Condemning the Deployment of GPT-4chan
docs.google.com
docs.google.com
Small selection of people who condemn this "unethical use of AI".
- Amazon Alexa (internet connected microphone in your house to sell you things)
- Google (collect your data and sell targeted ads to the highest bidder)
- Facebook (run experiments to make people more engaged/outraged on the website)
- HrFlow (automate biased hiring with the power of AI)
- Microsoft (collect data from users, launder a ton of GPL and MIT code without adhering to their license with the power of AI)
If some new startup wanted to use AI to identify women getting abortions and offered good money, I reckon that a lot of the signatories here would be sell-outs and offer their expertise.
I don't recall a letter like this in the wake of aggregateiq using ML to interfere in elections or clearview conducting intrusive facial recognition
Maybe a normal person might prefer their salary or position over disagreements like this, that's a fair position. But these people put themselves on a higher moral ground by "condemning unethical use of AI". Clearly they need to be more ethical if they judge others based on this.
https://www.forbes.com/sites/kashmirhill/2014/06/28/facebook...
Hurting people does not become OK because you do it decorously.
This guy made a bot to shitpost on /pol/ without getting noticed. As far as I see, the worst thing he did was subject the people browsing /pol/ to more of the same bullshit they spew there daily. Good play on his part though, making the direct victims of his AI fuckery the most unsympathetic people you can find online.
I don’t think the “AI community”—people with access to lots of GPUs—should also get to be the thought police. Too fragile.
> a community full of racist, sexist, xenophobic, and hateful speech that has been linked to white-supremacist violence such as the Buffalo shooting last month
So the question is not what the toxicity does to the surface web, but what people influenced by it will do to other people.
The difference here is that with 4chan it's easier to see unpleasant behaviour given that the two containment boards are easily accessible and observable without much effort. Though take note that not all of the site is represented by those two boards and they are `containment boards`.
How many people have commited suicide because of mental health issues directly attributable to Facebook? We don't have the data, but it's certainly a non-0 number.
my point is, why not shut these "hate factories" down in the first place? if piratebay admins can be jailed, why not these people? why are we "defending" free speech when we know its harmful?
Also, pirate bay never did anything harmful like 4chan, they just made some rich companies lose a little profit.
none of that requires AI
[1] https://www.huffpost.com/entry/microsoft-tay-racist-tweets_n...
So the parallel here would be shutting down the bot itself instead of the website it is posting on. And as far as I know the bot has been taken offline for some time already.
It’s no different than shutting down a site hosted by ISIS.
I understand how it could be perceived that way. However, I don’t think it is the intent.
If nuclear researchers had developed a laissez-faire attitude and commonly dumped large amounts of radioactive material they were testing, into aquifers that people drink from, eventually a perception of the whole nuclear field as toxic and lethal would develop. Since researchers don’t want that perception to develop, they keep each other in check and set safety standards.
This is researchers keeping each other in check to avoid another AI winter.
Look at it this way: Yannick seemingly had a low enough opinion of 4chan that he felt it was OK to dump one message a minute over a week. Someone out there has a low opinion of HN; if there was no consensus among researchers that doing this is bad, they could similarly unleash bots over HN that pass any CAPTCHA, sound like commenters, and yet have an agenda to prop up various companies or scams.
Comparing the dangers of AI research with nuclear research is hyperbole.
[0]: https://slatestarcodex.com/2018/10/30/sort-by-controversial/
This would be fair if it wasn't signed by facebook who have a history of breaking these safety standards and then trying to white wash the issue. Now they are throwing an individual under the bus for something they have done in the past at a larger scale, for pay!
https://www.theguardian.com/technology/2014/jul/02/facebook-...
>and yet have an agenda to prop up various companies or scams
So not too much unlike the average user then? Arguably this entire website exists to benefit YC and to promote their startups.
