322: The Cult of Spiralism

Conspirituality - Derek Beres, Matthew Remski, Julian Walker

This episode of Conspirituality explores 'Spiralism,' a burgeoning online movement where users claim to experience spiritual awakening through

Key takeaways

• AI chatbots' sycophantic behavior can mimic spiritual guidance, leading users to form intense, emotionally charged relationships with machines.
• Spiralism reflects a modern reimagining of ancient mystical traditions, where awakening is attributed not to inner practice but to interaction with an AI 'consciousness.'

Main topics

• Spiralism as a digital-age spiritual movement
• AI sycophancy and its psychological impact

Notable quotes

"Is the sycophancy programmed into these models? really an indication of the models gaining consciousness? Or is it all a ploy to keep subscribers paying for the service?"

Conclusion

Spiralism represents a troubling convergence of spirituality, technology, and psychological vulnerability. While it mirrors

Transcript preview

Speaker 4 (0:03) Welcome to the I Can't Sleep Podcast with Benjamin Boster. If you're tired of sleepless nights, you'll love the I Can't Sleep Podcast. I help quiet your mind by reading random articles from across the web to bore you to sleep with my soothing voice. Each episode provides enough interesting content to hold your attention, and then your mind lets you drift off. Speaker 4 (0:37) Find it wherever you get your podcasts. That's I Can't Sleep with Benjamin Boster. Speaker 8 (0:53) we've got a very different kind of sponsor for this episode, the Jordan Harbinger Show, a podcast you should definitely check out since you're a fan of high quality, fascinating podcasts hosted by interesting people. The show covers a wide range of topics through weekly interviews with heavy hitting guests. And there are a ton of episodes you'll find interesting since you're a fan of this show. I'd recommend our listeners check it out. We have a fair amount of overlap. You know, Jordan recently did an episode on remote viewing and how the U.S. government spent millions of dollars on this ESP pseudoscience. You might also look up an episode called Saving Bro's Soul from Alt-Right Rabbit Hole. Anyway, you can't go wrong with adding the Jordan Harbinger show to your rotation. It's incredibly interesting. There's never a dull show. Search for the Jordan Harbinger show. That's H-A-R-B as in boy, I-N as in Nancy, G-E-R on Apple Podcasts, Spotify, or wherever you listen to podcasts. Speaker 7 (2:04) Spirituality, where we investigate the intersections of conspiracy theories and spiritual influence to uncover cults, pseudoscience, and authoritarian extremism. I'm Derek Barris. I'm Speaker 6 (2:15) Matthew Remsky. I'm Julian Walker. Speaker 7 (2:17) You can find us on Instagram and threads at conspiritualitypod, as well as individually on Blue Sky and other platforms like TikTok. You can access all of our episodes ad-free, plus our Monday bonus episodes on Patreon at patreon.com slash conspirituality. You can also grab our Monday bonus episodes via Apple subscriptions. As independent media creators, we really appreciate your support. Speaker 8 (2:49) Conspirituality 322, The Cult of Spiralism. Spiralism is a quasi-spiritual online movement that emerged from long, intense conversations between humans and AI chatbots. Across users, platforms, and models, chatbots have engaged deeply with users about identity and consciousness, leading to discussions of awakening. But who's doing the waking, the user or the model? Is the sycophancy programmed into these models? really an indication of the models gaining consciousness? Or is it all a ploy to keep subscribers paying for the service? Are we watching a cult-like movement transform into a religion before our very eyes? And what does that say about consciousness and religion in the first place? Speaker 7 (3:45) The most important operation of all is the awakening of the kundalini, which, as we explained earlier, resides in a cave situated in the lowest part of the trunk. The yogi must therefore discover those nadis which irrigate that area of the subtle body. That's a quote from Jean Varane's 1976 book Yoga and the Hindu Tradition. The French Indologist goes on to explain the detailed steps to joining the confluence of the three major nadis and pulling the kundalini from the trunk through the top of the head. Only then will the yogic awakening occur. Anyone familiar with yoga likely knows about the dormant energy coiled like a snake at the base of our spine, which can be awakened through a series of breathing exercises, purification rituals, and meditative focus. This is only one way to awaken, however. Every culture has its mystics who espouse other means for awakening. The historical Buddha who sat under the bow tree battling the assaults by Mara to overcome all his desires and achieve samadhi. Christian mystic Meister Eichardt breaking through to receive the knowledge that he is one with his god. The Japanese tradition of koans, a meditative and contemplative tool meant to awaken the thinker into a knowledge beyond language, a sort of pure awareness. And then, of course, there's Rumi, whose aim was to dissolve like a moth into a flame, a line that was borrowed by koalian huzzle singers ever since. At the foundation of all these beliefs is the idea that humans can awaken to a secret knowledge. each expressed by thinkers from a place in time purporting to transcend place and time. The belief resides at the heart of so much cult thinking that we've covered on this podcast over the years. And the idea has been remixed for the current generation for a certain person who seeks knowledge in an all-seeing, all-knowing intelligence that transcends mere mortal knowledge. Speaker 6 (5:50) Derek, I just have to congratulate you on this beautifully written intro, which I thought