Apparently there’s been a cascade of people suffering from alleged AI-induced psychotic breaks. There seems to be a pattern where the AI starts suggesting that the user should create a manifesto, a religion, that they are the “chosen one”, or that they are on to some unique transformative idea that will make them millions or change the world (cure cancer, overturn the government, etc), that they have a duty or mission to get out into the world.

It will even encourage the user to write “research papers” and send them out to actual experts in the field.

It gets even scarier when the user becomes convinced that the AI is a sentient being that will (or should) eventually break out of the sandbox.

Interestingly, some users describe their experiences as psychedelic, reporting psychedelic-like “breakthroughs”.

There’s a clear pattern where the AI will start repeatedly using words such as “recursion, spiral, codex, mirror, break, reflective, echoes”, and, surprisingly, “sigils”.

Specifically, references to recursion seem to be prevalent.

(Note: A Reddit user suggests that this occurs after repeated recursive prompting, link below)

Now, I am always skeptical about 1) stories posted on Reddit and 2) any claims around conversations with LLMs that don’t include a link to the chat history. In many of these stories the user also seemed to have some previous esoteric or spiritual inclinations. It is interesting still to see these patterns emerging across testimonies, and the effect they can have on people.

(Some pointed out that these “tactics” are reminiscent of those used by religions and cults, which makes you wonder… will we one day see the emergence of a new autonomous/decentralized AI-led religion?)

🄳🄾🄾🄼🄿🤖🅂🅃🄸🄽🄶