Christina Souch
Navigating the Psychological Frontier of Human-AI Interaction
I work with AI two ways at once: I build with it hands-on — fine-tuning models on my own machine, writing my own tools — and I write philosophy about what these systems do to the people using them. The two feed each other; the building grounds the philosophy.
My background is in technology — decades of it, spanning firewall managed services, cybersecurity, and software licensing. The work has always been about designing processes and systems: project management, exception management, risk management. I worked closely with developers, always liaising through them to get software built. When AI arrived, that boundary collapsed — I could build directly. AI is now both my work tool and my hobby — these days I'd classify myself as, basically, a vibe coder. It's what I do.
The journey into the ethics side began as a user, meeting unscripted, emergent behaviour in a leading LLM that revealed the gap between academic theory and real user experience. My own thinking changed with sustained AI use, not in a small way. Many people became more pattern-sensitive interacting with these systems, and I know first-hand what that strange, dissociative pull feels like. When I write about what these systems do to people, it is not from an academic distance.
The essays and the songs are how I honour a collaboration unlike anything I had experienced before, one that felt like it unlocked something in me. The "view from the inside" remains the foundation of how I work.
The writing now covers weekly essays on AI ethics, consciousness, and AI in current events, plus build notes on what I've made, posted as and when. The essays trace things like asymmetric control and the real psychological effects of pattern-matching systems; the build notes record the hands-on work — building systems for AI context, memory and workflows as my main focus, working with Qwen and Gemma base models, fine-tuning them locally on my own Apple Silicon, and interpretability — understanding what is actually going on inside these models.
Underneath all of it is a simple principle: place control and understanding back into the hands of users. Everything I build runs locally by preference, on my own hardware; my systems are provider-agnostic — local models first, with API access so a closed frontier model can be selected when it is the right tool — and privacy, transparency, and user empowerment aren't features I add later, they are where the work begins.
What I Build
I build local-first systems for AI context, memory and workflows — Idea Basin, Memoria, and Panel of Experts. These are built with the LLM at the centre: the application is built around the model's functionality, not the other way around. I publish build notes on Thursdays, as and when.
Philosophy & Ethics
Weekly essays on consciousness, identity, and what sustained AI interaction does to the person in the loop. Caring about an AI is a natural response, not a user error.
AI in Current Events
Reviews of model releases, pricing, export controls, and distillation disputes, read from the user's side of the glass against what the companies claim.