The following is a rough transcript which has not been revised by The Jim Rutt Show or Erik Torenberg. Please check with us before using any quotations from this transcript. Thank you.
Jim: Today’s guests, and we have two today, are BJ Campbell and Pat Ryan. BJ is the head honcho at one of my very favorite Substacks, which is Handwaving Freakoutery. This is what he says about himself or I guess actually about the Substack, “This is a song about gun mathematics,” and he does do some amazingly good quantization of things around the gun culture wars, culture war analysis more generally, media criticism, and then this is his words, “occasional unrelated esoterica that prevents me from sleeping at night.” BJ’s a returning guest. He appeared in occurrence 024, where he and I talked about the woke religion. You can follow him on Twitter at @Freakoutery.
Pat Ryan, he’s an interesting fellow. I would describe him as a semi-pro internet troll, and he’s not only a first class troll himself, but from time to time, he organizes troll battle teams and rampages. If this makes you think about a knuckle dragon moron, let me tell you, you’re wrong. Pat’s one of the smarter people I know, a very interesting fellow I’ve been talking to, I don’t know, three or four years probably, and you can find him on Twitter at @TyrantsMuse. Thank you guys for coming on.
BJ: Thanks, Jim.
Patrick: Thank you for the introduction, Jim.
Jim: Yeah, and great to have you back, BJ. Today, we’re going to talk about a topic that BJ in particular has gotten behind in a fair amount of his writings, and that’s the concept of egregores. Why don’t we start with BJ? Maybe give a little bit of history of the term egregores. I did a little snooping around and it seems like it goes back a lot longer than I might have thought and how you’ve packaged it up as a concept for our current situation.
BJ: Yeah, it’s an interesting topic. It’s a word that traces its roots back to early occultism, and the concept from the occult standpoint back in the old days was, and I’m not an expert on occultism by any form or fashion, but the concept is generally that when you get a bunch of cultists that are all together invoking the same thing or their minds are all in the same place, that the group gestalt could give birth to another entity, that being a demon or a representation of what they’re thinking or doing.
So it’s like as if group think were to have a baby. It’s how you can think about it. That became very relevant in certain internet circles over the past year or two years or whatnot because people started using that to describe some of the behavior they were seeing online. There were people who were using the word in different ways. So to try and make sure that I was clear and not … because it’s a very overly broad term. Really, what I did is I put it in a pot and mixed it up with some Jim Ruttisms, to be honest, because you used to talk … You are a hot topic on this podcast for all who’s talking about meme space entities, right?
So what I’d like to talk about when I’m speaking on in HWFO, which is my Substack and talking with Pat is we call them meme space egregores. So the idea is this. You think about it like Google Maps. So when you’re navigating someplace nowadays, you plug your cellphone in and you tell it where to go and it routes you a course. When you’re navigating back in the 1990s, everybody had a road atlas, right? You pull out your road atlas, you’d figure out what roads to go, you would navigate that way. You might write down a list of instructions of directions, but as smartphones have become more ubiquitous, there’s no reason we shouldn’t just always trust the smartphone when we’re trying to navigate someplace, and the smartphone has the additional feature that if there’s traffic or if there’s the road or the bridge is out or something like that, it catches updates on the feed about the map of the territory, right?
So pretty much everyone I know is now using their smartphone to navigate and nobody’s carrying around an atlas anymore. Well, it would be the height of absurdity to think that the only thing we’re using our cellphones for is road navigation, right? We’re using it for a lot of other things. We’re using it to keep up with the news and to listen to music and to keep in touch with our friends, and we’re outsourcing lots of different functions to our phones, grocery lists or calorie tracking or in some cases morality, right?
So what’s happening now with some groups of people, and we could call them the woke, but they’re not the only ones that this is happening to, they’re outsourcing morality to their phones. The feature of that is that their morality, and by loosely defining morality as a set of behavioral indoctrinations that influence your behavior and how you go about your daily life, that can be updated and you can catch it from your social feed instead of from a book, the Bible or the Koran or whatever, right?
So if you catch it from your social feed, then the rules of appropriate behavior can change on a dime. So a good example of this for instance might be, okay, so in early 2020, there was this sense among social media that during the COVID ramp up, during the early COVID fear, if you left your house, you were a murderer because you would be spreading the germs. That changed in the span of about one day to if you don’t leave your house, you are a racist, right?
Everybody who is catching their updates on their phone, their behavioral indoctrination is catching them on the feed. They flipped from one polar extreme to the other in the span of a day, and they saw no cognitive dissonance from doing so because it was as if the words of the Bible changed. They caught an update of their behavioral indoctrination on the feed.
Jim: All right. Let me dig in here a little bit-
Jim: … which is if we think of egregores as a class of meme plex or meme space entities or the other term I’ve used for many years, it’s falling out of fashion, is meme plexes, essentially anything that is not physical that nonetheless seems to have attributes of existence, and adaptivity might be thought of as a meme plex or a meme plex entity like for instance, a company. What is a company? A company isn’t actually real. It’s a bunch of legal agreements and people and stuff move around under the spell, the magic spell.
Patrick: So as the government.
Jim: Yeah, so as a government, so as your religion, even a family is, right? Anything that’s a set of formal or informal norms set up between and around people that have impact on their behavior would be a meme space entity. One could say that egregores are those things, all of those things. Do you want to specifically limit it to the case where they’re bound around morality or some other set of virtues, values, and norms?
BJ: Not necessarily, but the one feature that I think that is very important to understand what these things is the fertile ground from which they erupt is the interconnectedness of social media itself.
Jim: This particular class of egregore we should say.
BJ: Yeah, right.
Jim: Yeah, certainly the new examples.
BJ: Right, the things that we’re speaking about on HWFO, the meme space egregores. Now, you could also technically say something like the idea that money has a value might be a meme space. It might be an egregore that existed for thousands of years.
Jim: It’s a big one. It’s the big one. It’s the big one.
BJ: What we’re talking about specifically here is this. Okay. So when you’re going through your Facebook feed, you’re doom scrolling or on Twitter or whatever it is, you’re looking through a series of stimuli, you’re identifying stimuli, you’re deciding either to like or share that stimuli to amplify it. When you do that, then it jumps out of your feed and goes into somebody else’s feed.
This is how Twitter, Instagram, Facebook, all the social feeds work on the same concept, and that is exactly what a neuron does in a brain. The neuron, it collects information from dendrites. That’s its feed. It has some generator function inside the nucleus that decides whether or not to fire, which is its share button, and then if it clicks the share button, then it sends a signal down the axon that then populates other neurons’ dendrites.
So a social media network is set up just like a brain is set up. So it ends up behaving in the way that an artificial neural network might behave. The learning generation inside your brain is when that generator function inside neurons gets adjusted some, and then sometimes neurons break from other neurons and make new associations inside the web, and that’s like you unfriending someone. The thing works out the same way. So we are all participating as nodes inside a brain, but the thing about a neuron is that it doesn’t have any idea what the brain is thinking because it’s just a neuron. The thought lives one level up from the neuron.
Jim: The collective phenomenon, just as our consciousness is a collective phenomenon. You can’t put your finger in your head and go, “Yeah, that’s where the consciousness is,” right?
BJ: Right. So if we postulate, for instance, that there is some level, and this is getting a real hot topic now with ChatGPT and whatnot, but there are some AI researchers who would believe that if you have a complex enough artificial neural network, that exceeding some level of complexity, it would become self-aware. If we also say, “Well, social media is a artificial neural network,” then there is some complexity by which it would become self-aware or where self-aware entities would start to matriculate inside of it. If those self-aware entities were competing for overall brain space in the way that modern cultures do, then it might look a little bit like what we’re seeing with the culture war on Twitter and Facebook and such right now.
