> As LLMs are not trustworthy, it's key to use them for things that are easy to check. If system A is only useful when it's output is confirmed by system B, th

> As LLMs are not trustworthy, it's key to use them for things that are easy to check. If system A is only useful when it's output is confirmed by system B, then you might conclude that the usefulness is entirely in System B, not in system A. In other words: What good is a LLM, if it can only give you trustworthy answers to questions where you already know the correct answer? What good is it if after getting the answer from a LLM, you then have to go and get the right answer somewhere else?