Imagine teaching a self-driving car not just to follow rules, but to drive with the nuance and adaptability of a human. That's the exciting potential of a new research project from the University of Toronto, which uses large language models (LLMs) to make reinforcement learning (RL) for automated driving more human-centric. Traditionally, training AI drivers involves complex reward functions that try to capture every aspect of safe and efficient driving. This research takes a different approach, using LLMs to provide real-time feedback on driving decisions, much like a human instructor. The LLM receives information about the current driving scenario, including the car's speed, the position of other vehicles, and the intended action. It then provides a reward signal based on how well the action aligns with human-like driving behavior. This approach allows the AI to learn more nuanced driving strategies, adapting to different situations and driving styles. Researchers tested this approach in a simulated highway environment, creating 'aggressive' and 'conservative' AI drivers by providing different instructions to the LLM. The results were impressive: the AI drivers exhibited distinct driving styles, mirroring the instructions given to the LLM. The aggressive AI prioritized speed and efficiency, while the conservative AI focused on safety and stability. This research opens up exciting possibilities for personalized automated driving, where drivers could customize their AI's driving style to match their preferences. It also suggests that LLMs could play a crucial role in making self-driving cars more adaptable and responsive to real-world driving complexities. While the results are promising, there are still challenges to overcome. The LLM's understanding of the driving environment needs further refinement to ensure accurate decision-making in all situations. Future research will focus on improving the LLM's ability to interpret complex scenarios and incorporate more sophisticated driving rules, paving the way for safer and more human-like automated driving systems.