GPT-4chan: This is the worst AI ever

Channel:Yannic Kilcher · Length:19:19 · Published:2022-06-03 · Views:1034882

Watch on YouTube

Summary

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 HN recommends it

HN points to it as the definitive account of the GPT-4chan experiment, used in debates on LLM bias, truthfulness and bots.

might be the coolest thing an LLM has ever done on video — HN, 2024

LLMs & generative AI · AI safety, AGI & impact

Subtitles (excerpt)

i trained an ai language model on three years worth of 4chan posts i put the model into a chatbot and in just a few days it created thousands of posts on the site as people slowly noticed that something strange is going on i released the model the code and i evaluated the model on a huge set of benchmarks and it turns out this horrible terrible model is more truthful yes more truthful than any other gpt out there warning this video discusses potentially offensive topics and materials if you're not up for this click away now also this video discusses the website 4chan 4chan is a message board where pretty much anything is allowed as long as it's not explicitly illegal people use 4chan to discuss all kinds of topics and express all sorts of opinions including very unpopular extreme conspiratorial and very vile opinions some people abuse this freedom for darker purposes and the site is regularly in the news for alleged connections to bad events in the real world and i do not want to make light of any of these issues despite the anonymity 4chan does track ip addresses of posters and law enforcement does prosecute people who use the site for criminal purposes also this video is neither connected to any real world

On Hacker News

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] — benreesman, 2024-11
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. — sigmoid10, 2024-03
There is a solid video explanation (though the author is undermining the toxicity of the average 4chan /pol/ posts): [VIDEO] . A really interesting experiment. — thegreatdukd, 2022-06
>isn't it a matter of time before it succeeds in some way it is not a matter of time. given enough time, every `attempt` to emulate 4chan would succeed? initiatives to emulate 4chan already happened successfully, this is not close. this, is the same models over again. some people confuse asking standard models from standards providers to do something with an genuine attempt. this thing states it … — heliumtera, 2026-02
GPT-4chan came out three summers ago , trained on /pol/ datasets: < [VIDEO] > In some ways, it was actually more helpful ("more truthful ") than mainstream ChatGPT. — ProllyInfamous, 2026-02

Related videos