I think the "context window" that people refer to with large language models means there's a maximum number of tokens that are retained, with the oldest being d

I think the "context window" that people refer to with large language models means there's a maximum number of tokens that are retained, with the oldest being discarded. The window is a sliding window.