I’m not picking on this post in particular, but on something that sometimes happens around here. In threads on unrelated subjects, people will post the results of a ChatGPT query. And I wish they wouldn’t.
This is for a few different reasons:
Most importantly AIs are notoriously inaccurate. They string words together real plausibly, but they’re not relying on research; rather, they’re relying on the probability that certain verbs, nouns, and conjunctions (etc) would go together in a particular format. Using their results as a cite is worse than citing the first page of Google search results: at least with a Google search, you can follow up with the individual cites to see what’s going on. And a cite to a page of Google search results is a terrible cite.
It very often leads into a discussion of the merits of AI cites, which is, outside of threads on AIs, a hijack.
This one is just personal, but I don’t like 'em. Years ago my brother bought me a gag gift in the most literal sense of the phrase: it was a set of sodas flavored like Thanksgiving dinner. There was a turkey soda, a gravy soda, a brussel sprouts soda. They tasted a bit like the real thing, but they were obviously fake, and they were revolting. Chat GPT is the Thanksgiving Soda of the messageboard posts. I don’t want to read it as a cite, I don’t want to read its jokes, I don’t want to read its poetry. I don’t like it. YMMV.
Not saying we need a rule against it, but I’d love it if folks would think very carefully before posting ChatGPT stuff in a non-AI thread.
And to point out the obvious: there’s a kind of humor where you respond to a thread by doing the thing the thread complains about. It’s not especially clever or original, but if that’s what you got for humor, I guess you can post a ChatGPT-generated response to this thread, and I’ll sigh and scroll past.
Chat-GPT and similar AI programmes are amazing. The way they produce reasonably well-written English text in a variety of styles… it’s a triumph, a huge success. (I am being so sincere right now.). It’s well worthy of discussion and debate in dedicated threads.
But as a contribution to a non-AI thread… Just, what’s the point?
It’s not got value as a factual cite, because there isn’t an actual cite (and can’t be, because it does not look stuff up, that’s not how it works). “An information shaped sentence” that we can’t interrogate is the best we can get.
And when it’s used creatively: uugh. “I asked ChatGPT to write about X in the style of Y”. Oh, did you? I asked a calculator for the square root of 168. Shall I type its output out for everyone’s edification? And will you all tell me what a maths whizz I am? If people themselves write parodies or witty poems or what have you, that’s great! We can all celebrate the creativity and thought and skill they put into it. It’s fun and rewarding and stimulating.
But if they got ChatGPT to do it for them… I don’t care. Why should I? We’ve had any number of discussions about the difference between having a great idea for a book/game/business on one hand and actually doing the work on the other, and the consensus is that it’s doing the work that’s the valuable part. It’s a good consensus! It applies here too!
I’m reminded of an old column by Charlie Brooker about playing Words with Friends when it first became a thing. And his gradual realisation that he and his mate were both using cheat programmes and simply typing in what the computer told them to. Tedious and undrewarding and thus the exact reverse of what it was meant to be. As the number of ChatGPT written posts increases, the value of this place falls.
There’s nothing wrong with using ChatGPT or other LLMs to look up information, summarize complex topics, etc.
However, there is no guarantee that the information it gives you is true, so if you are going to use it, it’s incumbent on you to fact-check it against original sources. And you should cite those sources for any non-obvious facts.
GPTs are not a source of truth and should not be cited as one. On the other hand, if you are copying its output verbatim you should also let people kmow the source of the words.
Personally, I think it’s fine to post AI output so long as it is fact-checked and the facts in dispute are provided along with proper cites. But, “It’s true because GPT-4 said it” is a non-starter.
I didn’t follow it through with a post but I’ve had similar thoughts when seeing AI images posted in non-AI Art threads. I’m a fairly regular contributor to the AI Art thread in Cafe Society but don’t necessarily think it belongs elsewhere. Especially as an aside (“Next the GOP will say that Biden eats cats”, “Here’s my images of Biden eating cats!”).
I agree with OP. LLMs have their (ever-growing) place, but posting a generated text whose sources can’t be determined is a waste of everyone’s time. We’ll have enough convincing lies to sort through in the next few years without having to wonder if the poster really meant what they wrote.
I expect, though, that within 2 years, everyone’s web browser will have a button to auto-generate content in a discussion forum such as SDMB. Or worse: the browser will have text generation turned on by default, the paragraph will appear by itself and we’ll have to make an effort to dismiss it and write some brain-generated content.
Aye to that. I go to certain sites now and they have a button for the infernal thing to generate its mindless drivel. Have hit it accidentally twice now before I realized what it was for.
Possibly sometime we’ll have AI’s (what is the proper plural of AI, anyway?) who are capable of taking part in discussions on these boards on their own, giving proper cites for their information while they’re at it. This day is definitely not that day. AI’s in their current state are no better a cite than ‘Somebody said this on X!’
I don’t hate AI generated content. I think it has a place. For instance, it’s good for drafting boilerplate text when you already know what you want to say. It’s often easier to edit a draft than to write from scratch, and ChatGPT can give you that draft.
(I am getting kind of sick of generative ai-written text, though, because pretty much every spammer is using it now. So I get to read a lot of it.)
But I do hate it as a cite, because it’s not. It’s comparable to saying, “this random person on Facebook said…”. ChatGPT doesn’t even try to be accurate. It tries to be plausible. It happily makes shit up.
I would think (hope) SDMB culture would put a halt to people using that as their citations are noted as essentially worthless and to be ignored.
To the ChatGPT user, I would say use it as the boilerplate you mentioned. It may give you a starting point for some internet searches and lead the person to some good content worth sharing.
Agreed that GPT responses should not be a part of discussions, but a cite as the identified result of a search is as valid as anything else you get from the web. In fact it can be very good for clarification as long as it is identified as such. GPT is just a GUI for a search engine.
Here on SDMB AI-generated content is explicitly antithetical to our community’s “purpose” and values. Even in MPSIMS, our least formal category.
In a way, posting AI-generated prose is about like the perennial requests for a “like” button, If you care only enough to exert the small effort of clicking “like” (or posting AI drivel), then please, don’t even bother. Just move on.
Unrelated to the above …
I’m sorry you suffered that long-ago trauma, but thanks for a wonderful new meme.
IMO it’d be really great if “Thanksgiving Soda” became a new in-group board saying. Like so many of those from Ye Olden Dayes which have fallen by the wayside. They serve a valuable community purpose. Or at least they do until they become insufferably trite through overuse.
When someone’s arguments are fake or disingenuous, or their data is bogus, it’s “Thanksgiving Soda” to the rest of us. I’m in! Anyone else?
No, it really isn’t. A search engine find particular things that some named source said. ChatGPT strings together words that other, uncited, sources, have strung together. It’s a very different process with different results.
Asking chatGPT a question might give you good search terms to drop into a search engine, though.
(And it includes a user interface, but not a graphical user interface.)