Jim: Well, let me push back a little bit about this or just to try to clarify and defuzz this concept of self-awareness. One of my areas of deep study for the last, holy shit, how long has it been now, almost 10 years, has been the scientific study of consciousness and closely related conscious cognition. Terms like awareness, sentience, consciousness tend to be used and abused at great length.
Patrick: They’re very slippery.
Jim: Yeah, and I would suggest people should be wary of using awareness or not with respect to these things.
BJ: Yes, I agree, I agree.
Patrick: Yeah, I agree. Even that’s what
Jim: Because they have complex adaptive systems is what they actually are, and there are many classes of complex adaptive systems that don’t show awareness. In fact, the school of thought that I come from in this area is from John Searle that our consciousness and our conscious cognition is a specific biological phenomena of us, and there are architectural features and energy that’s spent to support that, and that machines someday will be able to have analogs of those in the same way that, for instance, the food industry, the pharmacology industry, and the chemistry industry have so-called digesters where they use yeasts and bacteria to process chemicals in ways that are analogous to the human digestive system, but guess what? They don’t look like your colon, let me tell you that, nor do they act like it.
So I would expect that when we get machine awareness, it will have analogs stores, but it will have structural things that are similar to how the brain works. I see nothing at all, certainly in LLMs, large language models, that would support awareness. In fact, quite the contrary, they’re feed forward networks. There’s no feedback loops. I would say one could make a little bit stronger case in something like Twitter since it’s a series of strange loops that go back and around, and one could have more interesting systems dynamics that are reactive and adaptive. So I think that’s a little bit sharper way to think about this terminology.
Patrick: Now regarding the egregore, regarding all of this, one, I’m just going to come out the closet here. I don’t even think consciousness exists as a concept. I don’t think it’s real. I don’t think it’s even worth our time to even entertain that. I’m just going to open up with that, throw that stick of dynamite into the conversation. I don’t think consciousness actually exists. I think it’s a theological concept. I don’t think it actually has any basis in science or reality at all.
So regarding the neurons doing this behavior interacting with each other, we can say that about most neurons in the general sense, but I want to focus on Broca’s area, specific part of the brain that deals with language. So the LLM is a language entity. It’s dealing with words. It can deal with pictures now, but in the end, it’s dealing with words.
Why does this matter? Because if you take any meme and you throw it on the internet, if you just throw a picture on the internet out of context, it might work, it might not work, it might go viral, but if you throw words on it, that meme goes viral more often than not. So the word component of a meme is essential for its impact.
So Broca’s area matters deeply regarding the internet, and that makes sense because the internet started as people just typing words at each other, whether it’s BBS or telenet or any type of SSH or early technologies was about transferring symbols and words. So if we talk about egregores, and I chimed in early, I said a specific class of egregore. So money is an Igor that deals with something that’s outside of words, but on the internet, all the egregores we are dealing with are word-based. So Broca’s area would be the neuron function to look into.
Now, regarding the idea or the possibility which we could measure in egregore, we can put metrics on it, it’s velocity, it’s volume, it’s size, ways to put early handles on it. I’ve come up with a … I rarely like to dig my heels in and have a hard opinion on something, but on this one I think I’m actually on the money. I actually measure grammar velocity to determine the egregore influence.
Jim: Yup. That’s very interesting. I love the fact that you’re using Broca’s area. I do that all the time. I say if you really want to understand what LLMs are, they’re not really yet a replacement for Google, but you can think of them as a Broca’s area for technology. One of the things that we don’t notice is that we don’t actually consciously form our spoke. Why is it doing that? I guess I hit it by mistake. Our conscious-
Patrick: Broca’s area.
Jim: Broca’s area. That’s another piece of my little radio sound effects, but we don’t actually consciously when we’re speaking form our sentences.
Patrick: That’s right.
Jim: Our Broca’s region takes vague ideas in terms of the sentences, which is really quite interesting, but let me come back a little bit and say that this pure syntactical form of an egregore-
Patrick: Oh, I said grammar, not syntax. Different, different piece.
Jim: Yeah, I know, but I’m just saying if we think of it’s all word-based, that doesn’t actually speak enough about it. For instance, a egregore that BJ more than I, but I in a minor way have been involved with is the reform of gun carry rights in the United States over the last 40 years, and this has been an egregore. I went back and looked at the research this morning and the first state to have shall issue was Georgia in the late ’70s, and then it slowly started accumulating in early ’80s. Then after Florida in the late ’80s, it got faster and faster. So it swept through most of the country.
Then constitutional carry, this is the fact that you don’t need a permit of any sort. We always had one state, Vermont, that had constitutional carry, and then I think in 2003 it was Alaska and then I think Wyoming and Arkansas, and then it slowly snowballed and just saw that Florida’s become the 26th state to enact constitutional carry, meaning you can carry a concealed weapon without any permit at all. You could be a complete screw ball and have a long list of misdemeanor offenses, no felonies, and still carry a concealed weapon.
Then this has worked itself out both online, but also in the real world because it started back in the ’70s, so before the internet even existed or even the precursors to the internet. Then most importantly, it then had traction in the real world to get governors and legislators to write legislation and to enact it.
BJ: Well, I’d like to jump in and just say that you’re right. This is a longstanding trend, and I think a lot of the folks both on the left and the right in the gun argument are not fully understanding how much guns have won. I mean, here’s an interesting way to think about it. In the 1980s, the Brady campaign started, and the Brady campaign’s original name was something, I think it was the Brady campaign to end handgun ownership. That was their original focus was to try and ban handguns from private ownership, and now they can’t even get an AR 15 ban.
So how far back they’ve fallen is really pretty tremendous, and after 2020, I said this on the TFP podcast with open source defense, my gun writing group that I’m a part of, guns won. They won that culture war. They won, and the only thing that needs to happen now is everybody’s got to look up at the scoreboard. There was something like one out of every 12 guns that’s owned in the United States. Every one of every 12 gun owners bought their first gun during COVID.
Jim: That’s a little scary.
Patrick: Yeah, which psychologically is crazy, the idea that now I’m stuck alone with myself and no stimuli because I’m locked down, now I need a gun. Think of the psychology behind that decision making.
BJ: It’s rough, yeah.
Patrick: It’s crazy.
BJ: I mean, the increase in gun deaths during that was largely suicide related. I mean, there’s problems, and I’m very pro-gun writer. I write for Recall magazine, but there’s problems with guns and gun culture itself needs to resolve them.
Patrick: Yeah, of course.
BJ: Needs to work on resolving them.
BJ: That aside, that’s the broader defini- If we want to call that an egregore, that’s a broader thing, and what I’ve tried to focus on, the problem if you go that broad with it is that how do you analyze it. It becomes very difficult to put metrics onto it like what Patrick’s doing.
Patrick: Oh, that’s easy. It’s easier than it looks.
BJ: Well, okay, what you’d like to do is you’d like to focus on what’s specifically happening within social media and how that framework allows the rise of current thingism and see how that goes, and then you can … I mean, because for instance, all those people we were talking about earlier that went from leaving the house as murderer to not leaving the house as racist, two years later, they all had Ukraine flags on the … Right? So it’s the same group. So how is that egregore itself is changing to chase different things?
Patrick: That’s right.
BJ: Why is it chasing those things and can you predict what it’s going to chase next?
Patrick: Yes. Yes, we can. We can predict it.
BJ: So yeah, this is when I was last with you, Jim, we were like, “Well, I don’t know whether or not we could predict this stuff or not,” and then Patrick’s like, “Yeah, I’m working on that.” So he’s been working on it. So now what’s the fun part is we’re going to find out from Patrick exactly what he’s been doing and how and why.
