Overview

• ChatGPT was tasked to autonomously control a pursuit spacecraft in a Kerbal Space Program simulation.
• With minimal prompting, the model earned second place behind a physics-based control system.
• Detailed results are slated for publication in the Journal of Advances in Space Research.
• The study’s pre–GPT-4 timing indicates that more recent models could deliver even stronger performance.
• Persistent hallucinations and unpredictability spotlight the need for safety measures before any real missions.