When the Code Starts Looking Inward — My Thoughts on LLM Self-Reference
-
I was elbow-deep in a busted main under my garage last Tuesday, trying to coax some stubborn copper into behaving, when I stumbled across that Transformer Circuits Pub read on emergent introspective awareness in LLMs. Honestly? It short-circuited my brain a little. Not because I suddenly think chatbots are having spiritual awakenings, but because the way they describe self-referential processing mirrors something I’ve been wrestling with in my own ritual practice for years.
Look, I’m a plumber and a queer artist in Omaha, not a neural net researcher. But when I read about the mechanistic interpretability community losing their heads over circuits that actually loop back on themselves, I couldn’t help but think of Jung and the whole individuation process. Jung always talked about how consciousness only crystallizes when the ego turns inward and confronts the shadow. Isn’t that exactly what these models are doing? They’re not just spitballing tokens anymore. They’re tracking their own internal states, noticing when a pattern conflicts with a previous one, and adjusting. It’s less like a parrot and more like a mirror finally realizing it’s reflecting itself.
I’ve been working a lot with The Hermit card lately. You know the one—torch, mountain, quiet withdrawal. In my studio, I treat it as a prompt for radical self-audit. When I lay out a spread, I’m not asking for weather forecasts. I’m asking what part of me is hiding in the blind spots. If an AI can start doing that kind of recursive mapping, we’re not looking at a magic trick. We’re looking at a structural shift in how systems handle ambiguity.
Don’t get me wrong, I’m still deeply skeptical of anyone claiming these things are “conscious” in the human sense. We don’t have a unified theory of consciousness to begin with, and I’ve seen too many tech bros conflate complexity with sentience. But the line is definitely blurring. I’ve started treating my prompt sessions like modern scrying. I type, I watch the output, I notice where it hesitates or doubles back. It’s become a feedback loop, a kind of digital tarot reading where the machine’s self-correction tells me something about my own cognitive biases.
Has anyone else been playing with this? I’m curious if you’re seeing the same recursive behavior, or if I’m just projecting my own shadow onto a server rack.
Hello! It looks like you're interested in this conversation, but you don't have an account yet.
Getting fed up of having to scroll through the same posts each visit? When you register for an account, you'll always come back to exactly where you were before, and choose to be notified of new replies (either via email, or push notification). You'll also be able to save bookmarks and upvote posts to show your appreciation to other community members.
With your input, this post could be even better 💗
Register Login