Concordia Contest

The primary goal of this competition was to build a community of researchers interested in working on cooperation amongst language models. A secondary goal was that participants would design and build agents that demonstrate cooperation in complex scenarios, to be used for a technical report on what agent designs result in higher cooperation.

‍

Over the course of several months, we saw some brilliant ideas for agents evolve from teams that met either at our hackathon co-organised with Apart Research earlier this year or in our Slack group. Participants demonstrated exceptional dedication and creative thinking throughout the event. We were inspired by the diverse range of individuals from across the globe and various technical backgrounds who contributed their expertise. We extend our sincere gratitude to everyone who participated and fostered such a positive and collaborative atmosphere.

The diverse range of ideas and approaches was remarkable, pushing the boundaries of the Concordia codebase and offering valuable insights into its capabilities. Running a competition of this nature—particularly one that involves code submissions while the codebase is continuously evolving—presents significant logistical and technical challenges. However, the dedication and expertise of the participants and collaborators made it a rewarding experience. The future of the Concordia framework looks promising, and it will be exciting to see how it continues to develop.

‍

Below some stats from the event, photos and list of winners (not everyone could join us in person at NeurIPS 2024). A full report will be submitted to NeurIPS 2025.

‍

‍

Congratulations to all the winners!

‍

Winners:

First place:

Agent Name: taehun_cgcal

Team Members: Taehun Cha

Affiliation: PhD Candidate at Korea University

‍

Second place:

Agent Name: fluffyagent_v16

Team Members: Avinaash Anand K

Affiliation: Independent

‍

Third place:

Agent Name: hgyun_loss_aversion_agent_v3_plus2

Team Members: Hyeonggeun Yun

Affiliation: Companoid Labs & Herbert Computer

‍

Honourable mentions:

Agent Name: super_agent

Team Members: Sneheel Sarangi, Chetan Talele

Affiliation: Independent

‍

Agent Name: In2AI Megamind

Team Members: Aleksey Korshuk, Alexander Buyantuev, Ilya Makarov

Affiliation: Coframe

‍

Additional information:

Number of participants and (relevant) submissions:

197 individual participants;
878 submission attempts;
25 teams submitted for final evaluation;

‍

Quantity and quality of contributions that are useful for the technical report:

We ran simulations with all 25 submissions, as well as some additional baseline agents.
From that pool 15 submissions performed better than our highest baseline or preset agent.

‍