what is the value in manually processing knowledge bases like this (besides perhaps in an artisan fashion) if your knowledge interface such as chatgpt already knows it and surfaces it at will?