4 ms·
Sorry I really did not explain this well. I don't mean the people who make and officially train ChatGPT. I mean people who can adjust it's answers by engaging i
by tus666 3y ago
Sorry I really did not explain this well. I don't mean the people who make and officially train ChatGPT. I mean people who can adjust it's answers by engaging in debate with it. It does have a liberal bias by default, but you can easily get it to admit it's wrong on almost everything by pointing out correct facts that contradict liberal arguments. This feeds into answers it gives to other people.
- leafstrat 3y ago> This feeds into answers it gives to other people. According to who?
- tus666 3y agoOpen.AI and ChatGPT itself. Ask it. The conversations ChatGPT feed back into the model, it's one of the ways they improve it. In particular, if you can get it to admit it's wrong on something, and it verifies that internally, it won't make the same mistake again. It's self-correcting in that way.
- smoldesu 3y agoCan you cite something that doesn't have a disclaimer reading "May occasionally generate incorrect information" on this? It's news to me.
- smoldesu 3y ago> you can easily get it to admit it's wrong on almost everything by pointing out correct facts that contradict liberal arguments You can get it to admit pigs fly for whatever that's worth. It's a prompt answering machine, not a truth-seeker. It's "liberal bias" could be chalked up to the majority of freely-licensed online content having a liberal bias. It's present in all of these models, even the unfiltered ones. Perhaps the conservative powers-that-be will band together and make a new, equally stupid and biased model to compete with the current stupid and biased models. Elon Musk has played with the idea recently, is he the Atlas Shrugged savior you're looking for? > people who can adjust it's answers by engaging in debate with it This does not happen.
- tus666 3y ago> You can get it to admit pigs fly for whatever that's worth Have you used ChatGPT? I think you a greatly underestimating it. It really is a truth seeker, you just need to provide it facts it can correlate against actual historical record. It's quite impressive.
- smoldesu 3y ago> Have you used ChatGPT? As a matter of fact, I have. Let me read you one of the first things listed directly after you log in: > May occasionally generate incorrect information They admit it. They know, they don't accept 'bug reports' of incorrect responses because the system is working as-intended. > it can correlate against actual historical record. It correlates against it's training data. Part of that is truth and history, part of it is excerpts from Tolkien and Twilight fanfiction.