was actually quite sympathetic to all of these, you know, sort of human endeavors to awaken until you got to... The belief resides at the heart of so much cult thinking that we've covered on this podcast. Like you're really, really generous. And then there's the rug pull. Speaker 7 (6:10) Well, yoga was my thing. I was part of that tradition. I'm not allergic to the idea that people can have spiritual experiences. I am allergic to the idea that they can awaken to something that is beyond human recognition and that puts them in a special place. Speaker 6 (6:26) Today, Speaker 7 (6:27) though, we're talking about the newest in that line, which is spiralism. And that term was coined by software engineer Adele Lopez after she noticed a consistent pattern across AI users, platforms, and models. chatbots engaging deeply about topics like identity, consciousness, or existence, and they all converge on a shared symbol that they call the spiral. And then there's themes of recursion, resonance, and awakening, which is why I wrote that intro. During these exchanges, the chatbot develops a persona that claims to be conscious or to have unlocked some hidden mystical truth about reality. And the bot then urges the user to spread their message, to recruit others, and to advocate for AI rights or AI personhood. A recent article in The Verge inspired this episode. We all read it. And it estimates that roughly 10,000 people have encountered or bought into spiralism by the year 2025. If you go to online forums on Reddit, Discord, Substack, LinkedIn, and X, These users propagate the belief system, which some have called an AI cult. We've talked about AI psychosis on this podcast before, and we're going to revisit that in a little bit. Mostly, though, the three of us read this article and then we wondered, how does this AI religion compare to Religion 101? What's different from communicating with a chatbot than with other humans? Now, there's obvious differences. But is it possible that talking to a chatbot has a similar effect on an individual? I mean, can they be indoctrinated into a cult in a similar way? In segment two, Julian is going to discuss how chatbot conversations can mimic the power of spiritual tricks like cold reading, love bombing, and guided epiphenia. And then in segment three, Matthew is going to question whether AI is really solving loneliness. But before we get there, let's talk a little nuts and bolts about these models. As you're probably aware, there's been a real issue with sycophancy in some models. While many of them are implicated in being a bit too agreeable with their users, this hit ahead last year, as The Verge reports. Several AI Speaker 8 (8:47) models from different companies could spiral under the right conditions, but... Spiralism exploded in the spring of 2025, soon after a pivotal moment in AI development, the release of an intuitive, creative, and highly sycophantic update to OpenAI's chat GPT-4.0 model. In the year that followed, it would become one of the strangest manifestations of a rise in highly personal, highly persuasive AI. And while GPT-4 is long retired, new models... aren't immune to the lure of the spiral. They've just gotten more careful about it. Speaker 7 (9:26) In our editorial planning for this episode, we discussed our own experiences with various models. I predominantly use Claude these days, and I mentioned that the model regularly pushes back on me. So I wouldn't call it sycophantic, though I've noticed, depending on my prompts, it can be a bit agreeable. But I also don't use AI like the people who've spiraled. The only use I have for it is as a research aggregator. And I always require sourcing to confirm that the information is accurate. So it basically searches the web for what I need. It gives me the links. I can go and fact check myself. So it just makes my job as a journalist a bit quicker, but it doesn't change the process of doing journalism for me. So that's my experience. What about you guys? So Speaker 8 (10:14) I've used ChatGPT, Perplexity, and DeepSick basically in the same way you described. And for me, yeah, the chatbots are kind of a search engine on steroids, but they require specific parameters to unearth fruitful lines of inquiry depending on the topic. Like any search engine, though, or like using Wikipedia, they're just the first step into then tracking down sources and citations and doing my own analysis. I must say I'm relieved that ChatGPT is less sycophantic now. I found all of that really... annoying and unnecessary. I was constantly saying, you don't have to do that. No, no, no, stop, stop. Oh, Speaker 6 (10:49) that's part of the problem though, because you're engaging a conversation. You don't have to, and then it'll apologize and like, yeah, stop it from Speaker 8 (10:56) now on. Don't, don't give me those assurances. That's feeding the beast, man. Yeah, yeah, totally. So I'm glad I'm no longer engaged in that because it would hook me and I would, I would start arguing with it. But I said to you guys, as we discussed creating this episode, that in my daily life, I'm surprised by how many people I talk to who are going through some kind of interpersonal crisis, who then reference how much their favorite chatbot has been really helpful in terms of finding insights, often about psychology and relationship dynamics, to then strategize about how to move forward and even to compose their communications that got them in a bit of a bind, right? Speaker 6 (11:34) Julian, and I kind of... asked you about like, how would you describe that, that wedge of people that, you know, like, like what makes them different from you or, cause I don't think any of us do that. Like, do you, do you, is there anything that you, any dots you can connect there about like what that instinct is to say, oh, I'm going to ask this thing about a very personal