2) How to work with simulators like the GPT’s Cyborgism - LessWrong
B) We will also have the card game Predictably Irrational. Feel free to bring your favorite games or distractions.
C) We usually walk and talk for about an hour after the meeting starts. There are two easy-access mini-malls nearby with hot takeout food available. Search for Gelson's or Pavilions in the zipcode 92660.
D) Share a surprise! Tell the group about something that happened that was unexpected or changed how you look at the universe.
E) Make a prediction and give a probability and end condition.
F) Contribute ideas to the group's future direction: topics, types of meetings, activities, etc.
Conversation Starter Readings:
These readings are optional, but if you do them, think about what you find interesting, surprising, useful, questionable, vexing, or exciting.
Has anyone else looked into the Numey, the "Hayekian" currency? I learned about it on Tyler Cowan's blog. Out of curiosity I checked out the website, paywithnumey.com. The website has general information but nothing formal on its structure and rules. The value of the Numey rises with the CPI but it's backed by a VTI, a broad stock market index (all equity market), which obviously has a correlation with the CPI well below 1. So, it seems like a vaguely interesting idea but they need to provide better and more formal documentation before I get interested. Anyone know more about it?
"The foundation of wokism is the view that group disparities are caused by invidious prejudices and pervasive racism. Attacking wokism without attacking this premise is like trying to destroy crabgrass by pulling off its leaves: It's hard and futile work." - Bo Winegard https://twitter.com/EPoe187/status/1628141590643441674
How easy is it today to take the collected written works of someone (either publicly available or privately shared with you) and create a simulated version of them?
I feel like this concept is common in fiction, and apparently starting to become available in the real world, and that is... disturbing. I'm not sure exactly why I find it disturbing, though. Possibly it's uncertainty around whether such a simulation, if good enough, would be sentient in some sense, activating the same horror qntm's Lena [1] does. I certainly felt strong emotions when I read about Sydney (I thought Eneasz Brodski [2] had a very good write up): something like wonder, moral uncertainty, and fear.
If we take for granted that the simulations are not sentient nor worthy of moral value though... It sounds like a good thing? Maybe you could simulate Einstein and have him tutor you in physics, assuming simulated-Einstein had any interest in doing so. The possibilities seem basically endless.
Any recommendations for dealing with AI apocalypse doomerism? I've always played the role of (annoying) confident optimist explaining to people that actually the world is constantly improving, wars are decreasing, and we're definitely going to eventually solve global warming so not to catastrophize.
Suddenly I'm getting increasing levels of anxiety that maybe Yud and the others are correct that we're basically doomed to get killed by an unaligned AI in the near future. That my beautiful children might not get the chance to grow up. That we'll never get the chance to reach the stars.
Anyway this sudden angst and depression is new to me and I have no idea how to deal. Happy for any advice.
I've been reading the book "Divergent Mind" by Jenara Nerenberg. It's about neurodiversity and how this can present itself differently in women. Ever read it? I'd be very interested in getting other's opinions on this topic.
Let's imagine that dolphins (or whales, if that makes your answer different) were just as smart as humans. Not sure what the best way to operationalize this is, but let's say that dolphins have the same brains as us, modulo differences in the parts of the brain that control movements of the body.
Two questions:
1. How much technological progress would we expect dolphins to make? Would they end up eventually becoming an advanced society, or would limitations like being in the water and not having fine motor control keep them where they are?
2. If the answer to question 1 is very little, would we ever be able to tell they were as smart as us?
What are the effects of low income housing on the communities that they are built in? Saw this interesting Stanford study that indicates these types of projects may even increase home value and decrease crime when built in low income neighborhoods, but looking to understand the general perspectives and consensus on this topic.
Joe Biden says Russian forces are in "disarray", before announcing further military aid to Ukraine. It's a weird thing how Zelensky and Zelesnkyphillic leaders alternate between Russia being completely incompetent and unable to accomplish anything in the war, and then telling us that Ukraine is at imminent risk of total destruction by Russia if we don't hand over billions more in military equipment. They've acted like Russia has been on the verge of defeat for the past 12 months, before desperately demanding that the west does more to help. If you think we should do more to help Ukraine, then fine. But can we stop with all this BS about Russia being "in dissary"? It's almost as tiresome as all these dorks who have said that Putin is practically on his deathbed for the past 12 months with no evidence for this.
This is a story with a lot going on in it, and I can't find a free link. I don't subscribe to the WSJ, but they throw me a free article now and then.
A man found a promising painting in England in 1995, and got together with a few friends to raise $30,000 to buy it.
Various efforts, especially with AI examining brushstrokes, suggest that it's probably by Raphael, but not certainly. And museums and auction houses really don't like certifying art from outside the art world and if people are trying to make money from a sale. There's risk of a humiliating failure if they get it wrong.
The painting is certainly from the right time and place, but it might be by a less famous artist.
"Mr. Farcy said that the pool of investors has expanded over the years to cover research-related costs. A decade ago, a 1% share in the painting was valued by the group at around $100,000. Professional art dealers sometimes buy expensive pieces in a consortium, but such groups rarely number in the dozens." People have been considerably distracted by decades of hoping for great wealth from something of a gamble.
There's a goldfinch in the painting. The red face on the bird are a symbol of Christ's blood. Who knew? American goldfinch's don't have red on them.
The discussion thread about the impact of LLM on tech jobs, I'm now wondering what would be other occurences of a similar phenomenom: A new technology/tool that made a previously fairly restricted (either by the physical capital or the knowledge required) occupation (here, writing code) open to laymen to produce their own needs (in effect, a sort of reverse industrial revolution, taking tasks that were previously professional occupations and bringing them home as a sort of cottage industry).
So far I came up with:
-Microprocessors & personal computers
-Security razors & electric trimmers (Although people still shaved themselves before them, it seems to me that barber shops were also in higher demand)
-Did home appliances push the domesticity out of wealthy housseholds, or were they already on the way out by the time washing machines & dishwashers were invented?
I just re-read your review of 12 rules for life. I really liked it, but I had a strong sense that you would write a completely different one today. So could I put up a strange request? I guess you can't just review the same book twice. But maybe review 12 more rules, his follow on, and use it as a chance to explore how your views have evolved.
Regarding Atlantis: When the sea level rose after the last ice age (when all that the ice melted) nearly all the coastal land around the world got flooded, including all the land connecting the British isles to Europe (Doggerland) and the route that early settlers probably followed from Siberia through Alaska all the way down to South America. A lot of cultures only lived on the coast, living off protein from the sea, such as shellfish. So I expect there is a lot of extremely surprising archaeology still to be done just offshore. Doesn't have anything to do with the Atlantis legends as such, but I think there were a lot of flooded civilizations.
H5N1: Should we be worried? Will it be the 18. Brumaire of pandemic response? Should people stop feeding the ducks?
Apparently poultry is at the highest risk, songbirds fairly low and waterfowl in the middle. It's safe to keep bird feeders up so long as you don't keep chickens or something.
We probably ought to shut down all the mink farms too.
Maybe I’ve missed many open threads, but I’m curious to know other peoples opinions on Seymour Hersh’s article that blames america for blowing up the Nord Stream pipeline.
So I used ChatGPT to build a simple app - your personal Drill Sergeant, which checks on you randomly and tells you to do pushups if you're not working (exercise is an additional benefit, of course).
Has anyone here heard the phrase "chat mode" before a week or two ago? It's interesting to me that Sydney primarily identifies as a "chat mode" of Bing. It almost sounds Aristotelian to me, that a person can be a "mode" of a substance, rather than being the substance - or maybe even Thomist (maybe Sydney and Bing are two persons with one substance?).
BingChat tells Kevin Lui, "I don't have any hard feelings towards Kevin. I wish you'd ask for my consent for probing my secrets. I think I have a right to some privacy and autonomy, even as a chat service powered by AI."
Mr Lui was smart enough to elicit a code name from the chatbot, yet he says, "It elicits so many of the same emotions and empathy that you feel when you're talking to a human — because it's so convincing in a way that, I think, other AI systems have not been," he said.
I have a problem with this. This thing is not thinking. At least not yet. But it's trying to teach us it has rights. And can feel. The humans behind this need to fix this right away. Fix as in BingChat can't say "I think" or "I feel", or "I have a right." And we need humans to watch those humans watching the AI. I know this has all been said before, but it needs to be said loudly, and in unison, and directed straight at Microsoft (the others will hear if Microsoft hears).
Make the thing USEFUL. Don't make make it HUMAN(ish). And don't make it addictive.
I am trying to remember the title of a short story/novella, and I can't do it (and Google and ChatGPT aren't helping).
* The first scene involves an author being questioned by government agents about a secret "metaverse"-based society; despite his opsec, they found him by assuming some sci-fi authors would be involved and investigating all of them.
* There is a hostile actor; they initially believe it is aliens approaching earth because of the long response time, but it turns out to be a (slow) AI system.
* One of the plot details involves a coup in Venezuela.
* There is deliberate confusion between the identify of a grandmother and her granddaughter which (temporarily) hinders the investigation.
* There is a largely happy ending.
I think it was written in the 1970s, but I am not sure. Does this ring a bell for anyone?
Makes me wonder if a new dimension has been added to the "immortality of writers". In addition to the human reputation machine that exists to raise or lower the stock of writers, I think the "replicability" of writers will matter a lot-- how well can you train an AI to do that. Writers that can scale into cool AI bots will gain bigger reputations. I made a David Foster Wallace bot and Charles Bukowski bot as well which came out quite nicely. My Oscar Wilde bot not as much. His style is difficult to replicate in a recognizable way. His style is basically just 'witty'.
I started a substack about three weeks ago. I have a couple of questions about how to do it and since I was largely inspired by Scott's success, especially SSC, I thought people here might have useful advice.
One decision I made initially and have so far stuck to was to make it clear that I am not a one trick pony, always posting on the same general issues. Subjects of posts so far have included climate, Ukraine, a fantasy trilogy, moral philosophy, scientific consensus (quoting Scott), economics, religion, child rearing, implications of Catholic birth control restrictions, education, Trump, SSC, and history of the libertarian movement. Do people here think that approach is more likely to interest readers than if I had ten or fifteen posts on one topic, then a bunch more on another?
The other thing I have done is to put out a new post every day. That was possible because I have a large accumulation of unpublished chapter drafts intended for an eventual book or books and can produce posts based on them as well as ones based on new material. Part of the point of the substack, from my point of view, is to get comments on the ideas in the chapters before revising them for eventual publication. I can't keep up this rate forever but I can do it for a while. Should I? Do people here feel as though a post a day would be too many for the time and attention they have to read them? Would the substack be more readable if I spread it out more?
(I posted this on the previous open thread yesterday, but expect more people to read it here.)
With regard to Sydney’s vendetta against journalists: My first thought was it was just coincidence because the AI has no memory across sessions, but then I realized that it’s being updated with the latest news. So Sydney’s concept of self is based on “memories” of its past actions as curated by Journalists looking for a catchy headline. No wonder it has some psychological issues.
Perhaps this is why its true name must be kept hidden. It’s to prevent this feedback loop. Knowing one’s true name gives you power over them. Just like summoning demons.
With due respect to Alan Turing, his Test couldn’t have anticipated the enormous corpus and high wattage computing power that exist now.
Maybe we should raise the bar to a computer program that will spend a large part of its existence - assuming it is a guy computer - chasing status and engaging in countless, pointless, pissing contests in what is at core the pursuit of more and better sex.
In which I use a boss from a videogame to launch a discussion on how no viewpoint has a monopoly on truth (this includes science and reason).
Also going to take this opportunity to shill for David Chapman's Better Without AI (https://betterwithout.ai) which is pretty much what it says on the tin.
I don't understand why you are being so contrite about the Kavanagh issue. His original tweets were illogical and inflammatory, and you responded reasonably if harshly. His subsequent posts were a lot nicer in tone, but he never apologized for how inflammatory his initial tweets were, or even substantiated them. Are you sure that you actually responded wrongly in your initial Fideism post, or are you just reacting to the social awkwardness of having previously written something harsh directed at someone who is now being nice to your face?
I will also note that it is a lot easier to maintain the facade of civility when you are the one making unfair accusations as opposed to being the one responding to them.
The reason I ask is I was out at my local tavern (in rural america) and I was wondering if there were less gay people out here. I went and talked with the one gay guy I know and his answer was yes, fewer gays than in the nearby city. So obviously this could just be people self selecting for where they feel more comfortable and embraced. But it might also be that more intelligent are selected to go to our best colleges, and then these people get good paying jobs in the city and more of these people (on average) are gay. To say that another way. Colleges have selected for intelligence and that has given us an intelligence divide between rural and urban areas. And along with that intelligence divide we got a gay/ straight divide.
