New paper: Dynamics of Agentic Loops in Large Language Models

This work proposes a geometric framework to analyze agentic loops in LLMs, showing how prompt design controls convergence versus divergence in semantic space. By calibrating embedding similarity against human judgments, it offers a rigorous way to study stability, attractors, and exploratory behavior in iterative AI systems.
#AgenticAI #LLMs #AISystems
https://arxiv.org/html/2512.10350v1