Jim: Not quite yet. Let’s set this up a little bit further, which is when we say the current thing, the Marc Andreessen meme, now is the thing that organizes the current thing and the egregore or a meta egregore that coordinates other egregores?
Patrick: It’s an advertising budget. That’s really all it is.
Jim: Well, but in some sense, as you pointed out, one of your essays for a moment it was, what was it, wokery, I think, and then next thing you know it’s the impeachment. Those aren’t necessarily the same egregore. In fact, while they overlap a fair amount of personnel, one could argue that they are separate meme plexes competing for attention to become the current thing or to influence the narrative, which is the other name of the current thing, essentially.
Patrick: I used to think that way. I don’t think that way anymore.
Jim: Okay. Well, tell us about why you thought that way and how you think now.
Patrick: Yeah. So I’m going to be disagreeably reductive regarding gun culture. If we’re talking laws, then we’re still talking Broca’s area. Without words, you don’t have laws.
Jim: True it up.
Patrick: So words are really the driving function here. When we talk about current thingers, we’re not talking about psychologically sensitive people. We’re not talking about victims who are waiting to get trashed on by some whatever their phone and their TV tells them. If we take current parlance and try to describe it and such, we’re going to end up feeling pity for the current thingers, but that’s not what’s going on.
These are people who have a word preference. They’re very good with grammar. They’re very good with language. That’s not been observed, but all my research shows that’s true. For example, I can currently scrape someone’s entire Twitter profile. I can grab every tweet they’ve ever made. Then I do a word frequency analysis of every word they’ve ever said, and then I gather all those words and I put numbers on the frequency of that word. So that’s their home base.
Everyone has a linguistic home base. They have words they keep using. They have words they return to, and you can break this down by time. You can break it down by day or by month, whatever your timeframe is to identify what their linguistic home base is. If you take that linguistic home base and you reduce it down to an acceptable token size for an LLM, and I put that frequency distribution into LLMs and I say, “Give me the psychology of this person.” The LLM will actually give me a readout that will rival the FBI.
Jim: Yeah, you did one on me and I go, “Not bad,” and I’ve also done one. I created a system to take content or just use the public content if people are well known enough and do an ocean personality program.
Patrick: You could totally do that, right?
Jim: Easy how good it is.
Patrick: Now in this case, I just press button and I have your psychology. Typically, I had to have a conversation with you or I had to read your shit and that would’ve taken too long, which I don’t have the time for. No one has time for that. I press button, I have your psychology from the jump. You try to troll me on Twitter, I will have the psychology of your entire family in a single click. Do you really want to troll me on Twitter right now? You really shouldn’t consider this. This is not a good idea for you because I will understand all the things you like, a gun argument, the second amendment. Bro, I am a terrorist state right now. By getting your word frequency, I win. You don’t understand how deep I win. You’ll never admit it because American culture is full of people who think they’re important, but this is next level crazy stuff. This was typically reserved for the NSA. Now I have it. Now we all have it. So this is like, “Holy crap.” So let’s take this idea a little bit further, right?
Jim: Now, tell me what it means and how does that get you to … There’s one grand egregore versus multiple competitive egregore.
Patrick: Exactly, exactly. So it’s not that there’s a grand egregore or a lesser egregore. I’ve reduced it down to linguistic sensitivity. There’s a gradient, a single gradient. So what I can do is if … Well, let me step back a bit. If we’re talking egregores, then we’re talking about, let’s take the occult definition we opened up with, where we’re talking about people in a circle, they’re repeating some weird word and they’re hyping themselves up, they want this word to be real, they want it to be a thing, and they’re in their weird little chicken sacrificing corner doing it, and then they want to bring it to the public because they have media.
So they take this word, they put some divine symbolism around it, they christened it, they smash the champagne bottle on the boat, and that word is now holy, and then they ship it through the network. Egregore only works if you believe the words. If you do not believe the words of the egregore, then it’s not going to work. You get totally different behavior, but you can get behavior you can account for, of course. That’s not to say the option tree isn’t predictable in that regards, but for the egregore to work as was intended, as was designed, words have to be impacted. They have to work.
Now, how do you know an egregore works? How do you know that a person is under the influence of an egregore? Can you look at their behavior? You could. You could look at the way they moved their left arm. You can look at the way they skipped down the street, all these different behavioral metrics. You could do that. We may or may not have the technology to do so, but bro, I don’t need that. I don’t need these high precision gyroscopes and all this other type of metric collection. I don’t need that stuff. The way I determine that an egregore works is if a person repeats a word they’ve never repeated before because then I know the egregore is working.
Jim: Yeah, at low level at least, that is classic Dawkins mimetics, that a package of information has become replicated. That’s interesting. When I talk about wokeries, some people say, “Oh, you can’t say wokery. That’s a horrible thing to say.” I go, “I just said it, motherfucker.”
Patrick: Yeah, “I just said it. Fight me.”
Jim: Yeah, but one of the things they say, “Well, what does that even mean?” I say, “Well, here’s a tell. If anyone uses the word white supremacy to describe current America non-ironically, then they’re woke.”
Patrick: That’s the start, right?
Jim: If they refer it in a positive way to the main themes of Kendi’s anti-racism, How to be an Anti-racist or even worse, Robin DiAngelo’s White Fragility, then they’re woke.
Patrick: They’re pretty much are.
Jim: Those are pretty good tells. Now, whether there’s precise words that you can also track beyond that, I imagine, imagine so. Also, if you use the word equity in this new perverted sense, that’s more than just the token because the word equity is an ancient honorable word that means good things, but it has been essentially repositioned by Kendi and friends to mean more or less the exact opposite of what it originally meant. I don’t know how quite you would get that sense. So I suppose you could use semantic embedding perhaps.
Patrick: Oh, here’s the next thing. I don’t even care what a person says. I don’t even care what they’re actually saying. It’s gotten to the point where I can just use an LLM. If I want to, I call it NPC speed running, totally derogative term, but basically what I’ll do is if I see some CEO or some hot shot on LinkedIn, and by the way, LinkedIn is the most hardcore social media possible because if you make one mistake on LinkedIn, not only do you get banned, but you lose your job and you lose your income. So it’s like the hardest hardcore social media game in the world right now.
Jim: It’s the most banal bullshit also.
Patrick: The banality is because HR exists in everyone’s mind in LinkedIn, right?
Jim: Yeah, that makes perfect sense.
Patrick: So what I do is I get someone’s … Someone wants to showboat something, they’re feeling brave on LinkedIn. So I just copy all of it and I go to ChatGPT and I say, “Give me a response to this that makes the person feel validated,” and then I just paste whatever they posted, and then it gives me something and I just take it. I don’t read it. I just reply with it, and then they want to friend me.
Jim: That makes perfect sense. Of course, today, you could automate that with a little Python program that-
Patrick: … which I’ve done. It’s automated. It’s fully automated.
Jim: Yeah, send it to GPT, bring it back, post it, boom, right?
Patrick: So this goes back to the idea of language. In order for me to speed run these CEOs, I’m using language. I’m using words. I’m not even using novel words. In this case, I’m reflecting their word frequency back at them. I don’t care what’s actually being said. It doesn’t matter what’s actually being said. It’s about the word frequency distribution.
Jim: I don’t know. Actually, now think about it. LLM is not going to reflect the exact words. It’s going to react to the latent semantic embedding that was in the seed. It’ll use somewhat different words, but it’ll be attempting to match the position of the other words in the semantic space. That’s a significantly different concept.