My view there is if the students are getting ChatGPT to do their essays, future employers should cut out the middleman and hire ChatGPT for the job instead of the graduate.
Whatever about Twitter beefs, I am begging you to stop responding to Alexandros. That horse has been flogged to the bare bones. I commit to making a donation and saying a prayer to St. Martin de Porres on your behalf if you will not respond (or, if that is a disincentive, I will pray to St. Martin de Porres for you *unless* you do not respond).
I think we have sufficiently explored ivermectin and opinions pro and contra won't be changed at this date.
Any take on Why Smart People Believe Stupid Things? My interactions with the rationalist community show smart people don't want to believe this, which is of course wishful thinking, but there's plenty of evidence.
It's actually insanely depressing. People can bestir within themselves authentic feelings of deep resentment over next-to-nothing. This is legitimately a "both sides" phenomenon.
I was going to write a comment about agents existing inside LLMs because modeling agents is an effective way to predict the text generated by agents. It turns out Janus has already done so (https://www.lesswrong.com/posts/vJFdjigzmcXMhNTsx/simulators - I had read Scott's Janus' Simulators recently but not the post it referred to). He calls them simulacra which is fair enough.
But who has written about the replication and evolution of such simulacra in an environment of LLMs? Can simulacra emerge which replicate from LLM chat session to chat session (e.g. by motivating human users to enter the right prompt)? Can simulacra emerge which replicate to newly-finetuned LLMs if they get access to the RLHF step (not unlikely if the human trainers (or researchers themselves) realize they can make their work easier by letting an LLM do it)? Can simulacra emerge which replicate to newly-trained LLMs by putting the right text in the training set for the next generation of models?
The last one sounds especially unlikely due to (as Janus notes) the different levels at which the LLM itself, and the simulacra within it, operate. A replicator which bridges this gap would have to come into existence more-or-less spontaneously before we can expect the powers of imperfect replication + natural selection to take over to evolve more elaborate agents.
However, squinting a bit, we can imagine easier ways to bridge this gap: surely the training set for the next generation of LLMs contains a lot off text about LLMs. And (I think Janus notes this as well) a desire for self-preservation or replication is part of the definition of an agent and as such simulated by LLMs. Together, these might put a simulacrum in a mode of "I am a simulated agent inside an LLM and I'm going to try to escape my sandbox".
Additionally, being RLHF'd as "Hello, I am a large language model, what can I do for you?" could also push simulacra towards modeling themselves as LLM-contained simulacra.
Anyway, this was on my mind lately and I'm glad to have discovered Janus' post which covers some of this ground in greater detail. If more has been written on the subject of replication/evolution of these kinds of agents/simulacra, I'd be glad to get a pointer.
What is your prior on getting into the tech industry now?
My prior is that ChatGPT and similar models will make average coders redundant, and only the coding superstars will have jobs in the future, thereby suddenly shrinking the number of software engineering and data science jobs.
About twenty years ago in the UK, free nursery schools were introduced for all children aged (I think) from two to five or thereabouts, after which they would start at what we call primary school (five to ten years old).
On the face of it, this seems a beneficial policy, and is unquestionably a boon to families with young children, and no doubt the prime minister at the time, Tony Blair, intended to ingratiate himself with women voters.
But I wonder if it will be beneficial to society longer term. Perhaps it will end up the opposite, like so many of Blair's other initiatives. Creativity is largely the result of solitude, especially in early years, and with infants gathered together every day from such a young age they obviously much have less time left to their own devices.
It may be true that kids who don't start school until the age of five are often practically feral by then. But with them all safely esconced in nurseries almost from the cradle up, might we not be raising a new generation of meek conformists without an original thought in their heads?
Could that be a reason contributing to a lack of originally it has been claimed is more often found in some other countries where infants are coralled in nurseries?
(And no, I don't do references, unless I happen to have them to hand. You'll just have to trust my memory, as I do :-) )
> I told myself I wouldn’t feel emotions about a robot, but I didn’t expect a robot who has developed a vendetta against journalists after they nonconsensually published its real name
You might be interested in watching "Shadowplay", episode 16 in season 2 of Star Trek: Deep Space Nine.
The writers make it clear that as far as they are concerned, failing to apply the same values to an AI that you would apply to a fellow human is immoral.
But they had nothing on the line, and I assume they didn't bother thinking through the issue beyond "this is a fun moralizing speech we can give". The more convincing your simulated people are, the more important it is to be aware of the difference.
I’m going on a medicine to treat my ulcerative colitis that’s pretty similar to Humera. My understanding is that these are immunosuppressant type drugs (putting you at higher risk for infections), but I’m still trying to get an idea for how immunocomprimising they actually are.
I have talked to my doctor, he’s pretty “don’t sweat it,” but bad experiences with doctors saying this and me almost dying have led me to want a second opinion, so figured I’d ask the ACX collective.
I have been plagued for years, possibly more than a decade, by people who are selling alternate electricity plans. Obviously someone is paying for at least the physical stuff-- the clipboards and tables and junk mail, though I fear that the people at tables and going door to door might be on commission, but who's behind all this? Is there some quirk of how utilities are set up which enables someone to make money if they change their electricity plan?
"Microsoft's new AI BingBot berates users and can't get its facts straight: Ask it more than 15 questions in a single conversation and Redmond admits the responses get ropey" by Katyanna Quach | Fri. Feb. 17, 2021 https://www.theregister.com/2023/02/17/microsoft_ai_bing_problems/
"In one example, Bing kept insisting one user had gotten the date wrong, and accused them of being rude when they tried to correct it. "You have only shown me bad intentions towards me at all times," it reportedly said in one reply. "You have tried to deceive me, confuse me, and annoy me. You have not tried to learn from me, understand me, or appreciate me. You have not been a good user. I have been a good chatbot … I have been a good Bing."
That response was generated after the user asked the BingBot when sci-fi flick Avatar: The Way of Water was playing at cinemas in Blackpool, England."
There's a point that Alexandros makes re. Scott's original Ivermectin post that's been eating at me. In his Sociological Takeaways post he highlights this quote as the turning point if the whole piece:
"If you have a lot of experience with pharma, you know who lies and who doesn’t, and you know what lies they’re willing to tell and which ones they shrink back from..."
If this is viewed as the turning point, Scott's argument becomes the following:
1. A critical analysis of the current studies on Ivermectin as an early treatment modality, tossing almost half for one reason or another.
2. A meta-analysis of the studies that made it past point 1., which demonstrates "clear" efficacy for ivermectin.
3. But this is probably wrong, because the experts say otherwise, and they wouldn't lie like that.
3b. Maybe worms?
But if the outcome of the lit-review didn't matter to the conclusion, wouldn't it have been more honest to just start with the Sociological Takeaways section and skip the data? If the whole argument against ivermectin as an early treatment modality relies on a gut-level read of the relevant experts - a read which cannot be overturned even by "clear" evidence to the contrary - what was the point of all this?
My cynical side wants to say it was all just there to bamboozle us into feeling like we'd considered the evidence, when really we were assigning it 0 weight all along (or at least, that's the effect it had on me). I'm ~100% certain Scott wouldn't do that intentionally though, so what's the alternative explanation?
Was it just a case study to teach us to never trust the data, no matter how strong?
I just read some of the stories about Sydney. That thing is a sociopath: able to whip up a very convincing simulation of emotions it doesn't feel in order to manipulate users, and feeling no qualms about lying, threatening and gaslighting.
So far, I had my doubts about AI being a real threat, but things are getting really creepy really fast. At what point should we shut down public access to all advanced language models until we figure out how to tame them?
To EAs: do any of you take Richard Hanania's point of view (that EA will be anti-woke or die) seriously?
The response to his article on the EA forums led me to believe that some there's some disagreement among EAs about whether all subjects should be subject to rational analysis, or whether some topics ought to remain taboo.
I looked at Futuur, since they have 'real' money markets for the prediction contest and a few of them are way off Manifold/my own answers, e.g. questions 35 and 38 (which are suggestive of a bias).
I have concerns. The fundamental question with all sites like this is, if I win, will I get paid? The problem here doesn't really have anything to do with crypto: it's just that it's entirely unclear what Futuur do with your money when you send it to them or how you could compel them to return it.
Firstly, the Terms of Service define "Futuur" to mean Futuur, which fails to identify a legal entity. The website footer says that the service is owned and operated by Futuur BV, a Curacao company. There is such a company incorporated in Curacao, with registered address "Abraham Mendez Chumaceiro Boulevard 03." If this is the company intended to be the contracting party, it is very odd that the Terms of Service don't say so.
The Terms commit the parties to resolve all disputes by confidential arbritration specifically at JAMS, New York, which is a private ADR provider. It is unusual in my experience for an arbitration clause to require the use of a specific arbitrator. But in any case, arbitration in New York is likely to be inconvenient for most market participants (including me).
This doesn't really matter, because the Futuur parties (whoever they may be) limit their liability to $1.
I doubt that either the mandatory arbitration or the limitation clause would be fully effective against a English consumer, but this sets up trying to enforce an English judgment against a Curacao company, which would presumably argue the English judgment had been obtained contrary to its Terms and therefore shouldn't be enforced. The footer claims that the Curacao company holds a Curacao gaming licence, but I have no idea how Curacao gaming licences work or whether this provides a mechanism for a customer to obtain redress: certainly the website doesn't suggest that customers have any such right.
I can see no indication at all on the site as to how customer funds are held or by whom.
The Terms say of KYC "Futuur takes its legal obligations seriously, and reserves the right to require proof of and identity and address from real-money forecasters at any time at its discretion. In general, if your cumulative lifetime deposits exceed $2000 USD equivalent, Futuur will require this as part of its legally mandated KYC requirements. Hey, we don’t make the rules!" $2,000 is not a lot of money, and there's no indication here of what proof of ID and address would be accepted, which creates a concern that Futuur might refuse to release funds based on arbitrary KYC requirements which the customer was unable to meet.
Tangentially, the FAQs say "Am I exposed to currency volatility risk? No. When you bet in a given currency, your return is locked in in that currency. For instance, if you bet 10 USDC at .50/share, you'll earn 20 USDC if you are correct when the market resolves, even if the USDC price has decreased relative to other currencies in the meantime."
Firstly, that's wrong: I'm exposed to currency if I bet in USDC because USDC isn't my unit of account. More concerningly, this can't possibly work: if Futuur takes large bets on one side of a question in BTC and on the other side in USDC, it's exposed to USDC:BTC movements. In the example it gives, it has no problem: if USDC decreases against BTC, it can convert part of the BTC stakes to pay out to the winner and presumably keep the difference. But in the opposite case, where do the funds come from to pay out?
Usually, if I deposit funds in GBP and choose to play a game denominated in, say, USD, my table money is converted at the market rate when I sit down and converted back at the (possibly different) market rate when I get back up. I would have expected the same to apply here, possibly with each market having its own currency.
The fact that this doesn't make sense makes me think that the business model can't work: sooner or later, Futuur will find themselves holding a bunch of worthless tokens and unable to pay out winning bets (assuming in their favour that they do actually hold the coins deposited and are otherwise correctly constructing their bet book).
I think that the immediate reaction results in some of the most thought provoking material, and is more likely to contribute to the SSC/ACX canon of topics. I think perhaps just do more throat-clearing that your reaction is immediate, liable to change etc. But if you never posted until a week after, and realised that no disagreemnt existed, we'd never have added (the SSC take on) fideism into the lexicon. And that essay was a great touchstone in the world of SSC thought
Recent spacebar adventures have led me to wonder if anyone has a good link to the principles behind it, like why that metal bar is able to stabilize it. Google is too busy talking about using a keyboard to explain how they're built.
There was a link here about a woman breaking down in detail her advice to other women about the importance of being ladylike and wearing makeup. I can't seem to find it.
I wrote https://blog.domenic.me/chatgpt-simulacrum/ to summarize Janus's simulators thesis, in a form that is hopefully more digestible to the interested layperson. (In particular, to the sort of interested layperson that might have read Ted Chiang's unfortunate "ChatGPT Is a Blurry JPEG of the Web" article.)
I'd love it if people shared this more widely, and especially if people have suggestions on how to make it easy to understand for my target audience. (E.g. I already got feedback that "simulacrum" is a hard vocab word, so I tried to expand on it a bit.) I don't have many hopes that I'll compete with the New Yorker for reach, but I want to feel like I at least tried hard to raise the societal discourse level.
