This has always been the premise. They can't fix the fundamental problems with LLMs but they can continue to optimise them for IE parsing large volumes of data quickly