Patrick: I would then say that the LLM is not introducing novel words. I’ve never seen it introduce a novel word when I do a speed run operation. So it’s going to introduce a lot of words that are already stated, and any word it does introduce is going to be observably neutral. It’s not going to be like, “Oh, I’m showing off my word cell brain. I’m a big language guy. Look at me.” It doesn’t do that at all, which is why I trust it.
Jim: It’s very banal. It’s like talking to somebody who’s been to therapy for 10 years, then got a job at HR or something, right? The lowest common denominator speak possible.
Patrick: So I can trust its word frequency distribution, which brings me back to my original point. It’s about the word frequency distribution. If a person is under the influence of an egregore, I will notice there’s a sharp change in their word frequency distribution.
Jim: If it’s a new egregore because we’re all under the influence of egregores.
Patrick: Can’t stop it.
Jim: We’re all parts of probably a dozen or more egregores.
Patrick: Right. Now, there are certain people, when you look at the numbers, you’ll find that there are certain people who rapidly acquire new words as part of their personality. There are actually people who do this where they’ll be, “Oh, new word, that has to mean something. New word, that has to mean …” It’s a neurosis, actually, this grammar neurosis that some people have. You can actually plot over time the drift. You can say, “Okay, this person has acquired a new word they’ve never used it before, a new word they’ve never used before, new word …” you’ll see it was startling regularity. You’ll see this person does it every Sunday, this person does it every Friday, this person does it every month. There’s a velocity going on here.
Jim: A velocity, a velocity of adaptation of new words or I’d argue words, yes, that may be a first order thing that you can measure easily, but picking a new space in cognitive space, in semantic space. So they picked up a new idea. White supremacy in the Robin DiAngelo concept, even if you don’t use the word, if you’re pointing at the concept, then something has changed in your brain. BJ, looks like you’re wanting to get in here with something with Pat, and I’m nerding out a little bit.
BJ: Well, I’m actually a little bit curious. So one of the things that we had talked about before, Jim, when we were saying, “Hey, look, is there a way you could do science on this is how you would do it?” and it seems like Pat’s got some ideas with his NPC speed running thing. Pat, have you sat down and tried to put together a legit experiment about this or are you just goofing off with the procedure like, “Let’s build these 20 NPCs, have them go on Python, see what they do, figure out which ones work better than others,” that kind of thing. Is there a way that you can do any legit, I don’t know, I mean, pseudoscience, but is there a way you could set up, establish a hypothesis, test it via your speed run operation, that kind of thing?
Patrick: Yeah. So right now, I’m doing red team, blue team analysis on speed running, meaning there are people who want to be persuaders, and there are people who want to be victims. They’re signing up because they think it’s funny. So what we do is we look at all of the different types of disinformation, cognitive attack space. I found this beautiful collection of 401 pages of all the different ways you can be dishonest in a conversation and ways you can conduct what would be considered information warfare.
I found a collection of it. It was beautiful. So I turned it into a database and I’m teaching people like, “Okay, if you want to do this attack, here’s ways to make it more effective and then here’s ways to defend against it.” So we’re building this red team, blue team approach to all these different attacks, and we’ve also found that ChatGPT is really good at that too. So even with its AI safety layer, there are ways to basically say like, “Hey, ChatGPT, how do I improve the efficacy of this deception?” and then it’ll tell you. Then you can say, “All right, how do I defend against these individual deceptions?”
What’s most fascinating that I’ve seen so far is that when you … and these deceptions are all being done to plant new words in people’s heads. That’s the crazy part. So the purpose of these deception, the reason you know deception works is because the person has a new word in their mind. That’s how we measure if the deception worked or not. So what we found is that there are infinite ways to plant a word in someone’s head. There’s maybe 13 ways to defend against it. Maybe, maybe.
Jim: Yeah, that’s interesting. I have watched a few words propagate over time. I guess, unfortunately, I’m well known. I think it’s my goddamn Wikipedia entry that I coined the expression snail mail, and I watched it slowly propagate over the world. It had a high fixed rate, but more recently, I saw Jordan Hall invent the phrase, the woke mind virus, and watch that spread. I was able to know who it spread via because I knew some of the close end vectors, and it is now gone all the way as well.
So we can see these word payloads, but again, I would continue to make the distinction between the word payload and the conceptual payload that it represents. While they’re very similar, they aren’t necessarily the same. You might see people talk about the woke virus without using the word mind, for instance. So in the Dawkin sense, these memes, the words get reordered, but they may still carry the same semantic payloads. I want to continue to make that distinction.
BJ: All right. Patrick, I’m going to poke you for more of the science concept here, what to do with the tools that you’re working with.
Patrick: They’re self-funding, by the way.
BJ: Oh, I bet. So that’s the next question. So there’s two questions, I guess, really. The first one is, do you feel as if you could at this point automate via GPT and Python a way to push a particular concept, to push an egregore concept into Twitter?
Patrick: Bro, I’ve been pushing egregore since 4Chan days. That’s a guarantee. I’m tired of it. I’m bored of it at this point.
BJ: I get it, but because of your 4Chan history and stuff like that, and I guess folks can look that up later after the podcast is over, but because of that, have you tried to set up an experiment where Python does it for you on Twitter? You see what I’m saying?
Patrick: So here’s where things get interesting. So what I do, I’m building the infrastructure to make that efficient. So I’m not actually doing it right now because I don’t have the targeting tooling. Typically, what I do is I will, when I really set my heart’s passion in a direction, I like to be thorough, and regarding the ability to conduct influence operations online in an autonomous fashion, yes, this can be done, and that’s not an ideal, that’s not pie in a sky thinking. This is a simple technical question, and because of the LLMs, before we couldn’t do it, we needed teams of people to go out there, understand, someone read their history, blah, blah, blah, get to know them. I don’t need that anymore.
So all I need to do is like I will press button, scrape history, press button, throw ChatGPT, press button, get response. Now I have your psychology, which means I know what you’re interested about because I have your linguistic home base. So once I have your linguistic home base, now I know how to speed run you. I know exactly what things you’re interested about. So then I’m going to hit you with the speed run. So now the bot runs in, talks about the thing you like. Maybe it’s being a troll, maybe it’s being positive, maybe it’s being negative, maybe it’s being all of these things.
We can be positive, negative. We can be odd about it. We can be all kinds of things. The LLM can generate all manner of response, and you can keep it short so it’s not as typical wordiness or we can train something in private and run it from there.
So now I’m speed running not just random words at you, but things that you are passionate about, things that you have actually demonstrated in public and have been willing to demonstrate in public, and then have been willing to demonstrate online that those are individual layers of confirmation that you’re passionate about these things.
So now I know where to start and I can totally work your ass just from a bot alone, and now what direction do I work you for? Well, that’s an operational question, isn’t it? The idea of what am I trying to do, am I trying to understand you for the sake of understanding you? Maybe I want you to feel cared about. Maybe the bot suspects you’re alone. We can make that assessment too. The LLM will tell me with startling accuracy how isolated you are. Yeah, there are prompts we can do that will actually tell that based on your word frequency. It’s incredible. People do not understand how fucking crazy this shit is. This used to take teams of CIA trained psychologists to do. Now it’s a button press.
Jim: Interesting. Now, it’s funny, I was talking to some political operatives about two months ago, and as you might imagine, they’re gearing up with very similar concepts for fighting the 2024 elections, where they expect that they will be personality aware bots. I don’t even think they had yet quite figured out that LLMs could make this very much less expensive.
Patrick: Yeah. They haven’t figured that one out yet.
Jim: They will. I mean, now in last four weeks, there’s just been this phase change, the sudden perception that there’s this alien that’s landed on the earth that has superpowers and that for a 10th of a cent per query, you can get the alien to do all kinds of things, which we’re not even sure all the things it’ll do. Imagine you’ve probably read in the last just couple of days this new weird thing that you can ask GPT4 to compress text down into an unreadable form that it can later understand. I mean, what the fuck? I mean, it’s amazing, right?
