I don't think it's particularly helpful to dive into transformer models, positional encodings and self-attention at the very start of an introduction to LLMs. Understanding how those works does little to help explain what LLMs can do and how you can use them.
I tend to stick with the higher level explanation that they can predict the next word (or next sentence) based on their training text, and then emphasize that while this sounds pretty limited it's actually capable of doing all sorts of impressive things once you scale it up enough.
I tend to stick with the higher level explanation that they can predict the next word (or next sentence) based on their training text, and then emphasize that while this sounds pretty limited it's actually capable of doing all sorts of impressive things once you scale it up enough.