The Hugging Face blog features a second entry in the Thousand Token Wood series, where five simulated AI labs compete in a financial drama scenario, each powered by a different small open-source model. The project explores how heterogeneous agents with distinct model architectures behave in competitive market conditions.
The simulation includes trading, resource bidding, and strategic alliances between the lab-agents, with each model bringing its own decision-making style to the economy. The project uses models under 7 billion parameters, demonstrating that complex multi-agent dynamics can emerge from relatively lightweight setups.
The work is part of the ongoing Build Small Hackathon, which challenges developers to create sophisticated AI applications using compact, locally-runnable models. The full source code, model configurations, and simulation logs are available on Hugging Face.