The Collaborative Abilities of ChatGPT Agents in a Number Guessing Game

Exciting Announcement: Second LLM Lunch Seminar

Dive into the fascinating world of AI at our upcoming seminar featuring cutting-edge research on ChatGPT-3.5 and ChatGPT-4!

When: Jan 19 at 12:00 (bring your lunch)
Where: Room L5DE13 and zoom

Presenter: Katja Sangati

Join us for an insightful discussion on the latest paper, to be presented at the 29th International Symposium on Artificial Life and Robotics.

Title: The Collaborative Abilities of ChatGPT Agents in a Number Guessing Game

Authors: Ekaterina Sangati, Federico Sangati, Marc Slors, and Kenji Doya.

Abstract: Unveiling the Minds of Machines
We present a new behavioral evaluation of ChatGPT-3.5 and 4 Large Language Models. Groups of 3 ChatGPT agents were asked to play a collaborative number-guessing game in which each agent had to submit a number, and the sum of the team's numbers needed to match the target number. ChatGPT-4 model showed better performance than ChatGPT-3.5. However, both models did worse than previously reported human participant results in this game. A deeper analysis of model errors shows that the two models failed for different reasons, and neither model adopted human-like strategies of social coordination.

Mark your calendars for Jan 19 at 12:00, Room L5DE13 or zoom! Don't miss this chance to delve into the minds of these language models and explore the future of AI.

Disclosure: This message has been generated with the help of ChatGPT.