Last year the IBEX lab received a subsidy from NWO to build a Multi-Actor LLM tool that can be used to automate experiments with LLM agents. A first version of the tool is now available on GitHub.
The tool consists of a Python package that allows you:
– to connect to a server that runs a game, as a python script, as a local server, or online
– initiate different players with different roles, different LLM models, and different prompts
– manage the information state of each of the each role actor
– specify how the players respond to the server to provide their chosen action
an example game, repeated prisoners dillema, is part of the package already. So are the implementation of more complicated choices of our Futarchy for Spatial Planning via Harberger Taxation | IBEX-lab experiment ( but you will have to provide the server for that).
Python library:
with documentation here:
https://econagents.readthedocs.io/
Recent Comments