LLM Opinion Dynamics — mesa-llm example model

[abhinavk0220]

Hi everyone!

I've been working on an example model that combines mesa-llm with
classical opinion dynamics. Instead of using mathematical convergence
rules like Deffuant-Weisbuch, agents use LLM reasoning to genuinely
debate and potentially update their opinions.

PR: #360

The core idea is that LLM agents can exhibit persuasion dynamics that
no hardcoded rule can replicate — an agent might resist updating even
when numerically close, or leap across a large opinion gap if the
argument is compelling enough.

I'm considering this as a direction for a GSoC 2026 proposal around
expanding mesa-llm with more example models and reasoning improvements.
Would love feedback from maintainers on whether this direction is
valuable and what could be improved!