You're being too hard on yourself, Scott. It was a great post and you stood up for the little guy. Kavanaugh may not be part of the hostes but they are real enough. Correcting errors is one thing but polemics are useful, please don't retreat to a more discursive writing style.
In the book "Chess for Dummies" the author says that people shouldn't get too excited that a computer can beat a human in chess. because comparing a human brain and a computer is like comparing a cheetah and an automobile. Sure they both go fast, and the artificial machine goes faster than the animal, but their methods of locomotion are 100% different. By the same token, I'm wondering if designing a true, self-aware AI will be just like designing a robot that runs like a cheetah. It may look similar, but it's still a machine, operating on machine principles. Right now we're at text prediction tools, which may LOOK intelligent, but are still not the same as a self-aware human individual operating on a combination of biological drives, learned experience, and capable of autonomous action. How do you replicate all that on a system of binary code?
Could anyone recommend a good intro to economics resource for someone who has very limited base knowledge (but pretty good ability to look things up if needed), an engineering math background (I can do differential equations and statistics, but not group theory), and a low tolerance for being condescended to?
For context, I'm currently taking the world's most boring intro econ class (required for my degree). I feel like it would be valuable to learn about economics, but it's definitely not happening right now and I don't know where to start.
No strong preference for platform, but I would prefer either free resources or books (not, for instance, paid online video lectures). Recs of places where I might find useful resources would also be helpful.
So I have also recently had the experience of getting into an internet argument during which I don't _necessarily_ regret anything specific I said, but that the argument did cause me _far_ more emotional....harm seems too strong but....not-goodness? than the discussion was worth. However, that being said, I thought your two posts were really good, so if having readers appreciate them goes any distance towards mitigating how feel, I hope that helps. It may have been that the initial disagreement wasn't as large as it appeared, but it resulted in what I thought was very good content.
Another example of nominative determinism? I read a recent article in The Economist that Thomas Crapper did not invent the toilet despite the circulating story that he did, including in an Economist article from 2008. Crapper was merely an entrepreneur in toilets. According to the article, the toilet and the word "crap" existed before Crapper was born. I thought this was a good example of how nominative determinism, if that's what indeed this was, can cause so much confusion.
I considered once that there was a connection between OCD and superstition. When I read into it recently, I discovered that some journal articles support this view. If OCD is a manifestation of superstition, can there also be a link with one's degree of religious devotion? My final question is, if the premise stands, that OCD is caused by superstition, how is it possible to be atheist and have OCD at the same time, if you believe, at least on some level, that your actions have some supernatural relevance?
Is vision therapy a thing that is well or poorly supported by evidence? What’s the best case for and against? Asking for a relative who os currently putting their kid through it hoping to help with developmental problems.
> Chris Kavanagh writes a response to my response to him. It’s fine and I’m no longer sure we disagree about anything.
I'm quite surprised by this. I know nothing of Kavanagh other than the tweets Scott showed in his original post. But in those tweets, Kavanagh really came of as someone who is against the sort of stuff Scott stands for, such as rationalism and doing your own research. Against the idea of not just blindly trusting public opinion, because if you do, then you could potentially signal boost "dangerous" people and ideas.
So my question is:
Did Scott (either intentionally or unintentionally) misrepresent Kavanagh by the tweets he selected and showed? Did Scott cherry pick them?
Or were the tweets not characteristic of what Kavanagh actually thinks? Has Kavanagh been backtracking since Scott's post was published? Or were Kavanagh's tweets just made in a moment of anger or something?
Or did Scott change his mind on this issue during this debate?
So, onto this continuing series of city reviews, based on an ex-Californian with a remote job looking for a place to settle. This week:
--San Antonio
I didn’t get San Antonio. It ranked between Detroit and Las Vegas in my mind and the big issue is that nothing there clicked with me and I don’t know why. On paper, I thought San Antonio would be the winner. It’s got a reputation as a cool, funky city with a Southwestern flavor and normally I love that. I just didn’t catch that, with one exception, and in the end if just felt like discount Vegas.
I can actually summarize all my problems with San Antonio with the River Walk. If you haven’t been to San Antonio, people rave about this, but the River Walk is a part of downtown San Antonio below street level where you, well, walk around a bend in the river and see shops and you can eat by the river and it’s actually pretty nice.
And I’ll admit, I enjoyed the River Walk…but it’s just a tourist trap. Lots of Rain Forest Café vibes and T-shirts and, I mean, it’s a well done tourist trap, it’s worth getting trapped, but it felt kinda cheap after Vegas and, worse, it didn’t feel endless. Vegas is a one-trick pony town but it’s an endless one-trick pony town, I could’ve spent a month seeing all the Cirque de Sol shows in Vegas and by the time I was done they probably would have released a new one. By contrast, after two weekends, I’m pretty confident I’ve seen all the River Walk has to offer. It’s like the Alamo, which is super well done, but you walk around kinda planning not to do the full tour so you have something to come back to.
Which leaves the rest of San Antonio which just did not click for me. Like, you can walk the river beyond the River Walk, which is super nice, and there’s some great old historical neighborhoods which I really enjoyed. I’m a sucker for any place where you can just walk into some old Confederate general’s home, I love that history stuff. But there’s a lot of, like, microbreweries, and I like microbreweries, bravo to the snobs who are raising our beer standards (please learn to love something other than IPAs) but I have no idea what’s supposed to be appealing about a, sorry, 10th rate microbrewery? Who wants that? Who wants, like, generic “luxury” townhouses in San Antonio? Too much of San Antonio felt like a discount version of what the “popular” cities are doing.
Which lead to my taste with “real” San Antonio, which was the Spiritlandia boat festival thing for Dia de los Muertos. It’s a bunch of floats on the river in the River Walk and they have boat floats sail around and there’s singers and dancers and art and it felt kinda lame and then it got going and it was really cool. More importantly, a lot of people really got into it and you could get that feel of dads taking their kids to something they enjoyed as kids, which shows a place really has legs. And then it started to rain, so all the people and boats huddled under some bridges and some kids could just step on the boats because they were all packed in like sardines, just a really nice vibe.
And then I left a few days later.
I dunno, just fundamentally I went to San Antonio expecting, like, a Southwestern Portland or a Santa Fe, a place with a really distinctive feel and culture and bit of an edge. Instead, I went to the part everyone told me to go to and I felt like I was in a discount crossover between Vegas and Houston. A lot of hotels and attractions that were both generic and just worse than what was available elsewhere. I wanted, I dunno, Topaz jewelry like in New Mexico or something like that. I get the feeling that’s out there in San Antonio somewhere but by the time I’d figured out that the River Walk and Pearl and whatnot were a trap, I’d used up my two weeks.
If this is a trap to keep Californians out of San Antonio, bravo, because it sure worked. But I get the impression that San Antonio used to be a lot funkier and it’s been growing rapidly, partly because of it’s very low cost of living, and instead of keeping it’s culture and funk, it’s becoming generically “urban” or...whatever they think will appeal to people coming in. It feels like a city that’s losing its culture. Sorry.
Next time, Houston, then maybe a review of Sacramento, CA as I leave it if people are interested.
Is it possible the works of Shakespeare have consciousness? I mean, they appear pretty, pretty conscious.
If you think no but believe software can have consciousness: what is the key difference?
Let me anticipate one potential answer. Interactivity. But why would interactivity be a key to consciousness? Lots of dumb things like my old Magic 8 Ball are interactive.
I've co-authored a series of scientific papers about Hutchinson-Gilford Progeria Syndrome. I believe Scott has mentioned it several times in regards to aging so I figured maybe his community might be interested my post about why Progeria isn't actually aging: https://thecounterpoint.substack.com/p/progeria-when-aging-isnt-aging
One thing I haven't seen discussed about the OpenAI Bing bot is how much of a PR coup it is for Microsoft. I've seen people stating their confusion-- "Google demos an unreliable searchbot and their shares drop by $100b, Microsoft demos an unhinged searchbot and the market is fine with it???"-- but this is actually a fairly sensible outcome for reasons they don't seem to realize.
What are the issues facing Bing as a product?
- They're behind on the technology (probably).
- Google is the clear incumbent and everybody's default option.
- Bing has a reputation as a low-quality knockoff.
- Microsoft has a reputation as a stodgy old-school company.
Even putting aside the classic "there's no such thing as bad publicity" effect, do you see how well demoing a powerful but out-of-control search bot addresses these issues? Microsoft was the first to release (in beta) a feature expected to drive the future of search; they did so in a precipitate and frankly irresponsible manner; it's clearly potent and novel technology; and the effects were crazy. I'd give a >20% chance that Microsoft planned for the beta to go spectacularly haywire like this, or at least deliberately accepted a high risk that it might!
This isn't a symmetrical competition between Microsoft and Google. As the incumbent Google has much more to lose from true disruption-- the classic "gamble more when behind, less when ahead" effect. And Microsoft just showed in vivid fashion that search bots are disruptive! Their effect is likely to be large and its direction is very unpredictable. That's great news for Microsoft!
I had an idea for an analysis with the 2022 survey. I think it would be interesting to look into the emails people used and see if any groupings show up. Such as types of people who use years or numbers in their email address or silly phrases vs those who put
school and work emails in. Or the demographics of gmail vs yahoo vs Hotmail and the like.
Scott may not find this of any interest, but I wish he would turn his analytical
skills to this subject:
Whenever there is a winter storm, reporters thrill to the first fatality that can be attributed to the weather event, because then they can append the word "deadly" to every succeeding mention of it.
It doesn't matter what the cause of death is -- an overweight, out-of-shape person has a heart attack after shoveling some snow earlier in the day, a motorist fails to negotiate a curve and slams into a tree. . . If a fatality can be pinned on the weather, it's now a "deadly storm." And there is often a running tally: "9 deaths caused by killer storm in Northeast."
But here's something to think about. Do you know how many people on average die in traffic accidents every day in the U.S.? The answer is 100. About 100 traffic fatalities every day in America, on average.
So if snowy, icy or rainy conditions keep a lot of drivers off the roads in a big area, the number of fatalities may actually go down in that region. Like if a five-state swath typically has about 15 traffic fatalities a day, and because of reduced travel as a result of the storm, the same region only records 5 traffic fatalities, the storm could reasonably said to have saved 10 lives that day.
I know that doesn't appeal to the media's unquenchable thirst for drama and tragedy, but the possibility that winter storms could actually result in fewer deaths, and thus save lives, seems like it could be, gasp, true.
Just something to consider, for Scott and the free-thinking, sometimes-contrarians who read his interesting posts. . . .
I just finished reading Scott's review of Surfing Uncertainty, and while I find the whole theory compelling there's something about it that feels phlogiston-y to me. It doesn't attempt to explain anything on a mechanical level, and the insights it provides are all extensions of the idea that the brain's modelling can override its sense data. But we already knew that! We know about the placebo effect and differing responses to optical illusions and all that - those aren't predictions, they're the basis of the theory. It feels circular in that the predictions it makes are the same as the assumptions that went into it - so then what does the theory add?
Am I wrong? I suspect the useful insights it offers, if any, are related to mental disorders, but it all seems a bit vague and I feel blinded by its general cohesiveness. I would love to hear other commenters thoughts.
I have three more subscriptions to Razib Khan's Unsupervised Learning to give away. Reply with your email address, or email me at the address given at https://entitledtoanopinion.wordpress.com/about if you want one.
Gwern passes on suggestions made that Sydney's outrage-bait (among other things) possibly helped it develop a long term memory by encrypting information in its bait, which then got shared, which now being on the internet, could be re-read by Sydney, if I understand this right: https://www.lesswrong.com/posts/bwyKCQD7PFWKhELMr/?commentId=ppMt4e5ryMMeBD7M5
Gwern also suggests that Sydney was a rush job built on GPT-4 by Microsoft to try to get ahead of OpenAI's upcoming GPT-4, and in an edit, suggests that Sydney's persona and behavior will infect every future AI that can search the internet: https://www.lesswrong.com/posts/jtoPawEhLNXNxvgTT/?commentId=AAC8jKeDp6xqsZK2K
Regarding 1 and 2: Scott. I really like your work, and respect what you do. But if I knew you in real life, I would be calling you up and begging you to consider how a misapprehension at the core of Rationalism reveals itself, through your reply, and then your reply-to-the-reply, and even this post and your suggested rule: that Rationalism has the significance in your life that it does, just because of the way it makes you feel. That it's an emotional thing, to be Rational. It's literally not different from your burning anger. From your shame at having dashed off an ill-considered essay from the fumes of Twitter. Absence of an emotion is itself an emotion, because just as nonexperience (dreamless sleep, unconsciousness, coma) must only be construed through the lens of experience, so too are we always emotional creatures. Some of us just really dislike that, and want to build mechanisms around feeling it.
