I'm not even sure. From a LLM standpoint that's even a good idea. From all my research on getting good output context matters. So if you throw too much junk in the middle it'll get evens stupider then it already is