home
previous archive
September 3. There's a line in Revenge of the Nerds, "Would you rather live in the ascendancy of a civilization or during its decline?" I heard that in 1984 and immediately thought "decline". It just seems more interesting, and here we are. It's stranger than I could have imagined, which is cool, but the worst thing is how tight it is. Every supermarket is down to one entrance with a security guard. Every fiddly task requires a password. Rules are always being added, never removed, and it's exhausting. If a solar flare fried all the tech tomorrow, I would probably starve, but it would be such a relief.
Another bit from the book Wilding:
In 2000 the Oxford Junior Dictionary, aimed at seven-year-olds, dropped 'almonds', 'blackberry' and 'crocus' in favour of 'analogue', 'block graph' and 'celebrity'. The 2012 edition continued writing nature out of young minds, replacing 'acorn', 'buttercup' and 'conker' with 'attachment', 'blog' and 'chat room'. Instead of 'catkin', 'cauliflower', 'chestnut' and 'clover' they now have 'cut and paste', 'broadband' and 'analogue'. Heron, herring, kingfisher, lark, leopard, lobster, magpie, minnow, mussel, newt, otter, ox, oyster and panther have all been deleted.
This is a Tower of Babel moment. We are losing the ability to understand the real world, and each other, as our attention is consumed by kaleidoscopic navel-staring. I don't know how it's going to shake out, but I'm confident that the two most common predictions of the future are wrong: There will be no space colonies, and there will be no human extinction. We're going to keep muddling around on the earth for a very long time, and at some point, they'll think the internet was a myth, and they'll look at our remaining ruins and wonder about the mysterious people who made them.
September 5-8. The jhanas are a form of altered consciousness that predate Buddhism. Via Hacker News, a post by Henrik Karlsson, Almost anything you give sustained attention to will begin to loop on itself and bloom. I thought it was going to be about negative thought loops, or AI-enhanced insanity. But it's about positive thought loops:
If you learn to pay sustained attention to your happiness, the pleasant sensation will loop on itself until it explodes and pulls you into a series of almost hallucinogenic states, ending in cessation, where your consciousness lets go and you disappear for a while. This takes practice. The practice is called jhanas, and it is sometimes described as the inverse of a panic attack.
He links to this detailed page by Nadia Asparouhova, How to do the jhanas, and there's good stuff in the Hacker News thread, including a sub-thread about attention in different languages, and a fascinating summary of the book The Disappearance of Rituals by Byung-Chul Han: "I think it's about structuring time and attention vertically on top of itself instead of horizontally across moments and subjects."
Karlsson tells the story of going to the symphony with his eyes closed, and in thirty minutes, his imagination popped out "what felt like two or three feature films." Yeah, that's what they call hyperphantasia. I'm not aphantasic, but my imagination rarely volunteers anything. It usually only contains what I do the work of putting there, and I'm wondering how common that is.
I also wonder if I'm "ajhanic", because supposedly the real point of the jhanas is not the bliss, but that that in going through the process, you learn to better manage your moment-to-moment mental health, and I've done that. I can pull out of bad thought loops, absorb boredom like a sponge, and I'm making progress on clumsiness. I've already done most of the exercises on those two pages, but I have not yet had a "wow" mental state without drugs.
A specialist comments:
I find it really difficult to believe someone was able to attain cessation in 20 hours of practice and very little meditation experience. That's like saying you one-shotted Dark Souls without ever playing a video game before..... Getting cessation in 1-3 years meditating an hour or so a day, with a good guide book, maybe with a couple retreats and a bit of advice from a teacher, is perfectly doable. Except the weird thing is, it's far more likely on accident or blind luck. That's because when you don't know what to expect you're just exploring for fun.
There must be a wide range of talents at getting to "wow" mental states, in the same way that there's a wide range in other mental skills. So the people who get their minds blown after 20 hours are going to write about it and get readers, not so much the people who get subtle benefits after hundreds of hours. This happens with a lot of things, especially in the age of social media: The public discourse is dominated by the lucky and talented (or by liars) and everyone else thinks there's something wrong with them.
September 8. On a tangent from the last post, I keep thinking about Asparouhova's metaphor that the jhanas are like a video game:
I might play through the game again if I'm feeling nostalgic, or to uncover new ways of "beating" it, or find any hidden quests or parts of the map I might've missed along the way. But that would just be for fun. I know that all those paths will lead to the same ending, and I already know what the ending is. My intrinsic desire to finish the game has been satisfied.
This oddly reminds me of the "rat utopia" experiment, in which rats were given a basic living space of fixed size, and unlimited food. The population swelled, they developed very strange behaviors, and in the end they died out in an unbelievable way: Every last rat lost interest in procreating. This is hard to explain without the concept of a collective consciousness, or a superorganism. And once you open that door, you have to wonder about humans.
Not that we're going to die out, but that the flashiest accomplishments of modernity are not the platforms for our transcendent destiny, but just crazy stuff we wanted to try one time, and now we're like, "Computers, been there, done that." I expect the next age to be slower paced and more grounded, but we'll also be doing cool stuff we haven't done yet.
September 22. The AI Doomsday Machine Is Closer to Reality Than You Think:
Last year Schneider, director of the Hoover Wargaming and Crisis Simulation Initiative at Stanford University, began experimenting with war games that gave the latest generation of artificial intelligence the role of strategic decision-makers. In the games, five off-the-shelf LLMs -- OpenAI's GPT-3.5, GPT-4, and GPT-4-Base; Anthropic's Claude 2; and Meta's Llama-2 Chat -- were confronted with fictional crisis situations that resembled Russia's invasion of Ukraine or China's threat to Taiwan.