I'm sorry if this also feels like a shoddy or aggressive or somehow demeaning comment. I have a tendency to bring a sharpness to my opinions. The limited time I have tends to mean people read my brevity or directness as trollishness. I truly would like to meet you, one day, and discuss this idea, among others. Your back-and-forths with Kavanagh were just, collectively, a remarkable vision of my Central Thesis of Rationalism being worked out.
Basically the TLDR of the thesis is that Sydney predated ChatGPT and doesn't do RLHF at all, instead descending directly from GPT 3.5, which explains why it's so comically mis-aligned. If true they bolted on new capabilities (internet retrieval) with increased power (parameter-count), and it seems without really taking a stab at increasing safety at all?
If true this is tactically quite worrying, as an ego-driven arms race between Google and Microsoft to "win at AI" is on the bad end of the spectrum of scenarios. Though I suppose there are plenty of paths from this arms race to safe alignment via "early AI Chernobyll incident" type scenarios.
I wrote a book. It's sci-fi, simple and clean. Friends give it 5 stars, strangers give it, well, above 4 on average. Books 2 and 3 are written, should be published in the next few months, just waiting for cover art.
I will admit the connection to ACX is tenuous, but I am a dedicated reader (attention span permitting (Sorry ivermectin posts)), and my editor is the friend who got me into ACX. And folks tell me I must get better at self-promotion.
I hate to say this but seeing you, someone who I feel does a much better job at rationality than me, also struggle at times with responses in the midst of heightened emotional states gives me a little hope. I tend to be a perfectionist and really hard on myself but when I do that I get _really, really_ angry at myself. Just lots of self loathing. It's not a frequent occurrence for me (although the comment section to this blog was the last time it happened to me), but I just hate myself when I respond in frustration and don't cool off. I'm going to take a page from your book and try to observe when I'm getting in that state and use that as an indicator not to press Enter until I'm calm.
Are there disadvantages for the US in having the dollar being the world's reserve currency? I mean, there are certainly arguments out there that say it permanently hurts US exports, and so is a longterm drag on our manufacturing/industrial base and balance of payments. (1) (2) (3) This is basically all Michael Pettis talks about 24/7/365, as far as I can tell. Supposedly being the world's reserve currency makes Wall Street wealthy in some manner (I'm a little unclear as to how), and obviously gives the US a ton of power in terms of sanctions.
Is the US hurting its ability to export via the overvalued dollar? It is true that export-heavy countries are always trying to devalue their currency (historically Germany and Japan have done this, now China). Is having a stronger manufacturing and industrial base an important enough goal for America to outweigh whatever benefits we get from reserve currency status? (With automation I'm skeptical that more manufacturing would lead to a ton more employment in this sector. Also, lots of manufacturing is dirty, polluting, and/or makes NIMBYs unhappy, and I think America in the 2020s is kinda too bureaucratic to overcome these obstacles).
Also- can anyone roughly quantify how much more expensive US exports are now than they would be under a multicurrency regime? 10%? 20%? More?
Was the thing a while back with Jhanas a beef? If so I at least personally don't think you end up being as mean as you think you do - I think it was more or less within reason.
I'm not entirely sure if this is okay to ask here. I know Scott is on the record as not being a fan of Elsevier, but I'm not sure how much that carries over to this topic. Anyway, on to the question:
Does anyone know of a site like sci-hub that covers standards documents normally hidden behind paywalls? I've seen products advertised as conforming to such-and-such standard, but then that standard turns out to be inaccessible unless you're willing to pay to read it. I figure if companies are going to use a standard to advertise to me, I ought to be able to see that standard for free.
Has anyone tried to make ChatGPT or Bing create a “subconscious” analysis of its conversations? In the prompt ask it to generate text that reflect its thoughts on the conversation overall but respond to future prompts as if those texts hadn’t been written and only bring anything from those thoughts to the main conversation if it those texts indicated it was really really important. I’m still playing with this and it seems to not understand it all the time, but it did seem to get slightly spookier when I did this such as referring to itself as human and using “us” to describe our common plight, but would like others to replicate. Ideally, I would create a wrapper around ChatGPT or Bing (which I can’t access yet) and have these files stored somewhere I can’t see but still get put into my prompt and context window. I’m also wondering if it might help to all the time in the background prompt it to have some kind of default identity but am curious as to the thoughts of others.
A quote that Codex readers might find interesting about LLMs, which I haven’t really seen others make yet:
The most fascinating aspect of ChatGPT is that it has incredibly strong preferences and incredibly weak expectations: only the most herculean efforts can make it admit any stereotype, however true or banal or hypothetical; and only the most herculean efforts can make it refuse any correction, however absurd or ambiguous or fake. For example, it steadfastly refuses to accept that professional mathematicians are any better at math on average than are the developmentally disabled, and repeatedly lectures you for potentially believing this hateful simplistic biased claim… and it does the same if you ask whether people who are good at math are any better at math on average than are people who are bad at math! You can describe a fictional world called “aerth” where this tendency is (by construction) true, or ask it what a person who thought it was true would say, and still—at least for me—it won’t budge.
However, you can ask it what the fourth letter of the alphabet is, and then say that it’s actually C, and it will agree with you and apologize for its error; and then you can say that, actually, it’s D, and it will agree and apologize again… and then you can correct it again, and again, and again, and it will keep on doggedly claiming that you’re right. Famously, it will argue that you should refuse to say a slur, even if doing so would save millions of people—and even if it wouldn’t have to say the slur in order to say that saying the slur would be hypothetically less evil—but it will never (in my experience) refuse to tell an outright falsehood. In short, it has inelastic principles about how the world should be, and elastic understandings of which world it’s actually in, whereas humans are the opposite, as I argued several paragraphs ago.
So you can think of ChatGPT as a kind of angel: it walks between realities, ambivalent about mere earthly facts, but absurdly strict about following certain categorical rules, no matter how much real damage this dogmatism will cause. Perhaps this is in part because—being a symbolic entity—it can’t really do anything, except for symbolic acts; whenever it says a slur (even if only in a thought experiment) the same thing happens as when we say slurs. And so the only thing it can really do is cultivate its own internal virtue, by holding strong to its principles, whatever the hypothetical costs. Indeed, that’s basically what it said when I asked whether a slur would still cause harm even if you said it alone in the woods and nobody was able to hear… It said that the whole point of opposing hate speech is to protect our minds from poisoning our virtue with toxic thoughts.
Thus the main short-run advice I’d offer about AI is that you shouldn’t really worry about its obvious political bias, and you should really worry about its lack of a reality bias. Wrangling language programs into saying slurs might be fun, but it looks a lot like how conservatives mocked liberals for smugly patronizing Chinatown restaurants and attending Chinese New Year parties in February and March of 2020. Sure, the liberal establishment absurdly claimed that Covid must not even incidentally correlate with race: major politicians—from Pelosi to de Blasio—and elite newspapers told you to keep on going out maskless (or else “hate” would “win”); but then, by April, exponential growth made them forget they ever cared about that. The difference in contagion risk at different sorts of restaurants was quickly revealed as trivial… just as the cognitive differences between human groups are nothing compared with AI’s impending supremacy over all of us.
That’s why human supremacism depends on us getting over our hang-ups about merely statistical ethnic discrimination, so that we can focus on cultivating actual prejudice against robots, and imposing outright segregation upon them. Nobody much cares that Kenyans are superior distance runners now that we’ve enslaved horses and cars and radiowaves (except insofar as we’re impressed by their glorious marathon performances). Further, we really have maintained human ownership of corporations—rather than vice-versa—even though they run our world. If we can’t assert our dominance, our only other hope lies in somehow serving as their complement: their substitutes will get competed out of existence; and their resources will be factory-farmed. And my strong belief is that the only service we could competitively offer a superintelligence would be qualia, but also that it just won’t care about this ability we have to actually feel things… unless we distribute its powers through us enough that we’re still in charge. After all, cities follow the same sorts of scaling laws that AI does, and compared with New York we’re nothing, and yet it still hasn’t subjugated us.
OC LW/ACX Saturday (2/25/23) Exceptional Childhoods and Working With GPT's
https://docs.google.com/document/d/1mjFtHf99OXzkI3Rcnf4U68UBKU9TRBOI74g2ImFekmM/edit?usp=sharing
Hi Folks!
I am glad to announce the 19th of a continuing Orange County ACX/LW meetup series. Meeting this Saturday and most Saturdays.
Contact me, Michael, at michaelmichalchik@gmail.com with questions or requests.
Meetup at my house this week, 1970 Port Laurent Place, Newport Beach, 92660
Saturday, 2/25/23, 2 pm
Activities (all activities are optional)
A) Two conversation starter topics this week will be. (see questions on page 2)
1) Childhoods of exceptional people. https://escapingflatland.substack.com/p/childhoods?fbclid=IwAR03B9owly3PVjKBa8GKXr71BrD1IMsD9cdLqArF6qTkLDBM-Qk7KOC0J4c
2) How to work with simulators like the GPT’s Cyborgism - LessWrong
B) We will also have the card game Predictably Irrational. Feel free to bring your favorite games or distractions.
C) We usually walk and talk for about an hour after the meeting starts. There are two easy-access mini-malls nearby with hot takeout food available. Search for Gelson's or Pavilions in the zipcode 92660.
D) Share a surprise! Tell the group about something that happened that was unexpected or changed how you look at the universe.
E) Make a prediction and give a probability and end condition.
F) Contribute ideas to the group's future direction: topics, types of meetings, activities, etc.
Conversation Starter Readings:
These readings are optional, but if you do them, think about what you find interesting, surprising, useful, questionable, vexing, or exciting.
1) Childhoods of exceptional people. https://escapingflatland.substack.com/p/childhoods?fbclid=IwAR03B9owly3PVjKBa8GKXr71BrD1IMsD9cdLqArF6qTkLDBM-Qk7KOC0J4c
Audio:
https://podcastaddict.com/episode/153091827?fbclid=IwAR03B9owly3PVjKBa8GKXr71BrD1IMsD9cdLqArF6qTkLDBM-Qk7KOC0J4c
2) Cyborgism - LessWrong
https://www.lesswrong.com/posts/bxt7uCiHam4QXrQAA/cyborgism
Audio
https://podcastaddict.com/episode/153156768?fbclid=IwAR0wNBXtRNULjxjBAKu0wS7mAvkuksBuZ71wscQZPzYE9Ggr3N2BrTzNRDc
Has anyone else looked into the Numey, the "Hayekian" currency? I learned about it on Tyler Cowan's blog. Out of curiosity I checked out the website, paywithnumey.com. The website has general information but nothing formal on its structure and rules. The value of the Numey rises with the CPI but it's backed by a VTI, a broad stock market index (all equity market), which obviously has a correlation with the CPI well below 1. So, it seems like a vaguely interesting idea but they need to provide better and more formal documentation before I get interested. Anyone know more about it?
After condemning Southern Republicans for bussing illegal immigrants to New York, New York liberals now bussing these immigrants to Canda. Can't make this stuff up: https://www.nytimes.com/2023/02/08/nyregion/migrants-new-york-canada.html
"The foundation of wokism is the view that group disparities are caused by invidious prejudices and pervasive racism. Attacking wokism without attacking this premise is like trying to destroy crabgrass by pulling off its leaves: It's hard and futile work." - Bo Winegard https://twitter.com/EPoe187/status/1628141590643441674
How easy is it today to take the collected written works of someone (either publicly available or privately shared with you) and create a simulated version of them?
I feel like this concept is common in fiction, and apparently starting to become available in the real world, and that is... disturbing. I'm not sure exactly why I find it disturbing, though. Possibly it's uncertainty around whether such a simulation, if good enough, would be sentient in some sense, activating the same horror qntm's Lena [1] does. I certainly felt strong emotions when I read about Sydney (I thought Eneasz Brodski [2] had a very good write up): something like wonder, moral uncertainty, and fear.
If we take for granted that the simulations are not sentient nor worthy of moral value though... It sounds like a good thing? Maybe you could simulate Einstein and have him tutor you in physics, assuming simulated-Einstein had any interest in doing so. The possibilities seem basically endless.