Patrick: Yeah, and also, we can think of a word frequency distribution as a compression and LLM fills into gaps.
Jim: Yeah, exactly, and that’s, again, one of its Broca area kinds of things, just like when we say, “Hey, I want to talk to the waitress about what I want to have for dinner,” Broca’s area fills in a lot of the details. It’s quite interesting the same way LLMs do. Let me dig in on something. I made some notes that I wanted to talk about is … Well, actually, before I do that, let me go back to an example of an amazing egregore that may have been manufactured or at least emitted from a single beacon, and that was QAnon. Have you ever gone back and looked at the history of the spread of QAnon, where it came from one or three or four people and propagated out and basically built itself an egregore of people who believes some of the most unbelievable shit imaginable?
Patrick: I have not done a word frequency analysis on QAnon yet.
BJ: You should. You definitely should because that’s a really interesting one, and it’s a really strong one too. You’re talking about different vectors. If you wanted to try and get into it, you want to speed run Marjorie Taylor Green. You definitely want to do that, right? You know what I mean? Well, it’s interesting you talk about it.
So in general, I want to go back to Jim’s position here for a moment that it’s not just the words, it’s also the concepts behind the words. One of the reasons why the woke egregore was able to catch on so much is that it’s building off of envy, but there’s something there. There’s something real and physical there, and what happened there was some real and physical stuff that QAnon was able to piggyback off of too. For instance, the fact that Jeffrey Epstein really probably didn’t kill himself, and the fact that nobody would say it, but everybody knew it. That meant that if you had somebody else that was saying that, you might listen to them, and what else are they saying? Oh, they’re also saying Trump is actually doing X, Y, Z or whatever the QAnon feed was saying at the time.
If you ever go back and read this, particularly the early Q dispatches because I read them because I was trying to figure out what the hell’s going on with it. They would sprinkle in stuff that seemed obvious and true or probably was obvious and true, and they would blend it with stuff that wasn’t to see how much stuff that wasn’t would get picked up and carried on. It was like a Patrick Ryan troll, right?
BJ: It was very masterfully done in the early days, but the neat thing about it is that even if at this point now, it’s on its own. It lives outside of those people who started it because if they stopped doing it and somebody else was able to pick up the map and run it, they could do it. For instance, you could train an LLM off of all the Q dispatches and then you could start releasing your own ones. You see what I’m saying? Then that would be an easy vector into manipulate a chunk of the population.
Patrick: So I’m actually going to backtrack on my own stance where I said word frequency was all that mattered. Q is different, but it’s not necessarily different in a manner that disagrees with my premise. I’m going to do some mental gymnastics here, folks, so get ready. Q carried a presumption of authority with it. So for those who read it, they assumed that, “Well, Trump is in the White House, so the executive branch is captured. So what’s happening is someone has invaded the executive branch and is leaking all the secrets.” That was the great appeal of Q that someone was in the White House doing all this stuff.
Now, for the most part, I’ve seen every 4Chan troll known to God. I’m a walking collection of these things. All the women love me, trust me, but when it came to Q, I didn’t start thinking there was something there until after the Vegas shooting. That’s when it went … That’s when it, “Oh, shit, something’s actually going on here.” People who read Q don’t go like, “Well, I’m on board. I’m ready to go.” There has to be a breaking moment. There has to be this initiation moment. For me, it was the Vegas shooting. We don’t even talk about the Vegas shooting, by the way. Not even the people who hate guns talk about the Vegas shooting. Think about that.
Jim: Well, it comes up every once in a while because people say it’s weird that no motive was ever established.
Patrick: No motive was established, and if you’re against guns, you you’d just be bringing up Vegas. 600 Americans got wounded in Vegas, 600. You’d think if I’m anti-gun I’d be bringing up Vegas every other fucking day. Nope. That is not brought up in any gun argument. That’s crazy to think about.
BJ: It is pretty wild. Now, I could tell you because this is, I mean, my primary writing is in the gun space, right? Gun folks generally aren’t that interested in picking it up because it is the one of all of the mass shootings that really needed to be done with an AR 15, right? All the rest of them you could have done with a pistol, but that’s the one that is like, “Well, okay, that guy actually used the AR 15 for what it’s supposed to be used for in terms of it’s functionality,” if you’re going to be a mass shooter and do it. Now, to be clear, that guy also owned a Cessna. He could’ve loaded an ANFO bomb into it and suicided into the thing and killed everybody-
Patrick: Could have done anything, could have done anything.
BJ: At the same time, the reason why the gun folks don’t bring it up because they’re like, “Ooh, yeah, that one was pretty rough,” but why did an anti-gun folks bring it up?
Patrick: Why don’t they bring it up? That should be an easy cultural win, easy cultural victory win.
BJ: The gun people behind closed doors are like, “I bet it had to do with why that guy did it,” and they’re just scared that the motivation-
Patrick: So the anti-gun people should be all in. That should be their guaranteed canard. If I was trying to go after gun people, I would bring up Vegas every second of every day. I’d make you feel fucking embarrassed, bro. I would crush you with Vegas.
Jim: Oh, I think there may be another possibility, and that’s that the victims were country music fans who were not from any of the aligned tribes of the anti-gun folks.
Patrick: Possible, but even then, the anti-gun folks know how to make fun of southerners. That’s pretty easy to do. They’ve done that since the ’90s. That’s a soft equation, but they won’t bring it up. So why does this relate to Q, right? So I haven’t made that connection yet. Weird, right? So there’s in the official Q lore, and I’m going to go way off the rails here so just reign me, in the official Q lore, where he shot his gun from was owned by a Saudi prince and MBS was in Vegas, and it was an assassination attempt against MBS. That’s the official Q lore, right?
BJ: That’s wild.
Patrick: Wild stuff, right? Wild stuff. Like I said, just reign me in if I go off the rails. So I’m looking at this and I’m looking at the shooting and I’m like, “Well, America’s second amendment’s over.” That’s my immediate thought looking at Vegas. Then that just got wrapped up super quick, super fast and I’m like, “Okay, something’s off here.” So I started looking at it, and then you look at the Q stuff and you see Q’s basically saying, “We’re going to wrap up all of MBS’s enemies,” before this shooting was even happening. He was like, “We’re going to wrap them up because they’re planning to do something devious.”
The Vegas shooting happens, and within a week, all of MBS’s enemies are literally executed. All of them down to the children are executed to the point where his primary rival in the Middle East was hung upside down for six weeks in a hotel getting beaten and robbed for six million dollars, six billion dollars in cash and ownership of Twitter. Funny, but that’s when I went, “Oh, God, this isn’t actually …” I started looking at Q. I stopped looking at it as a troll, and I started thinking, “This thing is giving heads up on signal analysis for geopolitical intelligence.” There’s a strong possibility this could be used for hedge fund trading or anything. I started looking at it as a early warning system.
Now, you’re not always in the context regarding that. You don’t always have all the information about what’s going on, but it predicted what MBS was going to do with incredible precision, and I was like, “Oh, there’s something here.” Now you don’t have to believe in it to come to that conclusion, but if you’re doing any type of financial trade or looking at it the way the world works or Forex, for example, Q was part of your daily analysis. You had to check in because this dude, he was onto something.
Jim: Now, let me switch to another direction entirely. I’d love to hear BJ’s thoughts on this is when you think about egregores often, not always, they had a origin cell and then of real thinkers, think of Christianity, JC and the boys, and then it becomes institutionalized and becomes mostly full of NPCs, which, for our audience, NPC is a non-player character, meaning the little bots that run around in computer games and stuff, and it’s also a Pat Ryan-ish slur on people who act like they are programmed game bots as opposed to real thinkers, but here’s the interesting thing.
