Ask HN: How LLMs know the context of some question?

Often at the end of an LLM dialog, they respond with something like "Happy coding."

I didn't think much about it until today: If LLMs are simply the product of a transformer decoding weights from an immense volume of text, how do they know I am coding? It's not written at the end of the answer, it's only written if I thank them!

Maybe it's in the training set, but I doubt that every answer on Stack Overflow ends with "Happy coding!"

So where does it come from?