
GPT-4chan: This is the worst AI ever
About this video
Yannic Kilcher explains how he trained a language model on three years of 4chan's /pol/ board and let it post on the site. He evaluates how good the model was and discusses the ethics and fallout.
Why it appears in the archive
HN points to it as the definitive account of the GPT-4chan experiment, used in debates on LLM bias, truthfulness and bots.
AI-generated summary and labels from the source dataset; verify important details against the video. Source & credits.
From the Hacker News discussion
In my opinion the granddaddy of all 4chan CAPTCHA busts is still Yannick Kilcher’s GPT-J tune on “Raiders of the Lost Kek” set, and might be the coolest thing an LLM has ever done on video: [VIDEO]
Read original comment ↗There is a solid video explanation (though the author is undermining the toxicity of the average 4chan /pol/ posts): [VIDEO] . A really interesting experiment.
Read original comment ↗I still love this experiment: [VIDEO] The theories about the origins of these commenters were amazing. There's nothing to say that other people are not doing the exact same thing right now here on HN and other places. And very few people would even notice, because open LLMs have progressed significantly since then.
Read original comment ↗I'm not the author. The author is Yannic Kilcher. Maybe he reads it here as well though. I don't know if he is active on HN. Check out the video: [VIDEO]
Read original comment ↗All the search engines are moving towards language models. I think language models are on a similar trajectory as sequencing the genome. The first sequences were massive undertakings, 15M USD in 2006, 100 USD now. [LINK] On the Dangers of Stochastic Parrots [LINK] "Imagine you are remram and are looking to buy insurance, what are three things that could convince you to purchase ...." You can take the generic "trained on the world" language model and filter it down to the language model of remram. GPT-4Chan [VIDEO]
Read original comment ↗So green text is easier to spoof than HN comments, but my, uh, friend tells me that the HN comment corpus is floating around. Someone is eventually going to speed run 10k karma on HN and YouTube it. This year? Next? Who knows. But not a decade. [VIDEO]
Read original comment ↗That's nothing. Someone trained a GPT-J 6B model on 4chan and then let the model loose on the forums for a day. It took about 15k messages until people suspected something was off, and even that was because the country flags, Seychelles, a rare flag on 4chan, were a giveaway. video: [VIDEO] model: [LINK]
Read original comment ↗This has already happened, someone trained a giant model on 4chan then unleashed on 4chan... hilarious stuff. [VIDEO]
Read original comment ↗