Concordia Contest at NeurIPS 2024 Competition
A language-model agent for cooperative social simulations, placing third at NeurIPS 2024 and contributing to a NeurIPS 2025 paper.
About this project
I designed a language-model agent for the Concordia Contest, a cooperative intelligence competition at NeurIPS 2024. I implemented the agent in Python using Google DeepMind’s Concordia framework. In text-based social simulations, it reflects on its own and other agents’ past actions and scores each option’s potential loss before acting, applying a loss-aversion approach to decision-making.
My entry reached the top-five final and placed third in the competition track. I presented the work in Vancouver. This participation also led to co-authorship of “Evaluating Generalization Capabilities of LLM-Based Agents in Mixed-Motive Scenarios Using Concordia,” published in the NeurIPS 2025 Datasets and Benchmarks Track.