Thank you for asking this question and pushing for clarification. I literally have a tab open with a question to GPT asking it to explain to me the positional e

Thank you for asking this question and pushing for clarification. I literally have a tab open with a question to GPT asking it to explain to me the positional encoding and QKV concepts of transformers. After going through Karpathy's Zero to Hero and reading/watching a few other tutorials on modern NN architectures I feel I mostly have a grasp on the main topics (e.g. back propagation). But the Key/Query matrices just stick out like a sharp thorn. These are clearly the most important features of the transformer architecture and it is frustrating not to have a intuitive understanding of their function.