Semantic Lenia: Modeling LLM Inference as Continuous Dynamics

This framework treats LLM inference as a continuous dynamical system within logit space using a non-linear homeostatic feedback loop. The method enables the emergence of autonomous semantic solitons that maintain generative trajectories at the edge of chaos.