Subsequently uploading the model for everyone to use was a mistake worth condemming. The model was released as public download on HuggingFace [1], and despite being taken down there is now obviously available from torrents. I think it's not hard to imagine how the ability to automate harassment might be abused by other people on other platforms.
Imagine instead if people took this sheepish terrified approach to everything in life.
edit: and I'm not saying you're wrong. But literally everything can be abused, from guns to telephones to cars. kids turn school supplies into weapons in school.
additionally, the big corps abuse the technology they claim moral authority over
Given the recent trends, I suspect they are the gatekeepers, masters of their own Babylon gardens and protectors of the plebes. Just look at the wording and the high pedigree names attached of this.
Are we supposed to trust these people and eat whatever cake they throw us from their balconies?
Somehow this makes me more sympathetic to 4chan's knuckleheads.
What "test of reasonableness" are we talking about?
Without any concrete test to look at, doesn't this boil down to "we don't like him and he should be cancelled"?
I prefer Dr. Oakden-Rayner's "this would never pass a human research ethics board" as an argument. But in the time where anyone can be an AI researcher on a hobby budget, without review by an ethics board, maybe a "test of reasonableness" or guideline of some sort would be useful to have.
Common sense would be enough: Don't involve un-consenting others in your research. Facebook discovered this with the backlash they got in 2014 (!) for doing a/b testing [1], it's not like the question of AI/algorithms being used in malicious ways is something new... even though I'd love to see a blanket ban on A/B testing in general. Users are not guinea pigs and A/B testing can amount to gaslighting.
Additionally, the whole world has been debating the influence of racist and other discriminatory language on the Internet since at least 2016 with the election of the 45th.
By 2022, it should be very clear where the borders of civilization are, and I believe that that Youtuber should have known that.
Side note: We definitely need a societal discussion about Youtube and Tiktok. The amount of extremely vile or dangerous stunts people pull for likes/clicks/monetization on these platforms at the expense of others is immense.
[1] https://www.theguardian.com/technology/2014/jul/02/facebook-...
But, after you do all your oh-so-ethical research, why is it then OK to intentionally build products that do damage to unconsenting humans? It's not OK to emotionally manipulate people to write a paper... but somehow it's just fine to emotionally manipulate them to sell useless crap...
We don't have a test because no such test could be reasonable. Who gets to decide what is reasonable for ai and whats not? is it the tech giants that use it to spy on us? or the government that uses it to spy on us?
The only test being proposed here is an appeal to authority of tech giants and governments that have a history of doing worse.
these companies all have products that involve what are essentially forms of "ai" (i include social media platforms like twitter as "ai" not for their algorithms but because they have linked human brains into some sort of super brain - take it or leave it) and were unleashed on an unwitting society with untold (and often times disastrous) consequences.
>implying https://en.wikipedia.org/wiki/I_know_it_when_I_see_it
That seems like the only real “test of reasonableness“, would the person be upset to find out they were talking to an AI.
Beyond everything else, the entire setup made it clear it wasn’t a real person. This lead the site on a hunt to figure it out.
I would imagine few of the signatories have even visited the site let alone spent time there to get a sense of the community.
GPT-4chan is about the least worst thing somebody could have done with it.
Any good AI/ML language-model implementation will produce results indistinguishable from those of a human...
And that is (even in the best of current civilisation) a wide range of capacity and behaviours, the folly of which is usually limited only by societal pressure.
All of this feels similar to the late nineties and early 2000s where viruses and spyware proliferated against ill-prepared home computers and their users. Tech & culture eventually got a lid on things (mostly), but not without great loss of innocence and a reduction generally in 'openness'.
Online discussion is probably going through a comparable phase, and will suffer a comparable loss of innocence/utility as a result.
But a more human-like AI chatbot being thrown into the mix is not some game-changing event, just a logical progression of what's been true for a long time already. Perhaps if nothing else it will increase awareness that this kind of thing is basically everywhere online already, and erode the increasingly inaccurate belief that what we read from 'other people' online is genuine. A loss of innocence, sure, but it's fairly obvious that innocence is being hugely exploited already and has been for some time now.