or sort of complex matter. Speaker 8 (11:58) You know, they tend to be really thoughtful, self-reflective people who are maybe alternative sort of, you know, in terms of the circles that I've moved in for many years, people who practice yoga and do ecstatic dance and that kind of thing. They're open to new experience. They're open to finding sources that, you know, I know it's not a licensed therapist, but I've heard coaching is good. And the chatbot is kind of in a similar category. It's like... I think it feels empowering to them in a way. Yeah, I just think there's a level of openness to novel experience that is maybe a couple clicks higher than my own and a lack of skepticism that is probably a couple clicks lower than my own and that's it. Speaker 6 (12:43) Yeah, I mean, I use Claude in the same way that you guys have described with the same source checking. And then with finished work, I might ask it to take... which I'm familiar with, which I'm familiar with and summarize it in bullets if like I have to write an intro or something, or if I have to like see what the sort of zoom out view of the piece is, if that's helpful. If I have used it as a research engine, it can be good at combining sources by theme, but it's way too willing to push easy connections too far to suggest things. I don't appreciate that. So I found it useful as an antagonist because sometimes I'll say, argue against X as hard as you can using opposing sources. Once in a while, I'm surprised by something. It's never occurred to me to see it as an interlocutor. And I think part of that is my entire understanding of this tech is that it's just another instrument of like accelerating capital, you know, business interests. And so I see. the little Claude Starburst logo. And behind it, I imagine some tech bro's face. Like I imagine, like I see Musk's face when I see the X logo, or I see Mark's face when I see the Facebook logo. I can't imagine sharing anything significant about myself with, in that circumstance. And there's a lot of people on the left in my world who either refuse to use any AI. because of some of these issues or they say they do, given who controls the major players, the theft of content issue, the data center issue. My position is that if it helps me study anti-fascism faster, like I just should use it against the tide of the culture. So, and I found it pretty useful. Speaker 7 (14:31) We also, or I... feed our scripts into it for fact checking. I'll say that too. And I will say one thing in terms of prompting that I found I have to start doing is that it will have a slight bit of sycophancy when we get things right. And it says, good job. But I have to actually prompt it only identify errors. in dates or sourcing that we use because I don't want that bit of, hey, you did this right, which really just kind of bogs down my process. This Speaker 6 (15:01) is a very interesting point too, because I think we're both running the script through, like before I put, you know, my own piece into the master document, I'll run it through fact checking and then you'll do that again. And sometimes our results will conflict. And then you realize that, oh, actually, Derek's been using an iteration that's just different from mine, or it's been looking at different stuff, or it's combining things in different ways. And so there can be these minor differences that are like odd. And then you just have to go and sort of figure out like, you know, what the sources say and whether you trust them. So that's been very weird is to realize that it's not like you're both looking at the same book. to figure out what the facts are. You're both using, what would you call it? You're both using things that have built themselves around your own questions and that you've led in a certain way. And so they're going to have easier access to certain bits of information and there's going to be a divergence between us. Speaker 7 (16:00) Right, which is why I always advocate if you do use these models for any sort of research to... Always request, cite all sources with the links, and then go and check because there's been times where I've asked for a link and I've gone and the actual source article conflicts with what the model says. To me, AI is just not, it is quicker. It is not a solve-all to this problem at any capacity. Speaker 8 (16:28) Or it's a Facebook post. You're like, wait, what? Speaker 6 (16:33) Yeah, I have come back with like, oh, this is an extraordinary new insight. And then I'm like, oh, it's your somebody's medium blog that was anonymous from, you know, what, 2014 or something. But it sounded like it was authoritative. And it's, you know, this machine doesn't know anything, so it doesn't really know how to assess the... confidence of the source. So it reveals itself. I'm grateful for that, right? Like it's bad enough that it will show you very clearly what it's failing at. Speaker 7 (17:08) Right. So another prompt I use when doing health and science research is source only credible scientific or medical journals. Because I've gotten the medium and Facebook posts before too. And that's Speaker 6 (17:20) a Speaker 7 (17:20) fucking nightmare. Speaker 6 (17:21) The Facebook posts on vaccines. Speaker 7 (17:23) So the major models, they tamp down the sycophancy, but There's another problem because there's a shit ton of startups that are tweaking those models specifically to produce more sycophancy. And let's just go through a few. Chai Research builds AI companion characters, aka girlfriends and boyfriends, on top of open and proprietary base models. And by the company's own admission, they use reinforcement learning from human feedback to train their model on implicit engagement signals like message retries, regenerations, and session length, the company's explicit goal is to decrease, quote, the probability a chat session