Large language models (LLMs) like ChatGPT are impressive, but they still struggle with basic common sense. Imagine trying to teach an AI that spilling a drink usually leads to cleaning up the mess. This seemingly simple concept is difficult for LLMs to grasp, making them prone to errors and outdated information. Updating these models with new commonsense knowledge is computationally expensive. Researchers are exploring efficient ways to "edit" an LLM's knowledge base without massive retraining. Traditional methods focus on correcting simple facts, like who the current president is. However, common sense is more nuanced and harder to pin down. A new study introduces a novel approach to editing commonsense knowledge within LLMs. Researchers discovered that, unlike factual knowledge, common sense is distributed throughout the model's "brain," particularly in the MLP and Attention layers. This makes editing more challenging. They developed a method called "Dynamics-aware Editing" (DEM), which pinpoints and modifies the relevant parts of the model responsible for a specific piece of common sense. Initial results on models like GPT-J and LLaMA-2 are promising, showing significant improvements in commonsense reasoning. This research opens exciting possibilities for making LLMs more reliable and aligned with human understanding. While challenges remain in scaling these techniques to larger models, it's a step toward building AI that truly "gets" the world.