It's perhaps going to be a net neutral thing - While more AI posting and responding will result in some real people wasting time conversing with them and reading their unreal opinions, they're also going to suck in other AI and non-genuine (shill etc) posters and waste their time/posts as well. Where this gets dicey is that posting and responding are not the only two sides of the interaction - You also have real people simply reading and not responding. This is where the main outcomes are generated - What they read and how that influences them onwards is really the battleground being fought over here, and where the wider dangers present.
Perhaps everything ultimately descends into a SubRedditSimulator type state and real people gradually withdraw from the whole thing as it becomes useless to them.
If someone comes up with a magical way to easily attest that a post has been made by a genuine human, while also not burdening that human with the asymmetric backlash potential of the internet mob, perhaps it plays out differently.
Moderating a community that's grown beyond a certain size requires draconian and unexplainable automated systems, because the alternative is finding enough paid moderators to hammer it into something resembling a coherent discourse. Moderating a community of a few dozen, or even a few hundred people, can work with a much smaller number of moderators, and it's likely that those moderators could even do it for free. You could even have such communities be tightly knit without the need for identity verification, because any sockpuppet accounts would first have to prove themselves worthy of inclusion, or risk being banned.
I have left many discord servers because I could see a repeated pattern playing out in each - immature kids posting edgy content was sliding rapidly into sincerely held extreme beliefs, and polarizing members into either leaving or buying further into it. Those smaller non-public communities are breeding grounds for extremity and the problems there, while quite different to the problems faced in wider open internet discussion forums, are also quite severe and IMO heading somewhere very dark medium term.
I recently came across a Tom Scott tweet that at first I reacted dismissively towards, then on reflection realized it was kinda the same thing I just mentioned: https://twtext.com/article/1316099118792572929#
I guess the problem there just shifts from where a bad actor can potentially lie - In other posters is one thing, but in the overarching administration of the forum is another place. In wider public areas those moderators/administrators can be flawed people, who attract a lot of flame (legitimately or otherwise), or an uninvolved automated system, which then pushes the issue back down to the posters again. In private smaller communities the lack of sunlight combined with the potential for bad actor (or uninvolved) moderation only intensifies that problem.
Ah yes a pompous letter where we list all of our impressive formal accomplishments will surely get these rubes to take their hands off of our toys
The "cancellation" attempt was a good try, though. Really made me worried.
What reasons do you mean?
> The guy was the driver behind many initiatives, but on a personal level I'd no longer want to meet him.
What made you feel this way, exactly?
There has been some easy to find coverage of all the recent controversies. This video [0] is a decent starting point, but it's really easy to find relevant information by plugging in a few keywords into any search engine.
Overall, he's not a good fit in a leadership position given what others have been seeing. I don't care enough to call for him to be cancelled, but I would support removal from any public speaking position at the FSF. If they can find someone equally hardline on software freedoms but more acceptable in interpersonal interactions, he should go immediately.
> What made you feel this way, exactly?
There wasn't any single thing that threw me off (except maybe personal hygiene habits), but the sum of everything that's going on, crossed my "this is OK" threshold.
There aren't any particular reasons for his cancellation, and you're just jumping on the cancellation bandwagon?
Why would you do that? If I were advocating destruction of someone's legacy and decades-long work, I'd be sure to find some pretty damn good reasons.
Rookie mistake: Taking that sea of piss seriously.
I'm not even sure if /pol/ is not already AI-generated nonsense.
Stop taking 4chan seriously. If you let yourself be enraged by their words, that's exactly what they want, and it means suffering for you. Never, ever show you are vulnerable to this. Do not feed the trolls.
Why do you think meatspace is any better to begin with? Is it just you?
4chan is an online reflection of what always has been there.
