Researchers Test LLMs as Autonomous Drivers for a Toyota Corolla

A team of three engineers recently conducted an experiment to determine if large language models could effectively operate a real-world vehicle. By integrating GPT, Claude, and Grok into a Toyota Corolla, the researchers aimed to see if these AI systems could navigate the car to an In-N-Out Burger location. The study highlights the current limitations of generative AI when applied to physical autonomous systems. While these models excel at text generation and reasoning tasks, the experiment demonstrated that they struggle with the real-time sensory processing and safety-critical decision-making required for driving. Only one of the tested models managed to complete the task successfully, underscoring the significant gap between general-purpose language intelligence and the specialized requirements of autonomous vehicle control. The project serves as a cautionary exploration of using LLMs for robotics and physical world interaction.