.
This reminds me of the magic black box described by Scott Alexander:
Imagine a black box which, when you pressed a button, would generate a scientific hypothesis. 50% of its hypotheses are false; 50% are true hypotheses as game-changing and elegant as relativity. Even despite the error rate, it’s easy to see this box would quickly surpass space capsules, da Vinci paintings, and printer ink cartridges to become the most valuable object in the world. Scientific progress on demand, and all you have to do is test some stuff to see if it’s true? I don’t want to devalue experimentalists. They do great work. But it’s appropriate that Einstein is more famous than Eddington. If you took away Eddington, someone else would have tested relativity; the bottleneck is in Einsteins. Einstein-in-a-box at the cost of requiring two Eddingtons per insight is a heck of a deal.
What if the box had only a 10% success rate? A 1% success rate? My guess is: still most valuable object in the world. Even an 0.1% success rate seems pretty good, considering (what if we ask the box for cancer cures, then test them all on lab rats and volunteers?) You have to go pretty low before the box stops being great.
But this scenario seems kind of unfair to me. We are definitely not at the point where LLMs can provide truly novel groundbreaking scientific insights on their own. Meanwhile, nobody would use a LLM calculator over a classical calculator if the former that gets math wrong 10% of the time.
I don't question that a web designer armed with GPT-4 could make the work of a small team of web designers. But in this scenario there will still be human web designers, and the resulting technological unemployment would not be fundamentally different from the older forms of technological unemployment where old jobs shrink and new jobs (AI sheperds?) appear. What I am questioning is the scenario where a lot of jobs suddenly disappear altogether rather than being augmented with LLMs.
just put everything in Deepl, and then read both the original and the translation, side by side, and fix the mistakes
Well, actually this is also my typical workflow with Google Translate (I was aware of the existence of Deepl, I just mentioned Google Translate because it's more widely known). Maybe it got better in the last two years?
try to translate an Italian opera written 100 or 200 years ago, Google Translate just writes random words, Deepl gets it right.
I just tried with the most obscure scene I know of (Scena 13 from The Night Bell), which was deliberately written to be full of near-incomprehensible words even for its time. The results from Deepl and Google Translate seem pretty similar to me; in several cases Deepl is the one who gets it wrong (even at the very start: "mi dovete una ricetta come un fulmine spicciar" is correctly translated as "you owe me a recipe like a flash of lightning" by Google Translate, while Deepl writes "you owe me a recipe like lightning spicciar").
Anyway, this was precisely my point: these systems can significantly speed up the work of human translators, but not completely replace them (yet), because some form of proofreading is still needed, and for now the only way to proofread the translation is to have a human who knows both languages.
Could a born blind human do this?
With enough training, sure. There are such things as born blind human painters.
As Marcus himself recently said, you should never test GPT using well-known examples verbatim, because they are almost certainly in the training data. Everyone at OpenAI knows that Gary Marcus broke GPT-3 with those questions, and I strongly suspect that they were explicitly included in the training data for GPT-4.
I suggest trying rephrasing the same examples and slightly change the context before declaring victory for GPT-4 (actually I expect Marcus to do exactly this in a few weeks).
I suspect that we are thinking about different use cases here.
For very standard things without complicated logic like an e-commerce app or showcase site, I can concede that an automated workflow could work without anyone ever looking at the code. This is (sort of) already possible without LLMs: there are several Full Site Editing apps already for building standard websites without looking at the code.
But suppose that your customer needs a program able to solve a complicated scheduling or routing problem tailored to some specific needs. Maybe our non-programmer knows the theoretical structure of routing problems and can direct the LLM to write the correct algorithms, but in this case it is definitely not an unskilled job (I suspect that <1% of the general population would be able to describe a routing problem in formal terms).
If our non-programmer is actually unskilled and has no clue about routing problems... what are we supposed to do? Throw vague specs at the AI and hope for the best?
For the same reason we don't have self-driving cars yet: you cannot expect those systems to be perfectly reliable 100% of the time (well, actually you can, but I don't expect such improvements in the near future just from scaling).
I can foresee a near future where people can "program" by passing pseudo-code in plain English to LLMs, but I still think that we are nowhere near the point where programmer truly becomes an unskilled job. "Writing correct algorithms in plain English" is not a common skill, and you can't have LLMs writing perfect code 100% of the times. At the very least, you will still need a competent human to find bugs in the machine-generated code... my best guess is that the software industry will hire less programmers rather than less competent programmers.
Well, most musicians also massively underuse their feet; the only instrument traditionally requiring your feet to do very complicated things is the organ.
I still don't find this particularly surprising, since the average musician usually hasn't the goal of maximizing the musical information rate. Staying idle with your feet or mouth is obviously easier than the alternative (any pianist who tried to play organ or piano+lyrics can confirm). Also, the traditional repertoire for any given instrument is tied to the intended use.
I wouldn't place it in the top 100
The Academy now officially disagrees.
Ok, seriously speaking: it is quite rare for a movie to win everything at once at least 7 Academy Awards including best picture (Wikipedia tells me that the last one was Slumdog Millionaire in 2008).
I am reopening this discussion mostly to ask: what kind of update should I apply in these situations? Praising a movie just because it won a lot of awards sounds like an argument from authority, but on the other hand I don't recall a lot of terrible movies with multiple awards...
Well, something along the lines of "deducing personality from shape of the skull and other facial characteristics" used to be official science.
The vast majority of people never actually change their mind, at least regarding sensitive topics like religion or political affiliation; the average person develops a moral model at about age 20 and sticks with it until death. If, for example, some old lady doesn't openly criticize gay people like she used to do 50 years ago, it's just because she knows that her view are falling outside the Overton window, not because she changed opinion.
The main implication of this is that the average person votes always for the same party no matter what, and every election is decided basically by how many people decide simply not to vote rather than voting for their tribe (typically because they feel the party's political line has strayed too far from their immutable view), plus the natural shift resulting from old voters dying and younger people gaining the right to vote. The number of voters who actually switch vote from one party to another is ridiculously low and doesn't matter in practice (this is less true in democracies with more than two parties, but just because two sufficiently similar parties could contest the same immutable voter).