Yeah, for a tech person like me it is even obvious. Take ChatGPT for example. All it is doing is tossing a many-sided dice to choose a next word to finish a text. Yes, the user prompt is considered as a text that should be extended by the neural network.

I am not downplaying the role of OpenAI engineers. They did a really amazing job to make a language model that assigns probabilities for words in a given context. But in the end, it is just a random number generator that chooses one word at a time from a given list.