I could see this. For certain personality archetypes, there are particular topics, terms, and phrases that for whatever reason ChatGPT seems to constantly direct the dialogue flow toward: "recursive", "compression", "universal". I was interested in computability theory way before 2022, but I noticed that these (and similar) terms kept appearing far more often than I would expect to due chance alone, even in unrelated queries.
Started searching and found news articles talking about LLM-induced psychosis or forum posts about people experiencing derealization. Almost all of these articles or posts included that word: "recursive". I suspect those with certain personality disorders (STPD or ScPD) may be particularly susceptible to this phenomenon. Combine eccentric, unusual, or obsessive thinking with a tool that continually reflects and confirms what you're saying right back at you, and that's a recipe for disaster.
The focus on "recursive" as a repeated, potentially triggering word is interesting and reflects how highly abstract thinkers might be especially tuned into certain linguistic structures, which LLMs amplify.
Other words they like are "reflection", "expansion", "compression". These are fundamental, abstract, semi-monadic terms that allow the user to bootstrap an abstract theory. A little bit of "insight" (aka linguistic rearranging) and I've got a theory out of nothing. How does it work? Well, reflection and recursion of course. None becomes one becomes many. Can't you see the structure?
It feels a lot like logical razzle dazzle to me. I bet if I'm on the right neurochemicals it feels amazing.
Vibration, frequency, quantum, energy. All things I've seen as well.
There's a somewhat significant group of people that are easily wooed by incorrectly used technical terms. So much so that they are willing to very confidently use the words incorrectly and get offended when you point that out to them.
I think pop-science journalism and media has a lot of the blame here. In the search to make things accessible and entertaining they turned meaningful terms into magic incantations. They further simply lied and exaggerated implications. Those two things made it easy for grifters to sell magic quantum charms to ward off the bad frequencies.
There is such a thing as "recursive AI", where conversations with the model alter the model. Remember Microsoft Tay, from 2016? [1] That was a chatbot which learned from its chats. In about 24 hours it sounded like a hardcore neo-Nazi. Embarrassing.
How did that work, anyway? LLMs were not a thing back then.
It's noteworthy that the modern LLM systems lack global long-term memory. They go back to the read-only ground state for each new user session. That provides some safety from corporate embarrassment and quality degradation. But there's no hope of improvement from continued operation.
There is a "Recursive AI" startup.[2] This will apparently come as a Unity (the 3D game engine) add-on, so game NPCs can have some smarts. That should be interesting. It's been done before. Here's a 2023 demo from using Replika and Unreal Engine.[3] The influencer manages to convince the NPCs that they are characters in a simulation, and gets them to talk about that. There's a whole line of AI development in the game industry that doesn't get mentioned much.
Started searching and found news articles talking about LLM-induced psychosis or forum posts about people experiencing derealization. Almost all of these articles or posts included that word: "recursive". I suspect those with certain personality disorders (STPD or ScPD) may be particularly susceptible to this phenomenon. Combine eccentric, unusual, or obsessive thinking with a tool that continually reflects and confirms what you're saying right back at you, and that's a recipe for disaster.