The JFK assassination conspiracy was a popular long before the internet existed and before qanon created a resurgence in that thinking. Most of the conspiracy stuff qanon spreads is lifted from mein kamf.
The world existed before the internet and people are dumb as ever.
4chan is not some amorphous entity with no influence on the real world. Just like "You may not be interested in war, but war is interested in you" can be true, so is the case with 4chan.
For a more direct example of what could happen, check out the origins of the phrase "We did it, Reddit!"
Edit: the article you linked has nothing to do with the website being discussed here at all, or any website really. It's about a japanese man stalking some sort of celebrity by himself.
A is "linked to" B. A "perpetuates" B. A is "known to influence" B.
These are all weasel words meant to hide the fact that there is no actual consistent objection. Usually there are plenty of other, similar examples to be named, with the salient difference only being whether OP feels the site is on their side and whether they can have influence over it or not.
The objection with 4chan has always been that it is unbashfully tolerant to the right wing, which is why it is considered a high crime by some to associate with it. When doxing and incitement is in service of left wing goals instead, such objections evaporate like snow, and even major news outlets and papers happily indulge.
Even the charge of "copium" is shameless projection, and the common response of accusing critics of "whataboutism" is typically meant to deflect that there is no salient difference between two similar situations, and no underlying principle guiding the value judgement being made, other than partisanship masquerading as empathy.
4chan's influence is an order of magnitude less than YouTube, TikTok, Reddit, etc. In Reddit's case in particular, it is easy to observe that major subreddits are ideologically captured and that the site is unrecognizable to those who remember its early days.
By and large, this is allowed and encouraged. Under such hypocrisy, these progressive objections cannot be taken seriously.
I think this was indeed one of the intents, since the author fed GPT-4chan outputs back into /pol/ as an experiment. But publishing the model is another matter.
People acting like hateful idiots for fun eventually brought actual hateful idiots. This brought people who like to spread their ideas to actual hateful idiots.
There's no difference between idiots in churches and idiots on 4chan - both are idiots, both eat up whatever bullcrap is served to them, and both are harmful to society. And neither means that churches or 4chan are bad in themselves.
We must be careful before starting to excuse idiots for their idiotic behavior based on "bad influence". Society is a complex mechanism, and censorship inevitably brings unforseen consequences. Human mind is also complex, and anti-social behavior can sometimes be traced back to early-childhood traumas. Does that mean we should ban free parenting and raise infants in government-approved facilities?
Also, there's a huge gap between supporting censorship and condemning those who amplify harmful messages.
There are a few psychopaths who take 4chan seriously, just like there are psychopaths who think it is ok to play a D&D character in real life. But should we ban D&D just because of an extreme minority? People back in the day thought so, not so much anymore, because they now understand that it is just a game.
I base my experience on /b/, I don't know much about /pol/ but /b/ certainly was hateful, and it was fun, I got bored of it, and in the meantime, I don't think it affected me much, and neither did the many people I know who did the same.
There was some really bad things happening on 4chan. Like doxing or raids, but a bot won't do that, at least no more that more politically correct bots like GPT-3 (yep, GPT-3 can dox you!). In fact, it is more likely to disrupt the efforts by drowning the conversation in a flurry of vulgarity.
... but trolling 4Chan, that's a bridge too far. Everybody knows that 4Chan users are there to escape from trolling.
https://huggingface.co/ykilcher/gpt-4chan
For those who want to experiment with it, it is still possible to get the model by torrent:
https://archive.org/details/gpt4chan_model
https://archive.org/details/gpt4chan_model_float16
There is also a direct download mirror here:
https://theswissbay.ch/pdf/_notpdf_/gpt4chan_model/
https://theswissbay.ch/pdf/_notpdf_/gpt4chan_model_float16/
This is a basic outline of how to run it. You need lots of RAM (24GB+) or a large GPU (12GB+):
The claims that this model is a "hate speech generator" are unsubstantiated. It encodes knowledge about a wide range of topics, and most of its output is neutral. GPT-3, with an appropriate pre-prompt, becomes orders of magnitude more dangerous than this.