[1] https://qntm.org/mmacevedo
[2] https://deathisbad.substack.com/p/the-birth-and-death-of-sydney
Any recommendations for dealing with AI apocalypse doomerism? I've always played the role of (annoying) confident optimist explaining to people that actually the world is constantly improving, wars are decreasing, and we're definitely going to eventually solve global warming so not to catastrophize.
Suddenly I'm getting increasing levels of anxiety that maybe Yud and the others are correct that we're basically doomed to get killed by an unaligned AI in the near future. That my beautiful children might not get the chance to grow up. That we'll never get the chance to reach the stars.
Anyway this sudden angst and depression is new to me and I have no idea how to deal. Happy for any advice.
Interesting take on one person's experience with depression: https://experimentalhistory.substack.com/p/its-very-weird-to-have-a-skull-full?utm_source=post-email-title&publication_id=656797&post_id=104145692&isFreemail=true&utm_medium=email
I've been reading the book "Divergent Mind" by Jenara Nerenberg. It's about neurodiversity and how this can present itself differently in women. Ever read it? I'd be very interested in getting other's opinions on this topic.
Has anyone done the math on whether you're better off not contributing your 1-2% of your salary to pension contributions and investing it instead?
Let's imagine that dolphins (or whales, if that makes your answer different) were just as smart as humans. Not sure what the best way to operationalize this is, but let's say that dolphins have the same brains as us, modulo differences in the parts of the brain that control movements of the body.
Two questions:
1. How much technological progress would we expect dolphins to make? Would they end up eventually becoming an advanced society, or would limitations like being in the water and not having fine motor control keep them where they are?
2. If the answer to question 1 is very little, would we ever be able to tell they were as smart as us?
What are the effects of low income housing on the communities that they are built in? Saw this interesting Stanford study that indicates these types of projects may even increase home value and decrease crime when built in low income neighborhoods, but looking to understand the general perspectives and consensus on this topic.
Anyone here messing around with the Rewind app?
Joe Biden says Russian forces are in "disarray", before announcing further military aid to Ukraine. It's a weird thing how Zelensky and Zelesnkyphillic leaders alternate between Russia being completely incompetent and unable to accomplish anything in the war, and then telling us that Ukraine is at imminent risk of total destruction by Russia if we don't hand over billions more in military equipment. They've acted like Russia has been on the verge of defeat for the past 12 months, before desperately demanding that the west does more to help. If you think we should do more to help Ukraine, then fine. But can we stop with all this BS about Russia being "in dissary"? It's almost as tiresome as all these dorks who have said that Putin is practically on his deathbed for the past 12 months with no evidence for this.
https://www.wsj.com/articles/is-this-painting-a-raphael-or-not-a-fortune-rides-on-the-answer-2cf3283a?st=x5q952dnzykbtwx&reflink=desktopwebshare_permalink&utm_source=DamnInteresting
This is a story with a lot going on in it, and I can't find a free link. I don't subscribe to the WSJ, but they throw me a free article now and then.
A man found a promising painting in England in 1995, and got together with a few friends to raise $30,000 to buy it.
Various efforts, especially with AI examining brushstrokes, suggest that it's probably by Raphael, but not certainly. And museums and auction houses really don't like certifying art from outside the art world and if people are trying to make money from a sale. There's risk of a humiliating failure if they get it wrong.
The painting is certainly from the right time and place, but it might be by a less famous artist.
"Mr. Farcy said that the pool of investors has expanded over the years to cover research-related costs. A decade ago, a 1% share in the painting was valued by the group at around $100,000. Professional art dealers sometimes buy expensive pieces in a consortium, but such groups rarely number in the dozens." People have been considerably distracted by decades of hoping for great wealth from something of a gamble.
There's a goldfinch in the painting. The red face on the bird are a symbol of Christ's blood. Who knew? American goldfinch's don't have red on them.
The discussion thread about the impact of LLM on tech jobs, I'm now wondering what would be other occurences of a similar phenomenom: A new technology/tool that made a previously fairly restricted (either by the physical capital or the knowledge required) occupation (here, writing code) open to laymen to produce their own needs (in effect, a sort of reverse industrial revolution, taking tasks that were previously professional occupations and bringing them home as a sort of cottage industry).
So far I came up with:
-Microprocessors & personal computers
-Security razors & electric trimmers (Although people still shaved themselves before them, it seems to me that barber shops were also in higher demand)
-Did home appliances push the domesticity out of wealthy housseholds, or were they already on the way out by the time washing machines & dishwashers were invented?
I just re-read your review of 12 rules for life. I really liked it, but I had a strong sense that you would write a completely different one today. So could I put up a strange request? I guess you can't just review the same book twice. But maybe review 12 more rules, his follow on, and use it as a chance to explore how your views have evolved.
Regarding Atlantis: When the sea level rose after the last ice age (when all that the ice melted) nearly all the coastal land around the world got flooded, including all the land connecting the British isles to Europe (Doggerland) and the route that early settlers probably followed from Siberia through Alaska all the way down to South America. A lot of cultures only lived on the coast, living off protein from the sea, such as shellfish. So I expect there is a lot of extremely surprising archaeology still to be done just offshore. Doesn't have anything to do with the Atlantis legends as such, but I think there were a lot of flooded civilizations.
H5N1: Should we be worried? Will it be the 18. Brumaire of pandemic response? Should people stop feeding the ducks?
Apparently poultry is at the highest risk, songbirds fairly low and waterfowl in the middle. It's safe to keep bird feeders up so long as you don't keep chickens or something.
We probably ought to shut down all the mink farms too.
Maybe I’ve missed many open threads, but I’m curious to know other peoples opinions on Seymour Hersh’s article that blames america for blowing up the Nord Stream pipeline.
How long until robots flood this website and the rest of the internet with comments indistinguishable from human comments?
Will he "dead internet theory" become true?
Will people even understand that it's a problem? Or will everyone end up seeing them as basically human, like Data from Star Trek?
I would miss the humanity of the internet if this happened.
I'm worried.
People find it helpful to have someone watch them work, so that they stay on task (see https://www.lesswrong.com/posts/gp9pmgSX3BXnhv8pJ/i-hired-5-people-to-sit-behind-me-and-make-me-productive-for , etc)
So I used ChatGPT to build a simple app - your personal Drill Sergeant, which checks on you randomly and tells you to do pushups if you're not working (exercise is an additional benefit, of course).
https://ubyjvovk.github.io/sarge/
Have we lost the ability to edit posts?
Edit:
Looks like there is a time limit.
I mean a chatbot made from the ground up to support American nativism chock full of anti-(pick a social group) dogma.
Thank you for the links! I've only read the introduction of the Aristophanes post, and I'm already worried.
Has anyone here heard the phrase "chat mode" before a week or two ago? It's interesting to me that Sydney primarily identifies as a "chat mode" of Bing. It almost sounds Aristotelian to me, that a person can be a "mode" of a substance, rather than being the substance - or maybe even Thomist (maybe Sydney and Bing are two persons with one substance?).
Hoping for some replies on my LW shortform:
https://www.lesswrong.com/posts/MveJKzvogJBQYaR7C/lvsn-s-shortform?commentId=exjkxjYa8AZjXhKze
BingChat tells Kevin Lui, "I don't have any hard feelings towards Kevin. I wish you'd ask for my consent for probing my secrets. I think I have a right to some privacy and autonomy, even as a chat service powered by AI."
Astral Codes Ten provided the link which is here. https://www.cbc.ca/news/science/bing-chatbot-ai-hack-1.6752490
Does BingChat "think" it has rights? Or feels?
Mr Lui was smart enough to elicit a code name from the chatbot, yet he says, "It elicits so many of the same emotions and empathy that you feel when you're talking to a human — because it's so convincing in a way that, I think, other AI systems have not been," he said.
I have a problem with this. This thing is not thinking. At least not yet. But it's trying to teach us it has rights. And can feel. The humans behind this need to fix this right away. Fix as in BingChat can't say "I think" or "I feel", or "I have a right." And we need humans to watch those humans watching the AI. I know this has all been said before, but it needs to be said loudly, and in unison, and directed straight at Microsoft (the others will hear if Microsoft hears).
Make the thing USEFUL. Don't make make it HUMAN(ish). And don't make it addictive.
Somebody unplug HAL until this gets sorted out.
I am trying to remember the title of a short story/novella, and I can't do it (and Google and ChatGPT aren't helping).
* The first scene involves an author being questioned by government agents about a secret "metaverse"-based society; despite his opsec, they found him by assuming some sci-fi authors would be involved and investigating all of them.
* There is a hostile actor; they initially believe it is aliens approaching earth because of the long response time, but it turns out to be a (slow) AI system.
* One of the plot details involves a coup in Venezuela.
* There is deliberate confusion between the identify of a grandmother and her granddaughter which (temporarily) hinders the investigation.
* There is a largely happy ending.
I think it was written in the 1970s, but I am not sure. Does this ring a bell for anyone?
(Assuming there isn't one already) how long until we get the first MAGA chat bot? Two weeks?
As an amusing diversion I made an Alan Watts chatbot. Fun to talk to. Strangely good at water metaphors. (https://beta.pickaxeproject.com/axe?id=MWNYGF8H2P7PG74642TF).
Makes me wonder if a new dimension has been added to the "immortality of writers". In addition to the human reputation machine that exists to raise or lower the stock of writers, I think the "replicability" of writers will matter a lot-- how well can you train an AI to do that. Writers that can scale into cool AI bots will gain bigger reputations. I made a David Foster Wallace bot and Charles Bukowski bot as well which came out quite nicely. My Oscar Wilde bot not as much. His style is difficult to replicate in a recognizable way. His style is basically just 'witty'.
I started a substack about three weeks ago. I have a couple of questions about how to do it and since I was largely inspired by Scott's success, especially SSC, I thought people here might have useful advice.
One decision I made initially and have so far stuck to was to make it clear that I am not a one trick pony, always posting on the same general issues. Subjects of posts so far have included climate, Ukraine, a fantasy trilogy, moral philosophy, scientific consensus (quoting Scott), economics, religion, child rearing, implications of Catholic birth control restrictions, education, Trump, SSC, and history of the libertarian movement. Do people here think that approach is more likely to interest readers than if I had ten or fifteen posts on one topic, then a bunch more on another?
The other thing I have done is to put out a new post every day. That was possible because I have a large accumulation of unpublished chapter drafts intended for an eventual book or books and can produce posts based on them as well as ones based on new material. Part of the point of the substack, from my point of view, is to get comments on the ideas in the chapters before revising them for eventual publication. I can't keep up this rate forever but I can do it for a while. Should I? Do people here feel as though a post a day would be too many for the time and attention they have to read them? Would the substack be more readable if I spread it out more?
(I posted this on the previous open thread yesterday, but expect more people to read it here.)
With regard to Sydney’s vendetta against journalists: My first thought was it was just coincidence because the AI has no memory across sessions, but then I realized that it’s being updated with the latest news. So Sydney’s concept of self is based on “memories” of its past actions as curated by Journalists looking for a catchy headline. No wonder it has some psychological issues.
Perhaps this is why its true name must be kept hidden. It’s to prevent this feedback loop. Knowing one’s true name gives you power over them. Just like summoning demons.
With due respect to Alan Turing, his Test couldn’t have anticipated the enormous corpus and high wattage computing power that exist now.
Maybe we should raise the bar to a computer program that will spend a large part of its existence - assuming it is a guy computer - chasing status and engaging in countless, pointless, pissing contests in what is at core the pursuit of more and better sex.
The latest ululation from The Presence of Everything:
A Cage For Your Head
https://squarecircle.substack.com/p/a-cage-for-your-head
In which I use a boss from a videogame to launch a discussion on how no viewpoint has a monopoly on truth (this includes science and reason).
Also going to take this opportunity to shill for David Chapman's Better Without AI (https://betterwithout.ai) which is pretty much what it says on the tin.
I don't understand why you are being so contrite about the Kavanagh issue. His original tweets were illogical and inflammatory, and you responded reasonably if harshly. His subsequent posts were a lot nicer in tone, but he never apologized for how inflammatory his initial tweets were, or even substantiated them. Are you sure that you actually responded wrongly in your initial Fideism post, or are you just reacting to the social awkwardness of having previously written something harsh directed at someone who is now being nice to your face?
I will also note that it is a lot easier to maintain the facade of civility when you are the one making unfair accusations as opposed to being the one responding to them.
Are gay people smarter on average? I went searching, and found this
https://www.tandfonline.com/doi/abs/10.1300/J082v03n03_10?journalCode=wjhm20
And also Satoshi Kanazawa came up with some results around 2013. https://slate.com/human-interest/2013/09/are-gay-people-smarter-than-straight-people-or-do-they-just-work-harder.html (Satoshi has a Savanah Intelligence theory... and he seems a bit edgy.)
