This site is dedicated to describing Large Language Models in terms on non-linear dynamics. This work is the basis of the Finite Tractus: The Hidden Geometry of Language and Thought.
The goal is to adddress the following points
Nonlinear Dynamics Perspective LLMs resemble dynamical systems, not stochastic samplers.
Recursion Failures Models enter self-reinforcing loops and lose coherence, indicating attractor dynamics.
Metaphoric Depth Models generate metaphor, they transcend pattern-matching. How and why?
JPEG Embedding Experiments Structured behavioral shifts emerge—not random degradation. Why?
The Chimpanzee Paradox If stochastic typing sufficed, how can we explain coherence, wit, and structure?
Lived Experience Daily use reveals behaviors that contradict stochastic expectations.