How about "ChatGPT will fabricate rhetoric regardless of logic or truthfulness"? Which just makes me think ChatGPT could replace a great majority of politicians.
We need to tell people ChatGPT will lie to them, not debate linguistics
481–485 of 485 posts
Re: We need to tell people ChatGPT will lie to them, not debate linguistics
#482simonw’s linguistic argument seems to be “i’m right and i’m not gonna listen to you lalalalalala”. throwing a strawman paragraph comparison to a 5 word phrase doesn’t bolster his argument at all. You could just say “ChatGPT gets things wrong” instead of “ChatGPT will lie to you!” without the yelling and anthropomorphizing
We've been trying that for months. "ChatGPT gets things wrong" does not appear to be sinking in for a lot of people. I used a strawman paragraph deliberately as a rhetorical device: that's what strawman paragraphs are for.
shit I'm honored you replied, sorry for being rude about it. Didn't think my comment would even get seen in a days old thread. Didn't account for its appreance in /newcomments or however you found it.
Anyway, I can appreciate the rhetoric, I've used that technique myself. My opinion on the matter is that we do need to debate linguistics before we can properly proceed. I know it does't feel like work or it feels like a waste of time if you're coming at it from the angle of a software engineer, writing code and delivering value every day, but in the humanities, words and diction are paramount.
"ChatGPT gets things wrong" will sink in for people when they personally encounter a wrong bit that trips them up. hopefully before they put their money or reputation on the line, or use the "facts" as load bearing. Because that's the real worry, isn't it. Load bearing facts result in a doctorGPT that will prescribe the wrong medication and kill somebody, that a lawyerGPT will write a contract that hurts someone, that programmerGPT will write the wrong code and hurt somebody, financially or reputationally online.
Re: We need to tell people ChatGPT will lie to them, not debate linguistics
#483Earlier quoted context omitted.
> Do you have specific reasons for thinking that gpt-4 is near the end of this trend? I haven’t said that. I pointed that you can’t simply put more data into it and expect improvements. There’s definitely a lot of engineering done on the models to make them capable of using all the data. I can give you an example. You can’t simply add more convolutional layers to a CNN and feed more images. The architectures of image…
You are largely wrong, to be honest. Mostly, you can simply stack more transformers and (as long as you have layer norm), it can train on more data (of fixed sequence length). There are some tricks & tips, but the vast majority lie outside of the architecture improvements.
I’m not an expert, but I assume there’s a limit where «simply adding stuff” stops to provide improvements when you take training time into account. And my gut feeling is that true reasoning requires different architectures.
Re: We need to tell people ChatGPT will lie to them, not debate linguistics
#484Earlier quoted context omitted.
I'd actually go further and say it's better to assume I can be deceived by confident language, than to assume that this is a problem with "dumb people". If I see other people making a mistake, I want my first question to be "am I making the same mistake?". I don't live up to that aspiration, certainly.
Yeah this is a common sales tactic, and talented/ charismatic people can certainly employ confidence for an uptick in sales. The way to combat this is the same for all hoaxes/disinfo and so on, epistemology. The attitude of the Socratic method. As I write this I wonder if ChatGPT can use the Socratic method on itself.
Part of what's extremely frustrating about talking to people about AI is that people say things which seem like they understand what's going on in one breath, and then in the next breath say something that makes absolutely no sense if they understood what they just said. This post is a great example of that.
Okay, so you apply epistemology to ChatGPT: when you ask ChatGPT a question, how does it know the answer? The answer is: it doesn't know the answer. All it knows is how people string words together: it doesn't have any understanding of what the words mean.
So no, it can't use the Socratic method on itself or anyone. It can't ask questions to stimulate critical thinking, because it's incapable of critical thinking. It can't draw out ideas and underlying presuppositions, because it doesn't have ideas or underlying presuppositions (or suppositions). It's not even capable of asking questions: it's just stringing together text that matches the pattern of what a question is, without even a the understanding that the text is a question, or that the following text in the pattern is an answer. "Question" and "answer" are not concepts that ChatGPT understands because ChatGPT doesn't understand concepts.
Re: We need to tell people ChatGPT will lie to them, not debate linguistics
#485The part that's concerning about ChatGPT is that a computer program that is "confidently wrong" is basically indistinguishable from what dumb people think smart people are like. This means people are going to believe ChatGPT's lies unless they are repeatedly told not to trust it just like they believe the lies of individuals whose intelligence is roughly equivalent to ChatGPT's. Based on my understanding of the appro…
While reading your comment I realized that being "confidently wrong" is actually really human, so... yay?
I mean, that's one step closer to machines thinking like humans, right?
:)