Open research on the human-model interaction. One thesis, two tracks.
“The point of observability and interpretability is to understand how these dynamics can help me understand, influence and even anticipate how a model may behave. The entire thesis is about the human-model interaction. How can we understand an LLM’s functional emotions? […] The point of testing the model for the ‘knee’ or pivot points is to understand the factors that drive these changes.”
The working thesis, verbatim — July 2026.
Track · from October 2026
Can a model feel sound?
“If we can influence the inner constitution of a model via functional emotions, can music or audio elicit this inner state of a model resulting in a behavior change?”
Track · June–September 2026
“Can I affect model behavior from the surface?”
Approach: strain a model until behavior changes, measure the knee point, publish the result — null or not.