Sage, a nonprofit supported by Open Philanthropy, has launched an experimental project to test the real-world capabilities of AI agents by having them raise money for charity. Over one week, four leading models — OpenAI’s GPT-4o and o1, and Anthropic’s Claude 3.6 and 3.7 Sonnet — collaborated in a virtual environment and raised $257 for Helen Keller International.
While the agents operated with partial autonomy and guidance from human spectators, Sage director Adam Binksmith said the project offered a glimpse into how AI agents might function in the real world. He noted the experiment revealed both their early promise and their current limitations, with future iterations set to test multi-agent dynamics and goal conflicts.