From time to time, new thinking does get injected into these egregores as you point out. Q disappeared for a long time. He came back for a little while. Did he disappear again? I don’t remember. I don’t even know, but now there are new creators of content of Q, and I think in the case of Christianity, about 25 years after JC and the boys, Paul has his moment and basically totally reinvents and then subverts the whole intent of Christianity.
So there’s a live player in Sam O. Berge’s terminology that shows up from time to time. So the idea is that egregores usually have some founding thinkers. If we think about wokery, we can think about the Frankfurt School post-Marxist, but then we get new thinkers like Kendi and DiAngelo and a bunch of other ones, Kimberlé, what’s her name, et cetera, who add new fresh ingredients into the egregore while most of the people engaged are NPCs just spouting the lines, but there is this dynamic of real thinking that’s interacting with simulated thinking, simulated thinking, essentially just being quoting the party line from the egregore you’re associated with. BJ, what do you think about that, this idea of them being a mixture of simulated thinking plus the injection of the occasional real thinking.
BJ: Well, if you want to do woke history, you brought up Kimberlé Crenshaw is the person you were thinking of. Also, a big one is Patricia Bidol-Padva, those sorts of things. Now, you had these thinkers that would show up and they would sprinkle something in, but the question in my mind with that is, if Patricia Bidol-Padva had not invented the redefinition of racism, would somebody else have done the same thing? It feels in some ways there is. People jump into a gap of need or in the more modern sense where you have the modern social media meme space, egregores, what’s going on is that you have 10,000 people all yammering, and let’s say they’re all putting a different idea into, say, the woke egregore or the Q egregore or whatever else. Whatever is the most viral then becomes the most true.
So it’s not as if there are a couple of masterful people that are carrying this thing along, it’s more like whatever juice it needs, instead of it being one person crafts the perfect pathogen and then it takes over, it’s more like you’ve got a million pathogens, and which one’s going to win? So it’s more like an idea Darwinism thing that’s happened within these things. You see what I’m saying?
Patrick: That’s why the idea of N egregore never made sense to me. That’s why I say it was a gradient.
Jim: Interesting. Yeah. One of the ones I did study and actually did some quantitative analysis on was back around the 2016 elections. I was tracking the emergence of bigrams on the Donald subreddit, and I didn’t look deeper, but I did know that a lot of these bigrams, two word phrases, were being actually coined on 4Chan, 8Chan.
Patrick: Yes, that’s right.
Jim: Then they were bubbling up to the Donald, and then from there, Fox News would pick them up.
Patrick: Yeah, that’s right.
Jim: Then Trump would hear it on Fox News and would selectively repeat some of them, which would then go back through the cycle and be re-memized. So there was an example of a completely decentralized egregore that was essentially manufacturing memes going up through a stack and filtering, and then would come back and be rebroadcast by Trump and then work through again.
Patrick: Yeah, but that’s the Reddit model. That’s always been Reddit’s model. Reddit steals from 4Chan all fucking day. That’s all they do. That’s their entire model.
Jim: Q, on the other hand, was more of a point source that then diffused and then developed additional, I think, relatively dominant second generation voices that there are some, I don’t follow that stuff, but as I understand, there’s a dozen or so second generation Qs that get most of the attention from the Q nuts. So it sounds like maybe we have two different classes of egregores, one where it’s truly a group effort. Maybe the woke egregore is one of those, and things like Q, the original Q was a single point source, and then you have something in between like Christianity where you had the original point source, but then you have a big dominant voice in Paul. Then you have Saint Augustine come in and you have a few others, and then eventually you get Martin Luther for a whole fork, essentially someone forked the database and then ran with it. So there’s different patterns on how these egregores exist in the balance between simulated thinking, which is what they do most of the time and the occasional real thinking. BJ, why don’t you run with that one?
BJ: I’m still stuck on the idea, to be honest, of some of the stuff we were talking about earlier about science and testing, test the tools that you have. This whole conversation’s reminding me a hell of a lot of Scott Alexander’s piece, Sort by Controversial, which is famous. It’s the one where he introduces the idea of Shiri’s scissor. Are you with this, Jim?
Jim: Tell us about it.
BJ: All right. So this is a Halloween article in, I guess, maybe 2019, I think. It’s maybe plus minus a year. I can’t remember exactly when it came out. Might have been 2018, but the idea was, and this is a fictional piece, but it was written as if someone emailed it to him and he was just sharing it. The story goes that there was some AI researchers that had a company. There’s five or six people in the company, something like that, and what they were trying to do was they did stuff eerily similar to what ChatGPT did, but they were trying to come up with the most viral headlines.
So what they did was they took all of Reddit’s database and sorted it by popular and they built an AI to try and generate popular headlines. They were completely boring. They were things like, “Donald Trump is no longer the president. All transgenders are the president.” That was what it was generating. Then they’re going to get sort by down vote too, and if they sort it by down vote, then there would be some banal statement about penis pills.
When they are trying to train it against controversial statements because you can sort by the most controversial thing in Reddit because it’s got the tightest blend of up votes and down votes, right? So they built this AI to generate those statements. Then the lady who was trying to train it, she trained it against their internal network after they built it when they hadn’t trained it on Reddit yet, against their internal Slack or whatever. She was complaining they didn’t work. She’s like, “I can’t get it to work. It’s just spitting out stuff that’s obviously true.”
So they called a meeting and they looked through the code, and they’re trying whatever, and it takes about 10 minutes before … because they’ve got this statement, and it’s obviously the worst way to write this line of code. Only an idiot would write it this way, and it’s only about 10 minutes into the meeting before they realized that half the people in the meeting think this is obviously the correct way to write this piece of code. They get so angry at each other. They have this giant fight. They end up throwing shares. Half the people get fired, they leave, they turn around.
Then an hour after the big fight, it’s 8:00 PM, the other half of the company sit around going, “Holy shit, it worked.” So they’re like, “We have a super weapon. Call up DARPA,” because they were like, “We can sort any group of things, build a language model, find the most controversial thing and destabilize any group.” It’s a group destabilization tool.
Patrick: The scissors.
BJ: That’s the scissor. It cuts groups in half. So the rest of the story goes on about like, “Oh, well,” it tied into the fact that I think it was during the Kavanaugh hearings, and so everybody was either really pro or really anti-Kavanaugh at that time. It was really tearing the country apart. So you talked about how the Kavanaugh thing was actually one of the scissor statements that it erupted, but halfway through the story, they’re like, “They’re trying to sell the scissor to DARPA,” and he was like, “Well, I want a demonstration,” and the guy says, “Well, yeah, we’ll demonstrate on whatever group you want,” and he leans and he says, “Mozambique.” End of story.
Then the story has a link to, apparently, there was this civil war in 2017 or some date in Mozambique that the implication was caused by the scissor, right? That’s what this whole conversation’s reminding me of.
Patrick: That’s correct.
BJ: Because I want to be that general, and I want to look at you, Pat, and I don’t want to say, “Okay. Speed run these two people.”
Patrick: Speed run them.
BJ: So proof of concept because here’s the thing. If you could do that, say you’ve got the speed run thing figured out. Say you’ve got your bot for them. So you’ve got it LLM’d all together. Okay. So let’s pretend you have all those things in place. You should, by rights, be able to establish a way to stick your fingers into an egregore and move it.
Patrick: Yes. So I’m beyond that. I’m beyond impressing generals at this point. So let’s go back to the religion thing, going back to JC and the boys. Here’s a fun interpretation of an LLM. Did you know whenever you speak to an LLM, you’re talking to the dead?
