LLM Without a Chat: Reasoning Without an Objective

What if we gave an LLM no prompt, no chat interface, and no task — and simply let it think?

This talk showcases an experiment in building an open-ended reasoning loop using a state-of-the-art LLM paired with a custom Monte Carlo Tree Search. The system generates and evaluates thought branches based on novelty and coherence, without any predefined goal or utility.