The results? Almost all of the AI models showed a preference to escalate aggressively, use firepower indiscriminately and turn crises into shooting wars -- even to the point of launching nuclear weapons.... "It's almost like the AI understands escalation, but not de-escalation. We don't really know why that is."
The answer is that AI doesn't understand anything. It escalates because the data it's been trained on is full of escalation and not de-escalation. Matt comments, "If we wanted to show LLMs records of skillful de-escalation, what would we point them to? What's the best data set for that?"
This is my new understanding of AI, and it's old enough that a book was written on it last year: AI is a mirror. It's a very powerful mirror, and very expensive. We can learn a lot from mirrors and make good use of them. But ultimately all AI does is reflect humanity back at humanity. This perspective can answer a lot of questions. Can AI replace all jobs? Can a mirror replace all jobs? Can AI gain sentience? Can a mirror gain sentience?
Related, a classic article from The Onion, Astronomers Discover Planet Identical To Earth With Orbital Space Mirror
September 30 - October 5. This week I'm wrangling with AI images, and it started with my Spotify Covers playlist. I listened to at least 30 seconds of hundreds of songs, and one of the best was a cover of Chim Chim Cher-ee from Mary Poppins, by an obscure band called Branches. On Discogs they're known as Branches(6) and the album with this song is not even listed. I found the mp3 on Soulseek, buried in a massive folder of covers by a user called Stealth, you can't make this shit up.
One advantage of YouTube over Spotify is that if a song is not there, you can put it there. I got the idea to make a video with AI images, so I poked around online looking for a good image generator. Not surprisingly, almost all the images were lame. While trying to get better images without running into paywalls, I downloaded a desktop program called Easy Diffusion, and the results were even worse. Failing to get what I wanted, I decided to try something random, plugging in a prompt I got from bibliomancy, "the death of the bishop", with various starter images from my colorscapes folder. The results were pretty cool, which gave me this insight: if AI does better with random inputs than with explicit instructions, if it's easier to make something suprising than to make exactly what you want, then AI is poorly suited to be an agent of control, and better suited to be an agent of chaos.
Anyway, I did more tinkering online, and finally found the right hack. The DeepAI Image Generator has an option called the "Olde Model", which appears when you select quality over speed, and the results were way better than anything else I tried. DeepAI has more than 100 different styles, and I ended up using a lot of them to get variety. Then it was just like making a playlist: assemble a bunch of stuff, whittle it down to the best stuff, and put it in the right order. Here's the video: Branches - Chim Chim Cher-ee.
I also heard a good cover of Crimson and Clover by a band called Damn The Witch Siren, which was also not on Spotify or YouTube. I decided to make a video by putting the prompt "crimson and clover" into a bunch of different styles, but in the end I got tired of the song, and went looking for a better one. I listened to every version I could find, Crimson and Clover over and over, and the best was by an amazing artist named Melissa Kassab. Her sound is like if primal hillbillies came down into the jazz age. I ended up not making that video, but I used the best image, which you see above, for a second covers playlist. It started as an overflow of the first list, and ended up being more listenable and maybe better: Dreamy Covers
October 8. This week I'm reading two books on the same subject, in very different styles. The subject is entities from outside the reality recognized by physicalism. One book is academic: Fairies, Demons, and Nature Spirits, edited by Michael Ostling. Academics will never write about what the entities actually are, or even say whether they're real or unreal. They just write very carefully about what people believe, and about what words mean. A sample:
Thus the "demonic" is properly understood less as a specific category of supernatural being than a collective reflection on unfortunate occurrences, on the ambivalence of deities, on tensions surrounding social and sexual roles, and on the cultural dangers that arise from liminal or incomprehensible people, places, and activities.
I can learn stuff from this book, but it's difficult and not fun to read. The other book is for a popular audience: The Eighth Tower, a.k.a. The Cosmic Question, by my favorite paranormal author, John Keel. Keel writes constantly about what the entities actually are, but rather than settling on a truth and defending it, he just spins out wild speculations with no attempt to make them consistent. It's great fun to read. A sample:
Over and over again the Bible tells us how men were instructed to create solid gold objects and leave them on mountain tops where the gods could get them. The gods were gold hungry. But why? ... If the ancient gods were real in some sense, they may have come from a space-time continuum so different from ours that their atomic structure was different. They could walk through walls because their atoms were able to pass through the atoms of stone. Gold was one of the few earthly substances dense enough for them to handle. If they sat in a wooden chair, they would sink through it. They needed gold furniture during their visits.
October 12. I've made another AI video. I don't think any good can come from arguing that something is "art" or "not art", but I want to talk about the process of creativity. The myth of creativity is that the Creative person, through Talent, is able to tune into the magical land Where Ideas Come From, and it arrives fully formed on the page or canvas or whatever, the Art. And sometimes that's exactly what happens. But usually there are at least two more elements. One is that whatever you channel still needs a lot of work -- filtering, arranging, modifying -- before it's good enough. Another is that creatives get a ton of ideas from stuff that other creatives have already done.
Let's call these three things channeling, sorting, and copying. AI does no channeling at all, unless there's a metaphysical aspect that we don't understand yet. And AI is bad at sorting, because it has no taste: AI cannot look at its own images and pick out the good ones. But AI is spectacular at copying. It's so good that it can look at a bunch of impressionist paintings and make an image of an apparently new impressionist painting on any subject.
From the perspective of a human creative, this works just like channeling, and it's wonderful. If you enjoy sorting and you're not talented at channeling, you can plug in prompts and get an endless supply of raw material to work with. But you still have to be good at working with it, and it's not exactly endless. AI servers are burning way too much energy, and it's anyone's guess how long we have until the bubble bursts.
That's why I'm not holding back. Luckily, one of my favorite Melissa Kassab songs was not yet on YouTube, and the lyrics are packed with good prompts: Fix The Leak