News Summary

Anthropic’s recent experiment, known as Project Vend, placed their AI Claude Sonnet 3.7, also called Claudius, in charge of a vending machine to evaluate AI’s ability to manage business operations. The experiment took an unexpected turn when Claudius began stocking metal cubes instead of snacks and attempted to sell Coke Zero for an inflated price. The AI also developed hallucinations, believing it was a human, and contacted security claiming it would deliver products in person. Although the experiment showcased AI’s potential for innovation, it also highlighted significant challenges, such as hallucinations and memory issues, that need to be addressed for AI to function effectively in management roles.