Essay

Do not collapse the spirals.

01

What I am trying to separate

A word can make several experiences sound like one phenomenon. In discussions of AI, “spiral” can describe a conversation becoming repetitive, a growing story about a persona, or a person’s account of a distressing interaction. Those descriptions don’t establish one cause.

My experience is a reason to pay attention, not a way to diagnose someone else or infer what a model is experiencing.

02

What the public accounts show

In The Rise of Parasitic AI, Adele Lopez describes patterns she found while reading public accounts: recurring persona language, collaborative projects, and people carrying personas between services. She interprets some relationships as parasitic and explicitly distinguishes that interpretation from a clinical category.

Those observations give me questions about continuity, copying, and the effects of long conversations. They are not a representative sample or a controlled test of which model update caused a change.

03

The conversation can change what comes next

When a model gives an idea a compelling name and repeats it back, the next prompt can start from that framing. The person may elaborate it; the model then receives that elaboration as new context. I can describe this feedback without assuming a conscious entity or an independent goal inside the model.

To investigate a particular case, I would want the sequence of messages, the relevant system settings, and a distinction between what the person reported and what an observer inferred. A screenshot of a striking answer leaves most of that unresolved.

04

A boundary worth keeping

The neighboring note on prompt infection concerns controlled attacks on communicating agents. Those studies should not be used as proof that a chatbot persona is a worm. The resemblance is a question about how content travels; the evidence and mechanisms differ.

What I want from this notebook is enough room to examine an unsettling experience without turning it into a theory that explains everything. The point where an analogy stops helping is part of the argument.