Hi Amit — I think that the ability of ChatGPT to do multiple tasks, including code generation, isn’t due to explicit internal models to tackle any particular task. For example, you can get it to do navigation of the New York or London subway systems, and there’s no way that models were explicitly created for these tasks. It’s the breadth of the training set and the intrinsic flexibility of LLMs (not just ChatGPT) that makes it capable of doing so many tasks.