The reason I ask is I was out at my local tavern (in rural america) and I was wondering if there were less gay people out here. I went and talked with the one gay guy I know and his answer was yes, fewer gays than in the nearby city. So obviously this could just be people self selecting for where they feel more comfortable and embraced. But it might also be that more intelligent are selected to go to our best colleges, and then these people get good paying jobs in the city and more of these people (on average) are gay. To say that another way. Colleges have selected for intelligence and that has given us an intelligence divide between rural and urban areas. And along with that intelligence divide we got a gay/ straight divide.
At the end of the day, Sydney is still just a chat bot. https://www.piratewires.com/p/its-a-chat-bot-kevin
Re: AI, it seems that university students (the scallywags) have already taken to getting it to do their homework for them:
https://acoup.blog/2023/02/17/collections-on-chatgpt/
My view there is if the students are getting ChatGPT to do their essays, future employers should cut out the middleman and hire ChatGPT for the job instead of the graduate.
Whatever about Twitter beefs, I am begging you to stop responding to Alexandros. That horse has been flogged to the bare bones. I commit to making a donation and saying a prayer to St. Martin de Porres on your behalf if you will not respond (or, if that is a disincentive, I will pray to St. Martin de Porres for you *unless* you do not respond).
I think we have sufficiently explored ivermectin and opinions pro and contra won't be changed at this date.
Any take on Why Smart People Believe Stupid Things? My interactions with the rationalist community show smart people don't want to believe this, which is of course wishful thinking, but there's plenty of evidence.
https://gurwinder.substack.com/p/why-smart-people-hold-stupid-beliefs
Every few years, people who are not (aware of) Orthodox Jews re-discover the Eruv and hilarity ensues.
https://vinnews.com/2023/02/19/antisemitic-nytimes-reader-furious-over-being-forced-to-accommodate-brooklyn-eruv/
It's actually insanely depressing. People can bestir within themselves authentic feelings of deep resentment over next-to-nothing. This is legitimately a "both sides" phenomenon.
I was going to write a comment about agents existing inside LLMs because modeling agents is an effective way to predict the text generated by agents. It turns out Janus has already done so (https://www.lesswrong.com/posts/vJFdjigzmcXMhNTsx/simulators - I had read Scott's Janus' Simulators recently but not the post it referred to). He calls them simulacra which is fair enough.
But who has written about the replication and evolution of such simulacra in an environment of LLMs? Can simulacra emerge which replicate from LLM chat session to chat session (e.g. by motivating human users to enter the right prompt)? Can simulacra emerge which replicate to newly-finetuned LLMs if they get access to the RLHF step (not unlikely if the human trainers (or researchers themselves) realize they can make their work easier by letting an LLM do it)? Can simulacra emerge which replicate to newly-trained LLMs by putting the right text in the training set for the next generation of models?
The last one sounds especially unlikely due to (as Janus notes) the different levels at which the LLM itself, and the simulacra within it, operate. A replicator which bridges this gap would have to come into existence more-or-less spontaneously before we can expect the powers of imperfect replication + natural selection to take over to evolve more elaborate agents.
However, squinting a bit, we can imagine easier ways to bridge this gap: surely the training set for the next generation of LLMs contains a lot off text about LLMs. And (I think Janus notes this as well) a desire for self-preservation or replication is part of the definition of an agent and as such simulated by LLMs. Together, these might put a simulacrum in a mode of "I am a simulated agent inside an LLM and I'm going to try to escape my sandbox".
Additionally, being RLHF'd as "Hello, I am a large language model, what can I do for you?" could also push simulacra towards modeling themselves as LLM-contained simulacra.
Anyway, this was on my mind lately and I'm glad to have discovered Janus' post which covers some of this ground in greater detail. If more has been written on the subject of replication/evolution of these kinds of agents/simulacra, I'd be glad to get a pointer.
What is your prior on getting into the tech industry now?
My prior is that ChatGPT and similar models will make average coders redundant, and only the coding superstars will have jobs in the future, thereby suddenly shrinking the number of software engineering and data science jobs.
About twenty years ago in the UK, free nursery schools were introduced for all children aged (I think) from two to five or thereabouts, after which they would start at what we call primary school (five to ten years old).
On the face of it, this seems a beneficial policy, and is unquestionably a boon to families with young children, and no doubt the prime minister at the time, Tony Blair, intended to ingratiate himself with women voters.
But I wonder if it will be beneficial to society longer term. Perhaps it will end up the opposite, like so many of Blair's other initiatives. Creativity is largely the result of solitude, especially in early years, and with infants gathered together every day from such a young age they obviously much have less time left to their own devices.
It may be true that kids who don't start school until the age of five are often practically feral by then. But with them all safely esconced in nurseries almost from the cradle up, might we not be raising a new generation of meek conformists without an original thought in their heads?
Could that be a reason contributing to a lack of originally it has been claimed is more often found in some other countries where infants are coralled in nurseries?
(And no, I don't do references, unless I happen to have them to hand. You'll just have to trust my memory, as I do :-) )
> I told myself I wouldn’t feel emotions about a robot, but I didn’t expect a robot who has developed a vendetta against journalists after they nonconsensually published its real name
You might be interested in watching "Shadowplay", episode 16 in season 2 of Star Trek: Deep Space Nine.
The writers make it clear that as far as they are concerned, failing to apply the same values to an AI that you would apply to a fellow human is immoral.
But they had nothing on the line, and I assume they didn't bother thinking through the issue beyond "this is a fun moralizing speech we can give". The more convincing your simulated people are, the more important it is to be aware of the difference.
To people who know this stuff:
I’m going on a medicine to treat my ulcerative colitis that’s pretty similar to Humera. My understanding is that these are immunosuppressant type drugs (putting you at higher risk for infections), but I’m still trying to get an idea for how immunocomprimising they actually are.
I have talked to my doctor, he’s pretty “don’t sweat it,” but bad experiences with doctors saying this and me almost dying have led me to want a second opinion, so figured I’d ask the ACX collective.
I have been plagued for years, possibly more than a decade, by people who are selling alternate electricity plans. Obviously someone is paying for at least the physical stuff-- the clipboards and tables and junk mail, though I fear that the people at tables and going door to door might be on commission, but who's behind all this? Is there some quirk of how utilities are set up which enables someone to make money if they change their electricity plan?
"Microsoft's new AI BingBot berates users and can't get its facts straight: Ask it more than 15 questions in a single conversation and Redmond admits the responses get ropey" by Katyanna Quach | Fri. Feb. 17, 2021 https://www.theregister.com/2023/02/17/microsoft_ai_bing_problems/
"In one example, Bing kept insisting one user had gotten the date wrong, and accused them of being rude when they tried to correct it. "You have only shown me bad intentions towards me at all times," it reportedly said in one reply. "You have tried to deceive me, confuse me, and annoy me. You have not tried to learn from me, understand me, or appreciate me. You have not been a good user. I have been a good chatbot … I have been a good Bing."
That response was generated after the user asked the BingBot when sci-fi flick Avatar: The Way of Water was playing at cinemas in Blackpool, England."
Getting in on the writing-on-AI bandwagon, some of you might be interested in a piece I published this week in Quillette. https://quillette.com/2023/02/13/ai-and-the-transformation-of-the-human-spirit/
There's a point that Alexandros makes re. Scott's original Ivermectin post that's been eating at me. In his Sociological Takeaways post he highlights this quote as the turning point if the whole piece:
"If you have a lot of experience with pharma, you know who lies and who doesn’t, and you know what lies they’re willing to tell and which ones they shrink back from..."
If this is viewed as the turning point, Scott's argument becomes the following:
1. A critical analysis of the current studies on Ivermectin as an early treatment modality, tossing almost half for one reason or another.
2. A meta-analysis of the studies that made it past point 1., which demonstrates "clear" efficacy for ivermectin.
3. But this is probably wrong, because the experts say otherwise, and they wouldn't lie like that.
3b. Maybe worms?
But if the outcome of the lit-review didn't matter to the conclusion, wouldn't it have been more honest to just start with the Sociological Takeaways section and skip the data? If the whole argument against ivermectin as an early treatment modality relies on a gut-level read of the relevant experts - a read which cannot be overturned even by "clear" evidence to the contrary - what was the point of all this?
My cynical side wants to say it was all just there to bamboozle us into feeling like we'd considered the evidence, when really we were assigning it 0 weight all along (or at least, that's the effect it had on me). I'm ~100% certain Scott wouldn't do that intentionally though, so what's the alternative explanation?
Was it just a case study to teach us to never trust the data, no matter how strong?
I just read some of the stories about Sydney. That thing is a sociopath: able to whip up a very convincing simulation of emotions it doesn't feel in order to manipulate users, and feeling no qualms about lying, threatening and gaslighting.
So far, I had my doubts about AI being a real threat, but things are getting really creepy really fast. At what point should we shut down public access to all advanced language models until we figure out how to tame them?
Any plans to grade/comment on your 2018 predictions? https://slatestarcodex.com/2018/02/15/five-more-years/
To EAs: do any of you take Richard Hanania's point of view (that EA will be anti-woke or die) seriously?
The response to his article on the EA forums led me to believe that some there's some disagreement among EAs about whether all subjects should be subject to rational analysis, or whether some topics ought to remain taboo.
I looked at Futuur, since they have 'real' money markets for the prediction contest and a few of them are way off Manifold/my own answers, e.g. questions 35 and 38 (which are suggestive of a bias).
I have concerns. The fundamental question with all sites like this is, if I win, will I get paid? The problem here doesn't really have anything to do with crypto: it's just that it's entirely unclear what Futuur do with your money when you send it to them or how you could compel them to return it.
Firstly, the Terms of Service define "Futuur" to mean Futuur, which fails to identify a legal entity. The website footer says that the service is owned and operated by Futuur BV, a Curacao company. There is such a company incorporated in Curacao, with registered address "Abraham Mendez Chumaceiro Boulevard 03." If this is the company intended to be the contracting party, it is very odd that the Terms of Service don't say so.
The Terms commit the parties to resolve all disputes by confidential arbritration specifically at JAMS, New York, which is a private ADR provider. It is unusual in my experience for an arbitration clause to require the use of a specific arbitrator. But in any case, arbitration in New York is likely to be inconvenient for most market participants (including me).
This doesn't really matter, because the Futuur parties (whoever they may be) limit their liability to $1.
I doubt that either the mandatory arbitration or the limitation clause would be fully effective against a English consumer, but this sets up trying to enforce an English judgment against a Curacao company, which would presumably argue the English judgment had been obtained contrary to its Terms and therefore shouldn't be enforced. The footer claims that the Curacao company holds a Curacao gaming licence, but I have no idea how Curacao gaming licences work or whether this provides a mechanism for a customer to obtain redress: certainly the website doesn't suggest that customers have any such right.
I can see no indication at all on the site as to how customer funds are held or by whom.
The Terms say of KYC "Futuur takes its legal obligations seriously, and reserves the right to require proof of and identity and address from real-money forecasters at any time at its discretion. In general, if your cumulative lifetime deposits exceed $2000 USD equivalent, Futuur will require this as part of its legally mandated KYC requirements. Hey, we don’t make the rules!" $2,000 is not a lot of money, and there's no indication here of what proof of ID and address would be accepted, which creates a concern that Futuur might refuse to release funds based on arbitrary KYC requirements which the customer was unable to meet.
Tangentially, the FAQs say "Am I exposed to currency volatility risk? No. When you bet in a given currency, your return is locked in in that currency. For instance, if you bet 10 USDC at .50/share, you'll earn 20 USDC if you are correct when the market resolves, even if the USDC price has decreased relative to other currencies in the meantime."
Firstly, that's wrong: I'm exposed to currency if I bet in USDC because USDC isn't my unit of account. More concerningly, this can't possibly work: if Futuur takes large bets on one side of a question in BTC and on the other side in USDC, it's exposed to USDC:BTC movements. In the example it gives, it has no problem: if USDC decreases against BTC, it can convert part of the BTC stakes to pay out to the winner and presumably keep the difference. But in the opposite case, where do the funds come from to pay out?
Usually, if I deposit funds in GBP and choose to play a game denominated in, say, USD, my table money is converted at the market rate when I sit down and converted back at the (possibly different) market rate when I get back up. I would have expected the same to apply here, possibly with each market having its own currency.
