aside from what others have said, here's another point : it's not easy to know what you don't know. Every answer given by GPT is the best answer that specific

aside from what others have said, here's another point : it's not easy to know what you don't know. Every answer given by GPT is the best answer that specific model can give given the context and relational clues, but counting those 'nodes' of context and relation clues isn't actually a good metric for discovering the unknown and finding holes in the model.