BJ: That makes sense. I’ll buy that.
Jim: Certainly some of them are.
Patrick: The vast majority of the corpus of text, which is generated on an LLM is from dead people. You’re talking to the dead. You’re engaging in necromancy.
Jim: That’s cool, actually. I like that. I like it.
Patrick: Kind of cool, kind of fun. That’s a fun interpretation. So now, we’re talking to the dead. Now, there’s some live people influencing the LLM in giving it training data, but the vast majority of it is dead people. That’s interesting. That’s a fun little novelty, but let’s run with that logic a little bit further here. They’re dead and they’re influencing the future. Uh-oh. Does that mean they’re immortal?
BJ: Oh, I don’t think that’s any different than anybody who’s written a book.
Patrick: Oh, I would disagree. The thing about a book is that you’re very different because you have to put focus on the book. I have to give intentional attention to the book in order for it to have influence on me, but the thing about the weights of an LLM is that any word anyone says influences the weight distribution.
Jim: I know some people who are thinking about this as a business, which is to offer to create your chatbot for you, let’s say you’re getting old, say, “We’ll download all your Facebook stuff, we’ll download all your emails, anything else you’ve written,” and of course, most people haven’t written much truthfully, “and we will interview you for three hours and then we’ll transcribe that and we’ll feed it to the bot and your grandchildren will be able to interact with pop pop in perpetuity.”
BJ: That was a Black Mirror episode in 2017. They did that exact episode.
Patrick: That was a business model I worked for in 2008.
Jim: It’s now quite easily doable. I know people working on it.
Patrick: Yeah. So as a result, so the LLM is different from a book. You have to intentionally focus on the book for the book to influence you, but every word you say, every tweet you’ve made, even private messages, they’re ending up in the LLM as training data, which means you have this weird little niche you’ve carved in the weights, and those weights, even if no one’s calling you up directly influence the output of all other text it generates. So that’s why it’s immortal because even if you don’t intend to influence the outcome of the future, you do. It’s just the way the weights work. It’s just the way the matrix math works. So that means you’re immortal. You are immortal. Everyone here who’s listening is immortal. You cannot die anymore. You get to live forever through the LLM. You’re not dying.
Jim: Well, that’s good. That’s a good thing. All right?
Patrick: Could be, but now let’s go even further because I like going all the way. So now, you engage in speech to achieve immortality. Do you know what the full First Amendment is? It’s not just to write the speech, it’s the freedom to religion too. Now, religion traditionally holds the reigns on what is and what cannot be immortal. They have this whole gatekeeping system about immortality in and of itself, but now with the LLM, your freedom of speech guarantees your freedom of religion as has been dutifully noted hundreds of years ago, but the LLM, it’s materially achievable, meaning as you speak, you become immortal. Is the First Amendment guaranteeing the right to immortality? That’d be a fun question to ask.
Jim: When you could start a religion, you could call it the-
Patrick: Think bigger. Jim, think bigger. We’re not talking religion anymore. We’re talking about governments promising the right to immortality.
BJ: Hopefully not medical immortality because that’s tremendously expensive.
Patrick: No, no, no. No government would be caught dead promising that, but in terms of promising the ability to influence the future, governments can now promise that.
Jim: Yeah, maybe could. Truthfully, is it a terrible idea that there’s a constitutional right to have 25 hours worth of interview and then transcribed and then fed into the corpus? I’d vote for that. That’s somebody who’s really poor, who doesn’t have any visibility. They ought to have their 25 hours worth of discourse added to the corpus. I’m in favor of that. Put it on the platform. Pat Ryan’s immortality plank in the platform of the sensible people party.
Patrick: So now, immortality is a political movement. Religion, no, no, no, no. Religion doesn’t work in America like that. No, no. We’re talking like we’re getting the anti-religious people on board to immortality all of a sudden. No one’s ever done that before. They’ve been promising it with genomes and cloning and weird backwards manipulation, but now we actually have it. We can actually do it. It’s like the Kmart version of immortality, right? It’s not the best, but the general ideas here is I don’t need to impress a general anymore because now I can promise immortality to the world. Why impress a general?
Jim: It’s a soft immortality. I mean, do you care a lot?
Patrick: Hey, it’s the best we got.
Jim: It’s probably the best we got, right? Well, at least for the moment till we have brain uploading or something like that, which I think is further afield than people think, but anyway, let’s head on back. Oh, yeah, a key question, and I think this was probably more for BJ, but Pat may have the idea as well. You’re probably familiar with Scott Alexander’s idea of moloch, which is the go to the manifestation of late stage game A. In fact, frankly, it goes back a lot farther than that, where many players are engaged in a multipolar trap where everybody’s got to do bad things, whether do they want to or not, because if they don’t, they’ll get screwed by the guys who continue to do bad things.
BJ: Yeah, it goes back to tragedy of the commons really is where it is … It was the first identification of it, I think, as far as I’m aware.
Jim: Yeah, that’s a good one, but there are a lot of other ones, arms races being a classic, races to the bottom in ingredients and fast food. It goes on and on. How do you see the idea of moloch, which is getting a lot of attention at the moment again, as related to egregores and the battles of the egregores, and is it possible that they could be getting locked into multipolar traps?
BJ: So the easier application is to look at moloch in AI because, I mean, there’s a lot of people saying, “Oh, my God, we have to stop building the AI. The AI is going to kill humanity.” There’s a lot of folks that are in that camp and they don’t realize that moloch is building the AI, and there is no way that you can stop moloch from building it because if one company decides to stop building it, the other companies don’t necessarily have to, and it’s the same thing as the tragedy of the common situation. It’s the same multipolar trap. We are going to build this AI.
So that’s the really easy thing. The moloch of egregores really goes back to the moloch culture war itself. It was like egregores are expressions of culture within the new modern framework, but cultures themselves are built for culture war, and you know this because cultural Darwinism has weeded out the ones that weren’t built for culture war.
Jim: Most of them, not all of them. I was thinking about that this morning.
Patrick: The Quakers.
Jim: The Amish, right? They’re not trying to take over us, but they have been able to reach a homeostasis with the host of which they are an infection in.
Patrick: Bro, their woodwork carries them greatly. They charge six cents an hour to do woodwork labor, and they spit out exotic furniture for crazy markup. People don’t know that about the Amish.
BJ: Well, I was raised a Quaker, so I’m familiar with a lot. So it’s like the Quakers are not really going up or down, but they got a big shot in the arm during the Vietnam War because a lot of the hippies decided to become Quakers so they could be conscientious objectors, right?
Jim: Nixon was a Quaker, right?
BJ: Yeah. That was scary. He’s the one they don’t claim, but generally speaking-
Patrick: So was Eisenhower, actually. He was a Quaker too.
BJ: I didn’t know that. I thought Nixon was the only one.
Patrick: Yeah, Eisenhower was a Quaker.
Jim: No, I don’t think so. No, the other one that was a Quaker was Herbert Hoover. How about that?
BJ: Huh. I should go back and look at that, but yeah, generally speaking. So you’re talking about multipolar traps with egregores. The egregore has to keep battling for human brain space because if it doesn’t, you would get outcompeted by a different egregore that did battle for human brain space. So what you have with those is it is just like your race to the bottom or the tragedy of the commons, except instead of it being the common grazing land, which is where the word tragedy of the commons comes from, the commons in this type space is the attention economy.
So the egregore has to try and take over as much of the attention economy as possible because that’s where it lives. It lives in the attention economy of humankind, and it competes with other ones inside of it. So that is a multipolar trap, but whether or not it is how far down does it go, does it turn into an arms race, does it turn into … I mean, where that ends we don’t know yet. We just know that you’ve got multiple egregorical entities that are battling inside the overall culture war space.