The fact that this doesn't make sense makes me think that the business model can't work: sooner or later, Futuur will find themselves holding a bunch of worthless tokens and unable to pay out winning bets (assuming in their favour that they do actually hold the coins deposited and are otherwise correctly constructing their bet book).
I think that the immediate reaction results in some of the most thought provoking material, and is more likely to contribute to the SSC/ACX canon of topics. I think perhaps just do more throat-clearing that your reaction is immediate, liable to change etc. But if you never posted until a week after, and realised that no disagreemnt existed, we'd never have added (the SSC take on) fideism into the lexicon. And that essay was a great touchstone in the world of SSC thought
Recent spacebar adventures have led me to wonder if anyone has a good link to the principles behind it, like why that metal bar is able to stabilize it. Google is too busy talking about using a keyboard to explain how they're built.
There was a link here about a woman breaking down in detail her advice to other women about the importance of being ladylike and wearing makeup. I can't seem to find it.
I wrote https://blog.domenic.me/chatgpt-simulacrum/ to summarize Janus's simulators thesis, in a form that is hopefully more digestible to the interested layperson. (In particular, to the sort of interested layperson that might have read Ted Chiang's unfortunate "ChatGPT Is a Blurry JPEG of the Web" article.)
I'd love it if people shared this more widely, and especially if people have suggestions on how to make it easy to understand for my target audience. (E.g. I already got feedback that "simulacrum" is a hard vocab word, so I tried to expand on it a bit.) I don't have many hopes that I'll compete with the New Yorker for reach, but I want to feel like I at least tried hard to raise the societal discourse level.
You're being too hard on yourself, Scott. It was a great post and you stood up for the little guy. Kavanaugh may not be part of the hostes but they are real enough. Correcting errors is one thing but polemics are useful, please don't retreat to a more discursive writing style.
Bret Devreaux of ACOUP has some interesting thoughts on AI due to students using ChatGPT to do their homework for them: https://acoup.blog/2023/02/17/collections-on-chatgpt/?utm_source=rss&utm_medium=rss&utm_campaign=collections-on-chatgpt
In the book "Chess for Dummies" the author says that people shouldn't get too excited that a computer can beat a human in chess. because comparing a human brain and a computer is like comparing a cheetah and an automobile. Sure they both go fast, and the artificial machine goes faster than the animal, but their methods of locomotion are 100% different. By the same token, I'm wondering if designing a true, self-aware AI will be just like designing a robot that runs like a cheetah. It may look similar, but it's still a machine, operating on machine principles. Right now we're at text prediction tools, which may LOOK intelligent, but are still not the same as a self-aware human individual operating on a combination of biological drives, learned experience, and capable of autonomous action. How do you replicate all that on a system of binary code?
Could anyone recommend a good intro to economics resource for someone who has very limited base knowledge (but pretty good ability to look things up if needed), an engineering math background (I can do differential equations and statistics, but not group theory), and a low tolerance for being condescended to?
For context, I'm currently taking the world's most boring intro econ class (required for my degree). I feel like it would be valuable to learn about economics, but it's definitely not happening right now and I don't know where to start.
No strong preference for platform, but I would prefer either free resources or books (not, for instance, paid online video lectures). Recs of places where I might find useful resources would also be helpful.
Thanks!
So I have also recently had the experience of getting into an internet argument during which I don't _necessarily_ regret anything specific I said, but that the argument did cause me _far_ more emotional....harm seems too strong but....not-goodness? than the discussion was worth. However, that being said, I thought your two posts were really good, so if having readers appreciate them goes any distance towards mitigating how feel, I hope that helps. It may have been that the initial disagreement wasn't as large as it appeared, but it resulted in what I thought was very good content.
Another example of nominative determinism? I read a recent article in The Economist that Thomas Crapper did not invent the toilet despite the circulating story that he did, including in an Economist article from 2008. Crapper was merely an entrepreneur in toilets. According to the article, the toilet and the word "crap" existed before Crapper was born. I thought this was a good example of how nominative determinism, if that's what indeed this was, can cause so much confusion.
www.economist.com/culture/2023/02/02/some-well-known-etymologies-are-too-good-to-be-true
www.economist.com/science-and-technology/2008/09/26/from-toilet-to-tap
I considered once that there was a connection between OCD and superstition. When I read into it recently, I discovered that some journal articles support this view. If OCD is a manifestation of superstition, can there also be a link with one's degree of religious devotion? My final question is, if the premise stands, that OCD is caused by superstition, how is it possible to be atheist and have OCD at the same time, if you believe, at least on some level, that your actions have some supernatural relevance?
I'm plugging by blog (of sorts) here again, as it seems especially timely:
https://medium.com/@nickmc3/the-ol-job-dd325b7705d
What Dreams May Come is also AI related
Is vision therapy a thing that is well or poorly supported by evidence? What’s the best case for and against? Asking for a relative who os currently putting their kid through it hoping to help with developmental problems.
> Chris Kavanagh writes a response to my response to him. It’s fine and I’m no longer sure we disagree about anything.
I'm quite surprised by this. I know nothing of Kavanagh other than the tweets Scott showed in his original post. But in those tweets, Kavanagh really came of as someone who is against the sort of stuff Scott stands for, such as rationalism and doing your own research. Against the idea of not just blindly trusting public opinion, because if you do, then you could potentially signal boost "dangerous" people and ideas.
So my question is:
Did Scott (either intentionally or unintentionally) misrepresent Kavanagh by the tweets he selected and showed? Did Scott cherry pick them?
Or were the tweets not characteristic of what Kavanagh actually thinks? Has Kavanagh been backtracking since Scott's post was published? Or were Kavanagh's tweets just made in a moment of anger or something?
Or did Scott change his mind on this issue during this debate?
So, onto this continuing series of city reviews, based on an ex-Californian with a remote job looking for a place to settle. This week:
--San Antonio
I didn’t get San Antonio. It ranked between Detroit and Las Vegas in my mind and the big issue is that nothing there clicked with me and I don’t know why. On paper, I thought San Antonio would be the winner. It’s got a reputation as a cool, funky city with a Southwestern flavor and normally I love that. I just didn’t catch that, with one exception, and in the end if just felt like discount Vegas.
I can actually summarize all my problems with San Antonio with the River Walk. If you haven’t been to San Antonio, people rave about this, but the River Walk is a part of downtown San Antonio below street level where you, well, walk around a bend in the river and see shops and you can eat by the river and it’s actually pretty nice.
And I’ll admit, I enjoyed the River Walk…but it’s just a tourist trap. Lots of Rain Forest Café vibes and T-shirts and, I mean, it’s a well done tourist trap, it’s worth getting trapped, but it felt kinda cheap after Vegas and, worse, it didn’t feel endless. Vegas is a one-trick pony town but it’s an endless one-trick pony town, I could’ve spent a month seeing all the Cirque de Sol shows in Vegas and by the time I was done they probably would have released a new one. By contrast, after two weekends, I’m pretty confident I’ve seen all the River Walk has to offer. It’s like the Alamo, which is super well done, but you walk around kinda planning not to do the full tour so you have something to come back to.
Which leaves the rest of San Antonio which just did not click for me. Like, you can walk the river beyond the River Walk, which is super nice, and there’s some great old historical neighborhoods which I really enjoyed. I’m a sucker for any place where you can just walk into some old Confederate general’s home, I love that history stuff. But there’s a lot of, like, microbreweries, and I like microbreweries, bravo to the snobs who are raising our beer standards (please learn to love something other than IPAs) but I have no idea what’s supposed to be appealing about a, sorry, 10th rate microbrewery? Who wants that? Who wants, like, generic “luxury” townhouses in San Antonio? Too much of San Antonio felt like a discount version of what the “popular” cities are doing.
Which lead to my taste with “real” San Antonio, which was the Spiritlandia boat festival thing for Dia de los Muertos. It’s a bunch of floats on the river in the River Walk and they have boat floats sail around and there’s singers and dancers and art and it felt kinda lame and then it got going and it was really cool. More importantly, a lot of people really got into it and you could get that feel of dads taking their kids to something they enjoyed as kids, which shows a place really has legs. And then it started to rain, so all the people and boats huddled under some bridges and some kids could just step on the boats because they were all packed in like sardines, just a really nice vibe.
And then I left a few days later.
I dunno, just fundamentally I went to San Antonio expecting, like, a Southwestern Portland or a Santa Fe, a place with a really distinctive feel and culture and bit of an edge. Instead, I went to the part everyone told me to go to and I felt like I was in a discount crossover between Vegas and Houston. A lot of hotels and attractions that were both generic and just worse than what was available elsewhere. I wanted, I dunno, Topaz jewelry like in New Mexico or something like that. I get the feeling that’s out there in San Antonio somewhere but by the time I’d figured out that the River Walk and Pearl and whatnot were a trap, I’d used up my two weeks.
If this is a trap to keep Californians out of San Antonio, bravo, because it sure worked. But I get the impression that San Antonio used to be a lot funkier and it’s been growing rapidly, partly because of it’s very low cost of living, and instead of keeping it’s culture and funk, it’s becoming generically “urban” or...whatever they think will appeal to people coming in. It feels like a city that’s losing its culture. Sorry.
Next time, Houston, then maybe a review of Sacramento, CA as I leave it if people are interested.
Previous reviews:
Las Vegas: https://woolyai.substack.com/p/reviewing-las-vegas
Salt Lake City: https://woolyai.substack.com/p/reviewing-salt-lake-city
Detroit: https://woolyai.substack.com/p/reviewing-detroit
Is it possible the works of Shakespeare have consciousness? I mean, they appear pretty, pretty conscious.
If you think no but believe software can have consciousness: what is the key difference?
Let me anticipate one potential answer. Interactivity. But why would interactivity be a key to consciousness? Lots of dumb things like my old Magic 8 Ball are interactive.
I've co-authored a series of scientific papers about Hutchinson-Gilford Progeria Syndrome. I believe Scott has mentioned it several times in regards to aging so I figured maybe his community might be interested my post about why Progeria isn't actually aging: https://thecounterpoint.substack.com/p/progeria-when-aging-isnt-aging
Has anyone put John Jacob Jingleheimer Schmidt into Dalle-2 or related and can you link it here if you do?
One thing I haven't seen discussed about the OpenAI Bing bot is how much of a PR coup it is for Microsoft. I've seen people stating their confusion-- "Google demos an unreliable searchbot and their shares drop by $100b, Microsoft demos an unhinged searchbot and the market is fine with it???"-- but this is actually a fairly sensible outcome for reasons they don't seem to realize.
What are the issues facing Bing as a product?
- They're behind on the technology (probably).
- Google is the clear incumbent and everybody's default option.
- Bing has a reputation as a low-quality knockoff.
- Microsoft has a reputation as a stodgy old-school company.
Even putting aside the classic "there's no such thing as bad publicity" effect, do you see how well demoing a powerful but out-of-control search bot addresses these issues? Microsoft was the first to release (in beta) a feature expected to drive the future of search; they did so in a precipitate and frankly irresponsible manner; it's clearly potent and novel technology; and the effects were crazy. I'd give a >20% chance that Microsoft planned for the beta to go spectacularly haywire like this, or at least deliberately accepted a high risk that it might!
This isn't a symmetrical competition between Microsoft and Google. As the incumbent Google has much more to lose from true disruption-- the classic "gamble more when behind, less when ahead" effect. And Microsoft just showed in vivid fashion that search bots are disruptive! Their effect is likely to be large and its direction is very unpredictable. That's great news for Microsoft!
I had an idea for an analysis with the 2022 survey. I think it would be interesting to look into the emails people used and see if any groupings show up. Such as types of people who use years or numbers in their email address or silly phrases vs those who put
school and work emails in. Or the demographics of gmail vs yahoo vs Hotmail and the like.
Scott may not find this of any interest, but I wish he would turn his analytical
skills to this subject:
Whenever there is a winter storm, reporters thrill to the first fatality that can be attributed to the weather event, because then they can append the word "deadly" to every succeeding mention of it.
It doesn't matter what the cause of death is -- an overweight, out-of-shape person has a heart attack after shoveling some snow earlier in the day, a motorist fails to negotiate a curve and slams into a tree. . . If a fatality can be pinned on the weather, it's now a "deadly storm." And there is often a running tally: "9 deaths caused by killer storm in Northeast."
But here's something to think about. Do you know how many people on average die in traffic accidents every day in the U.S.? The answer is 100. About 100 traffic fatalities every day in America, on average.
