Three engineers conducted an unusual experiment by allowing popular AI models to control a real Toyota Corolla during a drive to In-N-Out, according to WIRED. The researchers tested GPT, Claude, and Grok in actual driving conditions, with only one of the three AI systems successfully completing the task.
According to the WIRED report, the engineers put these large language models in charge of operating a real vehicle, representing a novel approach to testing AI capabilities in physical, real-world scenarios rather than simulations. The experiment involved navigating to the popular fast-food chain In-N-Out, requiring the AI systems to handle real traffic conditions and driving decisions.
While the article notes that only one of the three AI models succeeded in the driving task, the specific details about which model succeeded and the nature of the failures were not elaborated in the available source material. The experiment highlights both the potential and current limitations of applying general-purpose AI models to specialized tasks like autonomous driving.