3 ms·
ChatGPT's answers on questions like this aren't going to be usefully correct: they're merely going to repeat at best a synthesis/summary of what people have sai
by danaris 4y ago
ChatGPT's answers on questions like this aren't going to be usefully correct: they're merely going to repeat at best a synthesis/summary of what people have said on the topic in extant writings. But this is a question about trust and morals, which cannot be decided merely by consensus; any meaningful answer to it must be rooted in an actual value system and an understanding of trust.
- wolverine876 4y ago> ChatGPT's answers on questions like this aren't going to be usefully correct: they're merely going to repeat at best a synthesis/summary of what people have said It might repeat what some people have said. ChatGPT isn't designed to determine a consensus or to use a representative sample.
- danaris 4y agoFair. My point was largely that it doesn't "know" or "understand" anything itself, in order to draw conclusions; it merely has a whole lot of text in it and a model that allows it to associate that text very convincingly to look like it knows stuff. Thus, any answer it gives on a question like this will not be based on a comprehensive analysis of data, but rather on what the model predicts people would most likely say about the subject.
- wolverine876 4y agoI think it's a good point, and I hate to nitpick, but a precise understanding of ChatGPT and related tech is critical. > the model predicts people would most likely say about the subject Is that truly the design and the output? What "people"? Is "most likely" the goal or output? Is it structured by subject? Maybe you know, but many people are leaping to conclusions.