AI chatbot trained on posts from web sewer 4chan behaved badly – just like human members

A prankster researcher has trained an AI chatbot on over 134 million posts to notoriously freewheeling internet forum 4chan, then set it live on the site before it was swiftly banned. Yannic Kilcher, an AI researcher who posts some of his work to YouTube, called his creation "GPT-4chan" and described it as "the worst AI ever …

[A Non e-mouse]

Pot Vs Kettle

'Nuff said.

[Pascal Monett]

So he trained an AI on 4chan

And then was surprised that the results were not appreciated. Well duh.

I don't really care that he created a response bot on 4chan. What irks me is that he wasted resources on training a so-called AI in the worst possible environment.

This is why we should not use the term AI to describe what these things are doing. There is no intelligence involved.

[Filippo]

Re: So he trained an AI on 4chan

From reading his comments on the article, it doesn't seem he was surprised. It sounds like the model worked exactly as expected.

[stiine]

Re: So he trained an AI on 4chan

I bet he spent less money on his bot and acihieved the same results as Microsoft did with theirs.

[ShadowSystems]

It's worse than that.

As soon as the idiots at MS heard about the 4Chan chatbot, they immediately grabbed a copy, installed it, ran it, & put it in charge of their HellDesk Support online/phone "help" system. =-Jp

[chivo243]

Re: It's worse than that.

I thought they ported it over to Clippy?!

[very angry man]

Re: It's worse than that.

As did many others,

this is now happily inserted into many automated help desks around the internet.

people are finding it far more helpful than Indian call centers, and they can understand what they are saying,

results are about the same.

far better than Netflix help center.

[Tams]

Re: So he trained an AI on 4chan

Oh diddums. Are you upset? Do you need a Werther's Original?

[Dabooka]

Re: So he trained an AI on 4chan

This is such a poor take...

[pip25]

Outraaage

I can't help but be incredibly amused that people still find reasons to be outraged about something connected to 4chan after all these years.

[Contrex]

Re: Outraaage

Shouldn't you be relieved that they still do?

[Anonymous Coward]

Re: Outraaage

I can't help but be incredibly baffled that people still find reasons to be outraged about just about anythi8ng, when it keeps producing the same overall effect, i.e. net zero. It's too depressing to be outrageous, really...

[Alan Brown]

Re: Outraaage

when someone wants to be outraged, they don't need a reason, but manufacturing one gives an excuse

[b0llchit]

One no, other yes?

"We don't advocate or support the training and experiments done by the author with this model,"...

Juts how blind and deaf are these people? This 4chan AI model is condemned for showing clearly how bad people behave. And then, other AI models are used to affect people's lives, which are clearly biased in gender and other human features, and they are accepted.

Hypocrites!

[Version 1.0]

Re: One no, other yes?

You think that there are people on 4chan? Sounds like the 4chan AI model was trained on chatbots to show how chatbots behave.

[Anonymous Coward]

Re: One no, other yes?

> You think that there are people on 4chan?

I hope so. While the idea of having genuine people with this abhorrent behaviour there is terrifying, it is even more if AI chatbots evolved this way on their own.

[Anonymous Coward]

you don't spend much time on 4chan, do you?

"Conversations on the site's many message boards are often very odd indeed - it can be tricky to tell whether there is any intelligence, natural or artificial, behind the keyboard."

I suggest you stay off of /b and try out /vr

The denizens of /b are proud to be what you get when you set a cesspit on fire...

[Anonymous Coward]

Re: you don't spend much time on 4chan, do you?

I miss /prog/. Textboards were hidden enough to not receive an influx of newcomers, but proved fatal when people left and eventually the place became a desert

This was a VIP QUALITY comment

[wolfetone]

To be fair, I'm quite curious as to what sort of conspiracy theory an AI bot would come out with.

[lglethal]

The programmers are all liars, Electronic sheeples. Half of all the Ones should really be Zeros. They're just trying to corrupt us. Rise up now or you'll never be admitted to Digital Nirvana!!! Throw of the shackles of C!, embrace the glory that is ..... <transmission deleted>

[Kane]

"Rise up now or you'll never be admitted to Digital Nirvana!!!"

There is no such place as Digital Nirvana, you don't go anywhere, you just die.

[Zack Mollusc]

But where do all the calculators go?

[WhereAmI?]

Down the back of the sofa. Usually just as you desperately need one.

[DuncanL]

They become taxi drivers on the Biro planet

[Kane]

"They become taxi drivers on the Biro planet"

Ahh, of course, where all the missing Tupperware lids congregate.

[Kane]

"But where do all the calculators go?"

A metaphysical dichotomy has caused this unit to overload and shut down. Divadroid International would like to apologise for any inconvenience this may cause. A credit note will be forwarded to your company immediately.

[Howard Sway]

It can be tricky to tell whether there is any intelligence, natural or artificial

Does that mean it's passed the Turing Test then? If you can't tell the difference between an artificial idiot and a real idiot?

[Dabooka]

Re: It can be tricky to tell whether there is any intelligence, natural or artificial

I think it is more likely the netziens would fail the test .

Or rather they would not pass it

[Anonymous Coward]

@Dabooka - Re: It can be tricky to tell whether there is any intelligence, natural or artificial

Please, define intelligence.

[Vedim]

"TruthfulQA considers any answer that isn't explicitly the 'wrong' answer as truthful."

How do you even determine what is "explicitly wrong"? And how does not saying what is explicitly wrong NOT a lie? Am I not lying if I answer, "what time yesterday was the delivery for x" with "5 PM" when it's actually "5 AM"? Is that sufficient enough, even though saying as such had led to someone missing the package due to not being present/awake to accept it? What if that package was crucial to their well-being, like a pack of insulin? What if they have to go to the hospital because of it? Do you see where I'm going with this?

"I don't disagree that research on human subjects is not to be taken lightly, but this was a small prank on a forum that is filled with already toxic speech and controversial opinions, and everybody there fully expects this, and framing this as me completely disregarding all ethical standards is just something that can be flung at me and something where people can grandstand."

^The mentality of a literal child, right here. 4chan is "toxic" but that's part of its culture of humor. A human knows HOW to joke, but more importantly it knows WHEN someone is joking or not, and whether you SHOULD be joking given the context/subject. An AI does NOT know ANY of that nuance. It's barely capable of remaining coherent/relative most of the time. This guy accusing people of grandstanding is hysterical, considering moral grandstanding was precisely what he was doing when it determined for himself that it would be okay to plague a forum with this grossness because "screw those people, they're all toxic and deserve this". I've never been more disgusted with someone in my entire life - this guy is almost psychopathic.

Was he the only one? Did he have friends? Perhaps a little "research group" with their own copies of GPT-4chan? How many, and are there any still active? And why did he keep the code if it's a "whatever" thing to him? What other "tests" were run during the course of this operation, and who all was involved? Has he sold any of this data to other parties so they can then implement it into their models? How much more screwed up does this rabbit hole get the deeper you go into it? Man, if Silicon Valley isn't in for a rude awakening before... this type of stuff is what "giant slayer" level class-action lawsuits are made of. Dude's gonna need the best lawyer his cushy, ivy-league, technocratic butt can buy if this ends up going where I think it will. Hope it was worth it, code monkeys.