The fact
The authors emphasize that relevance of retrieved information outweighs the sheer volume of tokens in the prompt.
This challenges the assumption that larger context windows (e.g., 128k tokens) automatically improve outcomes.
Click the link to read an article on the topic: