A team of three engineers recently attempted to integrate artificial intelligence into the driving process by installing large language models (LLMs) into a Toyota Corolla. The goal of the experiment was to see if AI agents, specifically GPT, Claude, and Grok, could navigate a real-world vehicle to a destination—in this case, a local In-N-Out Burger.

The project highlights the growing interest in applying generative AI to physical robotics and autonomous systems. While LLMs are typically used for text generation, coding, or data analysis, this experiment tested their ability to process real-time environmental data and execute driving commands. According to the report, the results were mixed: while one of the models was able to successfully navigate the vehicle, the other two failed to complete the objective. The researchers did not specify which model succeeded or the exact technical reasons for the failures of the other two, focusing instead on the broader implications of using these models for physical automation. This experiment serves as a proof-of-concept for how AI might eventually interact with vehicle controls, though it also underscores the current limitations and reliability issues inherent in using language-based models for high-stakes physical tasks like driving.