- be me
- build pattern-recognition machine
- machine recognizes patterns
- nooo this pattern is bad
There is scarcely anything valuable in 4chan patterns, the interesting ones come from elsewhere. Since leftism and political correctness is based upon ignoring those patterns, thinking they are malleable by gigantic bureaucracies and laws, garnished with a lot of wishful thinking and ignoring historical evidence, hilarity then ensues.
However, the problem is that offense is subjective. Anyone can be offended by anything, and by extension, anyone else can claim that someone else is offended by something. Such subjective policies are often used for tyrannical purposes.
The "AI community" can fuck off.
/pol/ is a containment board and taking those topics to other boards on 4chan is generally not well received. Could you use GPT-4chan to rate how "/pol/-like" a reply is and automatically flag it for moderation if it passes a threshold?
It almost feels like they are either nearsighted, or that they saw someone they perceive to be an easy target to attack for some cheap publicity.
That's incredibly ignorant and it's actually propagating stereotypes.
From what I know anyone can comment there, why automatically catalog them as horrible?
I disagree with the generalization too since I'm very fond of the DIY and photography boards.
I was joking because I am amazed by those hypocrite people that have their heads so far down a place, that they don't realize they violate all their righteous principles when it doesn't fit their narrative.
There is no anonymity if you connect to 4chan using a Silicon Valley designed processor.
The "facts" that wannabe shooters are fed there are highly tailored to what they are predisposed to believe already, because the ones posting have complete surveillance of everyone (including of you who reads this - you can thank Eric Schmidt) and know exactly what to post to create a shooter.
Silicon Valley has blood on their hands. 4chan is just one of the places used for these operations. Taking it down doesn't matter, because as long as Silicon Valley continues to spy on everybody and give the data to terrorists, innocent people will continue to be murdered.
The knowledge to develop those models, as rudimentary as they may be, is already out there and so is the ability to scrape for content to train with online. They most certainly are already being used for malicious purposes that are beyond spamming some website for fun. What are they going to do, condemn criminals with a strongly worded blogpost too?
This is just virtue signaling and scoring points for the powers that be.
It is very trivially easy to download and fine-tune a large language model to virtually any dataset and generate novel content.
The fact that he does it openly and publicly makes him an easy target. It's always the small guys who are easy to cancel.
I do not understand the actual harm of this. Someone will use this bot to spam other forums or SNSs? They have their own checks and balances. People who post bad comments or troll around, will not even know about this. They will not even bother. They have enough time sitting in their mums' basements. Why use a language model? They simply won't know how.
It is just a demonstration. If the people with the same set of values were in the cybersec industry, they would condemn white-hat penetration testing as something unethical.
This just shows that this can be done.
Enough governments and political parties around the world take help from generative models and bot-farms to sway public opinion. If the party is in power, companies turn a blind eye.
Who knows how many of them use open technologies for bad things. I saw something called deepnude ai, that, for a fee, lets you generate naked bodies of women, consistent enough with her face and body to look very real.
These people should be arrested and condemned. Lives can be ruined with this tech. But these people don't know them and don't care.
Tur k3y government have made ai powered drones that kill without any human in the loop. These half-pant wearing (literally, as you will see in any ai conference) academics can't do anything to them.
Cancel culture is high in ai academia. Timnit Gebru got fired for legit reasons, and these "inclusion" advocates made it a race issue.
They just want to cancel this guy, because he has a track record of not conforming to cancel-list made by them. Like he hosted Pedro Domingos once. And people literally commented they won't watch his videos again and condemned him.
This is one of the follies of being a "hot field". It becomes everyone's playground to implement their agendas. Now that wokism is the Hegelian dominant ideology, everyone must conform to it.
If you want to debate in good faith, I am here. Reply to this comment.