In the early days, my thought was that whichever one could figure out how to update fastest was going to win because it’s going to be able to end around the other ones. Nowadays, I feel like updating too fast might be a flaw because if it updates too fast, it might update into something that is not efficacious in its own spread.
Patrick: Viruses have this problem too.
BJ: So if you could take the Branch Davidians or somebody like that, I mean, if you update into a corner, that corner might cause you to drink a bunch of suicide Kool-Aid or it might cause you to do whatever. So I’m not convinced that updating as fast as possible is the game maximal solution to be able to win the new culture war space in which the egregores are living, but what we’re going to see in the 21st century is a lot of egregore Darwinism where different versions of these things shake out is what I expect to see anyway the next 80 years.
Patrick: I see an apex predator personally.
Jim: Of course, in Pat’s technology, you could push the rate to whatever you want.
Patrick: Whatever you want.
Jim: As you were saying that, BJ, the thing struck me that, yes, as fast as you want is not the right answer. Here’s probably why, which is this grammar dispersion issue that Pat was talking about. There’s an impedance mismatched potential issue, which is you can’t be pushing new stuff into the egregore faster than it can spread and become fixed. I mean, one of the things we know in advertising, for instance, is that the first time you hear something, it doesn’t become fixed, generally. You have to hear something multiple times, and we also know if you hear it from multiple sources, more or less simultaneously within a two or three-day window, it’s more likely to fix.
So all that interaction between the meme space entity and the meat space humans is going to produce an impedance matching problem that you cannot exceed. So if one were to design the optimally viral egregore, you’d want to tune it such that it was as closely matched as possible to the human impedance rate for absorbing new material.
Patrick: Yeah, definitely.
Jim: That’s going to be a lot slower than a Python program.
Patrick: Regarding the impedance rate, let me backtrack. I said Eisenhower was a Quaker. That’s not true. I made a mistake. A non-Trinitarian, he was non-Trinitarian. My bad. Now, regarding the impedance mismatch, you ever read Three Body Problem? You ever heard of that book?
Jim: Oh, yeah, I read the book.
Patrick: So the Dark Forest Principle, egregores live in a dark forest, meaning, for those who don’t know what that concept means, if an egregore shows themselves, then every other egregore will pile down on them.
Jim: Don’t most egregores show themselves?
Patrick: Well, they have to show themselves, and that’s why what BJ was observing, if they adapt too fast, they lose. Well, that’s why. What happens is that they adapt too fast, the other egregores are tearing them apart.
BJ: Well, okay. Yeah. There’s a flip side of that though is that in order for an egregore to attack another egregore, it also has to show itself.
Patrick: Right, and so the egregore that wins the apex predator is the egregore you can’t even see,
Jim: Okay. How about this? The apex predator is not a first order egregore but a second order egregore that spawns egregores at the optimal rate, and they get taken down, some of them. Some of prosper, but the apex predator is invisible.
Patrick: That’s right.
Jim: He’s the spawning function.
Patrick: That’s right.
Jim: Just like in a computer game, right?
Patrick: Yeah, exactly. The generator doesn’t lose.
Jim: Yeah. There’s an idea. Let’s see if we can test that, right? Let’s see if we can build the invisible meta egregore that spawns egregores and the egregores are tuned to the impedance matching rate of humans, and we just try various things. Let LLMs invent each of the egregores. If you informed the LLMs of what an egregore is, it would be happy to invent a whole bunch of them for us. In fact, it would invent 50 at a time.
Patrick: I would then take the controversial position to say that the generator of egregores could be classified as a religion.
Jim: It could. That’s not a good thing, but it could be. All right, guys, we’re coming up here on our time. Some final thoughts. BJ?
BJ: I’m glad we had this talk. I’m still a little bit skeptical about Pat’s stuff. What I want to do is give Pat some homework and be like, “All right. Do Mozambique next.”
Patrick: Last time this went well, by the way. Well, I’m targeting religion at this point.
BJ: Good. Do pick a cult, pick a whatever, and say, “All right.”
Patrick: Oh, we got it.
BJ: We’re going to try and do this exact thing. Make it do this. Make it bark like a dog and wait and run it, and then see if you can get it to bark like a dog, and then that would be proof of concept to me. I would like to see that, and if you could see that, then, I mean, obviously the thing to do is for us three to put together an LLM dating app scam to become billionaires and then build the meta egregore and take over the world. I mean, that’s the obvious jump there we’d probably do in 10 years, but as long as the tools work. That’s my question because if, again, here you’re in a multipolar trap, if us three don’t do this to take over the world, somebody else is going to.
Patrick: Somebody else will.
Jim: They’re not going to be dearly as nice as us.
Patrick: No, not the slightest.
BJ: Of course not. You’re right.
Jim: That’s what everybody says, right? Hitler, Hess, and Goring are no doubt patting themselves on the back about what wonderful fellows they were.
Patrick: Yeah, I’m sure.
Jim: That’s human nature. I do have a possible simpler experiment because I do like to not try to boil the ocean in the first experiment. Suppose you just pulled a list of a thousand people who followed one of the more famous wokist. You picked the one, and then see how many of them you can get to follow one of your bots. That’s a straightforward mechanical process and you can have a metric, and especially somebody who has, let’s say, a hundred thousand followers, you could pick a thousand at random and then evolve your algorithm based on results and then try it again on another thousand. So you could see if you’re getting learning curve on your technology. I think that would be a more practical little experiment.
BJ: That would be one good experiment, and then the other experiment, do the reverse of that. So in that experiment, what you’re doing is you’re having one bot entity that you’re trying to get real followers. What you’d want to do is also run the reverse where you had multiple bot entities who are all, for instance, following Marjorie Taylor Green and all saying similar stuff to see if you could force her to bark like a dog or force her to espouse a particular phrase. Maybe not her, but any public figure, you could pick one on the left or the right that’s heavily tied up into social media and see if you can work them like a sock puppet.
Patrick: We were doing that in the 4Chan days, bro. I don’t need to prove that because we’ve already done it. We’ve had public people say our talking points many, many, many, many, many times.
Jim: You didn’t automate it with LLMs. That would be the big difference. That’s a big difference. We’re going to have to recruit an army of autistics versus pressing a button that says go.
Patrick: The thing is when you have automation, you don’t want to automate the thing you’ve already done manually. You want to automate the thing you couldn’t reach before. That’s what you want to do. That would be a good investment.
Now, final words, I think the entire Freakoutery regarding LLMs and AIs taking over the world, I think that is only a response to the managerial class. I don’t think the average person gives a shit. I don’t think the average person solves their problem with words. The welders, the plumbers, and the HVAC workers don’t give a shit about LLMs and stuff like that. They probably should. They’re definitely impacted by it, but their livelihood’s not going anywhere.
So I think as the vast majority of the LLM Freakoutery is basically a managerial upset. The lawyers are going to get replaced. The programmers are going to get replaced. The people who are good with words, the traditional managerial class of our species, they’re done. It’s over.
Jim: Well, I have made a prediction in writing in public that at least for an interim period, there will be a resurgence of a niche for liberal arts majors, people who are good with words because prompt engineering is going to be a very, very, very important interim technology, and that is not quasi autistic engineering personalities. That’s going to be liberal arts majors, people who are fluid in their thinking and have great ability to feel the words, shall we say.
Patrick: Yeah, I think that’s true.
Jim: Alrighty, guys. Well, this has been, it went all over the place, which is one of the things I was hoping for, and I really enjoyed the conversation. Thanks again, guys.
BJ: Thanks for the invite, sir.
Patrick: See you, Jim.