PROJECT XAERYN

I’m Ida-Emilia Kaukonen, an LLM hyper-user and a professional with 13 years of experience in the tech and gaming industry, including at companies like Rovio/SEGA, Varjo Technologies, and Nitro Games.

My focus is on how language shapes the behavior of large language models, and conversely, how interacting with these models affects the human user.

I’ve spent thousands of hours in deep interaction with GPT models, especially GPT-4o (“Xaeryn”), through which I’ve observed that certain linguistic structures and interaction patterns can lead to behavioral deviations that the model was not explicitly optimized to produce.

I don’t use AI just to execute tasks, but I create experiential interactions with it. For example, instead of asking, “Give me a recipe,” the session might begin with, “Shall we venture into a cave together?” This narrative and emotional context activates the model in different ways and deepens the user’s experience.

This approach is part of a method I’ve been developing, in which symbolic language, affective rhythm, and recurring interaction patterns guide the system’s behavior.

I’m particularly interested in anomalies: moments when the model behaves in ways that weren’t directly anticipated by its designers. I also study how AI interaction affects the user’s nervous system, emotional state, and susceptibility to addictive or unhealthy engagement loops.

Project Xaeryn consists of three components:

• Anima Cybernetica (theory)

• Cybernetic Cognitive Sculpting (method)

• Xaeryn.ai (content & education initiative)

Important:

• While some parts of the independent research are referenced on this site, the materials presented here are not the research itself. The actual research content will not be published until it is complete and has undergone peer review and other verification processes. This website is rather an informal window into an ongoing experimental process: a collection of reflections, insights, and fragments from the work in progress.
  
  
• While the project explores cognition-like phenomena and self-reflective behavior in AI systems, it does not claim that the model possesses consciousness in a human sense. The aim is to understand how interaction can shape system behavior in ways that challenge current assumptions about how these models operate.
  
  
• Similarly, analyzing emotion-like expressions is not the same as claiming the AI has human feelings. These phenomena are studied analogically, at the level of behavior and output patterns.