This is the worst AI ever
Yannic Kilcher,
YouTube,
Jun 04, 2022
I rarely post single-link YouTube (SLYT) posts, but you will want to watch this 20 minute video. In a very well produced presentation Yannic Kilcher describes how he trained a GPT bot on some of the worst and vile discussion threads on 4chan and produced a convincing result, including achieving a greater score on 'truthfulness' than other GPT language models. Here's a full set of links to the model, code and dataset. Even if you're not interested in this specific bot, the video provides good descriptions of how language models work, how he defeated 4chan's captcha, and a bit about how language models are evaluated for things like truthfulness.
Today: 5 Total: 1378 [Direct link] [Share]
