Tech Workers Made ChatGPT Drive a Toyota Corolla (www.404media.co)

A group of San Francisco tech workers, known as DrivingBench, successfully connected various AI language models (LLMs) to a Toyota Corolla, enabling it to drive around a parking lot. This experiment stands out because it utilized general-purpose LLMs—GPT-6 Astra, Claude Fable 5.1, Grok 4.6, and GPT-5.6 Sol—rather than specialized self-driving technologies like those from Waymo or Tesla. While only GPT-6 Astra managed to navigate the course successfully, the attempt highlights the potential for frontier AI models to tackle real-world tasks without extensive training. The significance of this endeavor lies in its exploration of LLMs’ emergent capabilities and potential for real-world applications. By demonstrating that these models can drive—even within a limited environment—the team aims to set a new benchmark for LLM performance in tasks typically handled by specially designed systems. Key technical challenges included modifying prompts to convince the models to engage with the task and ensuring responsiveness under driving conditions, illustrating both the innovative spirit of the experiment and the limitations of current AI capabilities. Ultimately, this work points to a future where LLMs might increasingly influence real-world tasks beyond conventional screen-based interactions.