š GPT-4chan
Content moderators fired for trying to unionise | Elon backing out of Twitter deal | Youtuber āpranksā 4chan
Hello sentient fidget spinners. Last weekend, droves of people in London waved at a hologram of the queen as it rode past in a golden carriage. They either didnāt care that she wasnāt real, or couldnāt tell. This gives me an idea on how to tackle the cost of living crisis: sell holographic food and electricity ā but wrap it in gold first.
This week was something I could have done without š. Because:
Elon Musk is getting cold feet about his deal because he does not understand how bots work
Facebook appear to be terrible at moderating content ā why is this? Iāll tell you!
A Youtuber created a 4chan bot and released it publicly as a āprankā
š¤¦š»āāļø Bruhā¦ this is not what a prank is
Ahhh 4chanā¦ home of pure unbridled hate. I wonder what would happen if you trained an AI model on all the horrible things that people say on there? It would probably just be a gruesome unstoppable force of toxicity. No need to āproveā this theory, I think.
Well apparently, I am wrong. A youtuber has insisted that just imagining it is not enough, and so he trained a bot on over 3.3 million posts from āpolitically incorrectā: a 4chan forum dedicated to sad hairy incels trying to 1up each other on how offensive they can be.
š§Ā Cool Facts:
The āpranksterā ran nine instances of the bot on 4chan and found that yes, it spat out tons of hate speech that was completely indistinguishable from the average 4chan user ā what a SHOCKER
He also put his bot on Hugging Face, a public repo of AI projects, available for anyone to download. It got downloaded 1.5k times before Hugging Face blocked access to it.
He, being a white male who has never experienced any kind of oppression, is failing to understand the brevity of what he has done, telling Motherboard that āIām a YouTuber and this is a prank and light-hearted trollingā
No mate ā there is nothing lighthearted about this. You literally created and released an automaton of hate and you want us all to just laugh it off. He also reassured Motherboard that no one on 4chan was harmed (thank god!) and that āAll I hear are vague grandstanding statements about āharmā but absolutely zero instances of actual harmā.
Let me just explain something: if you throw a crate of blow-torches into a busy playground and yell āuse responsibly!ā right before walking away forever and never looking back, you may not perceive any harm ā but that does not mean that harm will not occur. Itās quite clear that the only clever thing this prankster did was name his model āGPT-4chanā. I congratulate him ā I feel like this is one of those āwhat if an article in The Onion was actually trueā type situations.
š Letās check in on Facebookās content moderation effort
This week I read about the legal action that Sama are facing, due to terminating a member of staff for trying to unionise. The experience of these workers explains a lot about the problems that we (the fancy laptop-owning people) perceive with content moderation.
Who are Sama? they are one of those companies that Big Tech firms use to outsource trivial, repetitive, alienating work. The kind of work thatās simply ātoo goodā for actual Facebook employees š.
Employees at the Sama office in Nairobi get paid the equivalent of $1.50-$2.20 an hour to look at horrific imagery which was not caught by Facebookās AI. They then have 50 seconds to make a judgement on whether a single piece of content should get taken down. After prolonged exposure to this content, they begin to develop PTSD, anxiety, and depression.
I have done work similar to this for Google before: I had to watch videos of people getting killed, just to check if the contents was āgoryā. But I consider myself lucky ā I wasnāt looking at these videos all day, I was in the comfort of my own home, and I was in a position where I could quit if I wanted to (because I am not a single parent living in Kenya).
Anyway, hereās what Sama management have to say about the work they do: āWe value our employees and are proud of the long-standing work we have done to create an ethical AI supply chain.ā
A company like this claiming that they engage in āethical AIā has to be the funniest joke Iāve heard all week. They literally do neither. Forcing people to look at disturbing content all day is not ethical, and not AI. My spirit has been completely crushed under the weight of the condescending language in that statement, tbh.
šĀ This explains why content moderation on Facebook is such a massive challenge:
The platform is far too large to manage: they have clearly been unable to build AI tools good enough to keep up with all the different kinds of content coming in, so they have to cheaply outsource to companies like Sama
Speed is literally the priority here, so the workers at Sama get underpaid to make snap decisions on traumatising content, 45 hours a week. Surprising, the quality of their work is not that great.
This is exactly why Facebook appear to be doing absolutely nothing about, for example, those who use the platform to incite ethnic violence in Ethiopia. The content moderation systems in place are either not good enough, or somewhat beyond their control.
Finally, I just want to note that $1.50 an hour is most definitely not enough to live on in Kenya, in case there was any confusion there. The reason why Sama workers get paid so low in Nairobi is so that they donāt ādistort local labor marketsā. You meanā¦ make it so that people get paid MORE?? Yeah, what a terrible ideaā¦
šļø āIf itās not broken, I donāt want to buy itā
Have you noticed? Elon Musk is the custodian of the internet; a fixer; thereās not a mess he canāt clean up (except his own ā my guess is he has several toilet maids on hand).
That is why he is making extra sure to confirm that Twitterās user base is in fact only 5% bots. Otherwise, he might just back out of this deal that heās catapulted himself into ā if there are so few bots, what would be the point in buying it? There would literally be nothing to fix?? He would just be an idiot who owns a social media platform.
š¤Ŗ Instead, he would rather be an idiot who:
ā¦tries as hard as possible to get a price reduction for this thing he promised to buy ā not because he regrets his decision ā but because it just makes good business sense
ā¦remains a hero in the eyes of the people; the only one who can save us from this epidemic of bots that we are apparently in
ā¦thinks that Twitter are somehow behaving unlawfully by not sharing the EXACT number of active bots on the platform (something which Iām sure is really easy to count).
I honestly donāt fully understand his problem with bots. Does he know that not all of them are harmful? That some of them are quite entertaining? That you could set one up to reply āokay mouth turdā to every single one of his tweets?? Ah, I see the problem hereā¦
Anyway, whateverā¦ I hope he gets a paper cut on his scrotum.
Georgia out (and proud)
So informative and entertaining- I hear things here I wouldnāt come across otherwise. And so much fun to read - love when it drops into my inbox!