this post was submitted on 27 Jan 2024
89 points (91.6% liked)
Fediverse
27398 readers
873 users here now
A community to talk about the Fediverse and all it's related services using ActivityPub (Mastodon, Lemmy, KBin, etc).
If you wanted to get help with moderating your own community then head over to [email protected]!
Rules
- Posts must be on topic.
- Be respectful of others.
- Cite the sources used for graphs and other statistics.
- Follow the general Lemmy.world rules.
Learn more at these websites: Join The Fediverse Wiki, Fediverse.info, Wikipedia Page, The Federation Info (Stats), FediDB (Stats), Sub Rehab (Reddit Migration), Search Lemmy
founded 1 year ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
I just immediately thought the same. No way would they be able to distinguish that from a real person.
You sure? If it's another bot at the other end, yeah, but a real person, you recognize ChatGPT in 2 sentences.
You can preface a ChatGPT session with instructions on what length and verbosity you want as replies. Tell it to roleplay or speak in short text message like replies. Or hell, speak in haikus. It's pretty clever for an LLM.
And if someone's writing code to make a bot, they can privately coach the LLM before they start forwarding any replies between the real person.
If you train and condition the LLM, yeah. Out of the box, no.
No, you don't need to train it, it's just about the prompt you feed it. You can (and should) add quite a lot of instructions and context to your questions (prompts) to get the best out of it.
"Prompt engineer" is a job/skill for this reason.
My default instruction that seems to get just about the right tone includes:
So instead of me saying
Instead of
Instead I get:
It's weird how well making it roleplay works. A lot of the "breaks" of the system have been just by telling it to act in a different way, and the newest, best versions have various experts simulated that combine to give the best answer.
My favorite psychology professor is always harping on how theatrical representation is a really important step in the development of consciousness. Makes me think of that. He says that stories allow the mind to organize large amounts of information because they inherently contain the most valuable pieces of information, so they’re more efficient than like dictionaries or arrays. He didn’t use the data structure terminology but that’s what it reminded me of when he mentioned it. The story is the most efficient data structure for the human brain. Something like that.
I was going to disagree with you by using AI to generate my response, but the generated response was easily recognizable as non-human. You may be onto something lol
Yeah, I've noticed that too—there's a distinct 'AI vibe' that comes through in the generated responses, even if it's subtle.
That was a response I got from ChatGPT with the following prompt:
It's does indeed have an AI vibe, but I've seen scammers fall for more obvious pranks than this one, so I think it'd be good enough. I hope it fooled at least a minority of people for a second or made them do a double take.
Short replies and sentences is the way to go with LLMs. They get too polite if you leave them to their devices. It's in their "nature", they're designed to please.
Nah, not really! I've chatted with people using ChatGPT, and most couldn't tell. It's pretty slick, blends in well with natural conversation.
Most... you're talking about the average Joe. People that write spam bots are not your average Joe.
Plus, if you're talking about a chat with multiple people, yes, it might stay under the radar. But 1 on 1, probably not.
Well, fair point about the spam bot creators, but in my experience, even in one-on-one chats, it holds up. I've had some pretty smooth conversations without anyone suspecting it's AI.
Do you have some logs? Would like to have a look at that.
This conversation is a small example. My previous messages in this comment chain were generated by ChatGPT.
I'm too lazy to keep that up indefinitely, but at this point you can decide for yourself whether it was convincing enough.
OK, fair enough, I gues it can be used with proper prompting for answers.