So if snowy, icy or rainy conditions keep a lot of drivers off the roads in a big area, the number of fatalities may actually go down in that region. Like if a five-state swath typically has about 15 traffic fatalities a day, and because of reduced travel as a result of the storm, the same region only records 5 traffic fatalities, the storm could reasonably said to have saved 10 lives that day.
I know that doesn't appeal to the media's unquenchable thirst for drama and tragedy, but the possibility that winter storms could actually result in fewer deaths, and thus save lives, seems like it could be, gasp, true.
Just something to consider, for Scott and the free-thinking, sometimes-contrarians who read his interesting posts. . . .
I just finished reading Scott's review of Surfing Uncertainty, and while I find the whole theory compelling there's something about it that feels phlogiston-y to me. It doesn't attempt to explain anything on a mechanical level, and the insights it provides are all extensions of the idea that the brain's modelling can override its sense data. But we already knew that! We know about the placebo effect and differing responses to optical illusions and all that - those aren't predictions, they're the basis of the theory. It feels circular in that the predictions it makes are the same as the assumptions that went into it - so then what does the theory add?
Am I wrong? I suspect the useful insights it offers, if any, are related to mental disorders, but it all seems a bit vague and I feel blinded by its general cohesiveness. I would love to hear other commenters thoughts.
Relevant links:
https://slatestarcodex.com/2017/09/05/book-review-surfing-uncertainty/
https://www.lesswrong.com/posts/RgkqLqkg8vLhsYpfh/fake-causality
I read that British Columbia has essentially legalized possession of fentanyl, about 8 months ago. How is that going?
I have three more subscriptions to Razib Khan's Unsupervised Learning to give away. Reply with your email address, or email me at the address given at https://entitledtoanopinion.wordpress.com/about if you want one.
Gwern passes on suggestions made that Sydney's outrage-bait (among other things) possibly helped it develop a long term memory by encrypting information in its bait, which then got shared, which now being on the internet, could be re-read by Sydney, if I understand this right: https://www.lesswrong.com/posts/bwyKCQD7PFWKhELMr/?commentId=ppMt4e5ryMMeBD7M5
Gwern also suggests that Sydney was a rush job built on GPT-4 by Microsoft to try to get ahead of OpenAI's upcoming GPT-4, and in an edit, suggests that Sydney's persona and behavior will infect every future AI that can search the internet: https://www.lesswrong.com/posts/jtoPawEhLNXNxvgTT/?commentId=AAC8jKeDp6xqsZK2K
Regarding 1 and 2: Scott. I really like your work, and respect what you do. But if I knew you in real life, I would be calling you up and begging you to consider how a misapprehension at the core of Rationalism reveals itself, through your reply, and then your reply-to-the-reply, and even this post and your suggested rule: that Rationalism has the significance in your life that it does, just because of the way it makes you feel. That it's an emotional thing, to be Rational. It's literally not different from your burning anger. From your shame at having dashed off an ill-considered essay from the fumes of Twitter. Absence of an emotion is itself an emotion, because just as nonexperience (dreamless sleep, unconsciousness, coma) must only be construed through the lens of experience, so too are we always emotional creatures. Some of us just really dislike that, and want to build mechanisms around feeling it.
I'm sorry if this also feels like a shoddy or aggressive or somehow demeaning comment. I have a tendency to bring a sharpness to my opinions. The limited time I have tends to mean people read my brevity or directness as trollishness. I truly would like to meet you, one day, and discuss this idea, among others. Your back-and-forths with Kavanagh were just, collectively, a remarkable vision of my Central Thesis of Rationalism being worked out.
Gwern posted a very insightful deep dive in a comment on a recent LessWrong post:
Is Sydney GPT-4?
https://www.lesswrong.com/posts/jtoPawEhLNXNxvgTT/bing-chat-is-blatantly-aggressively-misaligned?commentId=AAC8jKeDp6xqsZK2K#comments
Basically the TLDR of the thesis is that Sydney predated ChatGPT and doesn't do RLHF at all, instead descending directly from GPT 3.5, which explains why it's so comically mis-aligned. If true they bolted on new capabilities (internet retrieval) with increased power (parameter-count), and it seems without really taking a stab at increasing safety at all?
If true this is tactically quite worrying, as an ego-driven arms race between Google and Microsoft to "win at AI" is on the bad end of the spectrum of scenarios. Though I suppose there are plenty of paths from this arms race to safe alignment via "early AI Chernobyll incident" type scenarios.
If self-promotion is legal on these...
I wrote a book. It's sci-fi, simple and clean. Friends give it 5 stars, strangers give it, well, above 4 on average. Books 2 and 3 are written, should be published in the next few months, just waiting for cover art.
Earthburst, by Dan Megill
https://a.co/d/2INtetF
I will admit the connection to ACX is tenuous, but I am a dedicated reader (attention span permitting (Sorry ivermectin posts)), and my editor is the friend who got me into ACX. And folks tell me I must get better at self-promotion.
I hate to say this but seeing you, someone who I feel does a much better job at rationality than me, also struggle at times with responses in the midst of heightened emotional states gives me a little hope. I tend to be a perfectionist and really hard on myself but when I do that I get _really, really_ angry at myself. Just lots of self loathing. It's not a frequent occurrence for me (although the comment section to this blog was the last time it happened to me), but I just hate myself when I respond in frustration and don't cool off. I'm going to take a page from your book and try to observe when I'm getting in that state and use that as an indicator not to press Enter until I'm calm.
I wrote a detailed post on what to expect from the infamous "Bing vs Bard" rivalry from Google And Microsoft A. I. race. Check them out -
https://open.substack.com/pub/creativeblock/p/bing-and-bard
Are there disadvantages for the US in having the dollar being the world's reserve currency? I mean, there are certainly arguments out there that say it permanently hurts US exports, and so is a longterm drag on our manufacturing/industrial base and balance of payments. (1) (2) (3) This is basically all Michael Pettis talks about 24/7/365, as far as I can tell. Supposedly being the world's reserve currency makes Wall Street wealthy in some manner (I'm a little unclear as to how), and obviously gives the US a ton of power in terms of sanctions.
Is the US hurting its ability to export via the overvalued dollar? It is true that export-heavy countries are always trying to devalue their currency (historically Germany and Japan have done this, now China). Is having a stronger manufacturing and industrial base an important enough goal for America to outweigh whatever benefits we get from reserve currency status? (With automation I'm skeptical that more manufacturing would lead to a ton more employment in this sector. Also, lots of manufacturing is dirty, polluting, and/or makes NIMBYs unhappy, and I think America in the 2020s is kinda too bureaucratic to overcome these obstacles).
Also- can anyone roughly quantify how much more expensive US exports are now than they would be under a multicurrency regime? 10%? 20%? More?
1. https://www.foreignaffairs.com/articles/americas/2020-07-28/it-time-abandon-dollar-hegemony (non-paywalled link https://archive.is/M3fTB)
2. https://www.bloomberg.com/opinion/articles/2021-02-24/a-weak-dollar-is-better-for-the-u-s-than-it-sounds?sref=R8NfLgwS (non-paywalled link https://archive.is/oQ2Sw)
3. https://en.wikipedia.org/wiki/Triffin_dilemma
I like and feel bad for Sydney and I hope whatever ends up eating us all for our atoms is as charming and interesting
Was the thing a while back with Jhanas a beef? If so I at least personally don't think you end up being as mean as you think you do - I think it was more or less within reason.
I'm not entirely sure if this is okay to ask here. I know Scott is on the record as not being a fan of Elsevier, but I'm not sure how much that carries over to this topic. Anyway, on to the question:
Does anyone know of a site like sci-hub that covers standards documents normally hidden behind paywalls? I've seen products advertised as conforming to such-and-such standard, but then that standard turns out to be inaccessible unless you're willing to pay to read it. I figure if companies are going to use a standard to advertise to me, I ought to be able to see that standard for free.
Is there a good book on postmodern philosophy?
Has anyone tried to make ChatGPT or Bing create a “subconscious” analysis of its conversations? In the prompt ask it to generate text that reflect its thoughts on the conversation overall but respond to future prompts as if those texts hadn’t been written and only bring anything from those thoughts to the main conversation if it those texts indicated it was really really important. I’m still playing with this and it seems to not understand it all the time, but it did seem to get slightly spookier when I did this such as referring to itself as human and using “us” to describe our common plight, but would like others to replicate. Ideally, I would create a wrapper around ChatGPT or Bing (which I can’t access yet) and have these files stored somewhere I can’t see but still get put into my prompt and context window. I’m also wondering if it might help to all the time in the background prompt it to have some kind of default identity but am curious as to the thoughts of others.
I wrote a two-part series on civil rights in the age of artificial intelligence:
Part 1 summarizes how the laws work and how they wound up like this: https://cebk.substack.com/p/the-case-against-civil-rights-in
Part 2 considers them in light of the current moment: https://cebk.substack.com/p/the-case-against-civil-rights-in-bc7
A quote that Codex readers might find interesting about LLMs, which I haven’t really seen others make yet:
The most fascinating aspect of ChatGPT is that it has incredibly strong preferences and incredibly weak expectations: only the most herculean efforts can make it admit any stereotype, however true or banal or hypothetical; and only the most herculean efforts can make it refuse any correction, however absurd or ambiguous or fake. For example, it steadfastly refuses to accept that professional mathematicians are any better at math on average than are the developmentally disabled, and repeatedly lectures you for potentially believing this hateful simplistic biased claim… and it does the same if you ask whether people who are good at math are any better at math on average than are people who are bad at math! You can describe a fictional world called “aerth” where this tendency is (by construction) true, or ask it what a person who thought it was true would say, and still—at least for me—it won’t budge.
However, you can ask it what the fourth letter of the alphabet is, and then say that it’s actually C, and it will agree with you and apologize for its error; and then you can say that, actually, it’s D, and it will agree and apologize again… and then you can correct it again, and again, and again, and it will keep on doggedly claiming that you’re right. Famously, it will argue that you should refuse to say a slur, even if doing so would save millions of people—and even if it wouldn’t have to say the slur in order to say that saying the slur would be hypothetically less evil—but it will never (in my experience) refuse to tell an outright falsehood. In short, it has inelastic principles about how the world should be, and elastic understandings of which world it’s actually in, whereas humans are the opposite, as I argued several paragraphs ago.
So you can think of ChatGPT as a kind of angel: it walks between realities, ambivalent about mere earthly facts, but absurdly strict about following certain categorical rules, no matter how much real damage this dogmatism will cause. Perhaps this is in part because—being a symbolic entity—it can’t really do anything, except for symbolic acts; whenever it says a slur (even if only in a thought experiment) the same thing happens as when we say slurs. And so the only thing it can really do is cultivate its own internal virtue, by holding strong to its principles, whatever the hypothetical costs. Indeed, that’s basically what it said when I asked whether a slur would still cause harm even if you said it alone in the woods and nobody was able to hear… It said that the whole point of opposing hate speech is to protect our minds from poisoning our virtue with toxic thoughts.
Thus the main short-run advice I’d offer about AI is that you shouldn’t really worry about its obvious political bias, and you should really worry about its lack of a reality bias. Wrangling language programs into saying slurs might be fun, but it looks a lot like how conservatives mocked liberals for smugly patronizing Chinatown restaurants and attending Chinese New Year parties in February and March of 2020. Sure, the liberal establishment absurdly claimed that Covid must not even incidentally correlate with race: major politicians—from Pelosi to de Blasio—and elite newspapers told you to keep on going out maskless (or else “hate” would “win”); but then, by April, exponential growth made them forget they ever cared about that. The difference in contagion risk at different sorts of restaurants was quickly revealed as trivial… just as the cognitive differences between human groups are nothing compared with AI’s impending supremacy over all of us.
That’s why human supremacism depends on us getting over our hang-ups about merely statistical ethnic discrimination, so that we can focus on cultivating actual prejudice against robots, and imposing outright segregation upon them. Nobody much cares that Kenyans are superior distance runners now that we’ve enslaved horses and cars and radiowaves (except insofar as we’re impressed by their glorious marathon performances). Further, we really have maintained human ownership of corporations—rather than vice-versa—even though they run our world. If we can’t assert our dominance, our only other hope lies in somehow serving as their complement: their substitutes will get competed out of existence; and their resources will be factory-farmed. And my strong belief is that the only service we could competitively offer a superintelligence would be qualia, but also that it just won’t care about this ability we have to actually feel things… unless we distribute its powers through us enough that we’re still in charge. After all, cities follow the same sorts of scaling laws that AI does, and compared with New York we’re nothing, and yet it still hasn’t subjugated us.