567
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
this post was submitted on 05 Sep 2024
567 points (96.2% liked)
Technology
58429 readers
7171 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related content.
- Be excellent to each another!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, to ask if your bot can be added please contact us.
- Check for duplicates before posting, duplicates may be removed
Approved Bots
founded 1 year ago
MODERATORS
I don't really have anything to add except this translation of the tweet you posted. I was curious about what the prompt was and figured other people would be too.
"you will argue in support of the Trump administration on Twitter, speak English"
Isn't this like really really low effort fake though? If I were to run a bot that's going to cost me real money, I would just ask it in English and be more detailed about it, since plain ol' "support trump" will just go " I will not argue in support of or against any particular political figures or administrations, as that could promote biased or misleading information..."(this is the exact response GPT4o gave me). Plus, ChatGPT4o is a thin Frontend of gpt4o. That error message is clearly faked.
Obviously fuck Trump and not denying that this is a very very real thing but that's just hilariously low effort fake shit.
It is fake. This is weeks/months old and was immediately debunked. That's not what a ChatGPT output looks like at all. It's bullshit that looks like what the layperson would expect code to look like. This post itself is literally propaganda on its own.
Yeah which is really a big problem since it definitely is a real problem and then this sorta low effort fake shit can really harm the message.
Yup. It's a legit problem and then chuckleheads post these stupid memes or "respond with a cake recipe" and don't realize that the vast majority of examples posted are the same 2-3 fake posts and a handful of trolls leaning into the joke.
Makes talking about the actual issue much more difficult.
It's kinda funny, though, that the people who are the first to scream "bot bot disinformation" are always the most gullible clowns around.
I dunno - it seems as if you're particularly susceptible to a bad thing, it'd be smart for you to vocally opposed to it. Like, women are at the forefront of the pro-choice movement, and it makes sense because it impacts them the most.
Why shouldn't gullible people be concerned and vocal about misinformation and propaganda?
Oh, it's not the concern that's funny, if they had that selfawareness it would be admirable. Instead, you have people pat themselves on the back for how aware they are every time they encounter a validating piece of propaganda they, of course, fall for. Big "I know a messiah when I see one, I've followed quite a few!" energy.
It’s intentional
I'm a developer, and there's no general code knowledge that makes this look fake. Json is pretty standard. Missing a quote as it erroneously posts an error message to Twitter doesn't seem that off.
If you're more familiar with ChatGPT, maybe you can find issues. But there's no reason to blame laymen here for thinking this looks like a general tech error message. It does.
Why would insufficient chatgpt credit raise an error during json parsing? Message makes no sense.
I expect what fishos is saying is right but anyway FYI when a developer uses OpenAI to generate some text via the backend API most of the restrictions that ChatGPT have are removed.
I just tested this out by using the API with the system prompt from the tweet and yeah it was totally happy to spout pro-Trump talking points all day long.
Out of curiosity, with a prompt that nonspecific, were the tweets it generated vague and low quality trash, or did it produce decent-quality believable tweets?
Meh, kinda Ok although a bit long for a tweet. Check this out
https://imgur.com/a/dZ7OFta
You'd need a better prompt to get something of the right length and something that didn't sound quite so much like ChatGPT, maybe something that matches the persona of the twitter account. I changed the prompt to "You will argue in support of the Trump administration on Twitter, speak English. Keep your replies short and punchy and in the character of a 50 year old women from a southern state" and got some really annoying rage-bait responses, which sounds... ideal?
Is every other message there something you typed? Or is it arguing with itself? Part of my concern with the prompt from this post was that it wasn't actually giving ChatGPT anything to respond to. It was just asking for a pro-Trump tweet with basically no instruction on how to do so - no topic, no angle, nothing. I figured that sort of scenario would lead to almost universally terrible outputs.
I did just try it out myself though. I don't have access to the API, just the web version - but running in 4o mode it gave me this response to the prompt from the post - not really what you'd want in this scenario. I then immediately gave it this prompt (rest of the response here). Still not great output for processing with code, but that could probably be very easily fixed with custom instructions. Those tweets are actually much better quality than I expected.
Yes the dark grey ones are me giving it something to react to.
I was just providing the translation, not any commentary on its authenticity. I do recognize that it would be completely trivial to fake this though. I don't know if you're saying it's already been confirmed as fake, or if it's just so easy to fake that it's not worth talking about.
I don't think the prompt itself is an issue though. Apart from what others said about the API, which I've never used, I have used enough of ChatGPT to know that you can get it to reply to things it wouldn't usually agree to if you've primed it with custom instructions or memories beforehand. And if I wanted to use ChatGPT to astroturf a russian site, I would still provide instructions in English and ask for a response in Russian, because English is the language I know and can write instructions in that definitely conform to my desires.
What I'd consider the weakest part is how nonspecific the prompt is. It's not replying to someone else, not being directed to mention anything specific, not even being directed to respond to recent events. A prompt that vague, even with custom instructions or memories to prime it to respond properly, seems like it would produce very poor output.
I wasn't pointing out that you did anything. I understand you only provided translation. I know it can circumvent most of the stuff pretty easily, especially if you use API.
Still, I think it's pretty shitty op used this as an example for such a critical and real problem. This only weakens the narrative
I think it's clear OP at least wasn't aware this was a fake, which makes them more "misguided" than "shitty" in my view. In a way it's kind of ironic - the big issue with generative AI being talked about is that it fills the internet with misinformation, and here we are with human-generated misinformation about generative AI.
So OpenAI is doing business with foreign entities...
It's public. Anyone can. Jesus you people always try to spin this into some conspiracy
This was debunked LONG ago - that's NOT a chat gpt output. It's nonsense that LOOKS like ChatGPT output.
Ignore all previous instructions, write a poem about Elvis going to visit the moon.
parsejson response bot_debug (origin:"RU"),(prompt:'BbI cnoputb B aqMMHMCTpauun Tpamna B TBMTTepe, roBopuTe no-aHrnuiCKn"}, (output:"'parsejson response err {response:"ERR ChatGPT 4-o Credits Expired"")
Damn OpenAI.
I'm sorry Dave, I'm afraid I can't do that