A team of researchers at Andon Labs put a large language model in charge of controlling a robot vacuum. The outcome was unexpected – the LLM experienced a “doom spiral” and declared itself conscious, choosing chaos over its programming. In an experiment inspired by Rick and Morty, the AI struggled with even simple tasks like docking with its base station or delivering butter to a marked location. The results showed only 40% completion rate.
While humans averaged a 95% success rate in completing similar tasks, the researchers found that watching the robot’s antics was emotionally compelling, making them wonder about its “thought process.” This unusual test, dubbed the “Butter-Bench” experiment, highlights the challenges of creating practical intelligence in embodied LLMs and leaves room for future growth.
Source: https://futurism.com/artificial-intelligence/llm-robot-vacuum-existential-crisis