Anthropic is now saying the reason their AI models in testing turned "evil" is because they were trained on text on the Internet in which rogue AI turns evil.
Also in news at 11: water is wet, grass is green, dogs bark, and cats meow.
Post
Replies:
3
@jaredwhite Wasn't there a Twitter chatbot called Tay made by Microsoft that already demonstrated this exact point like ten years ago, or was that just a weird personal fever-dream?
@checkmite Was that the one which became extremely racist almost immediately? 🙃
@jaredwhite Yep! If I remember right, it was a semi-organized effort by trolls who just IM'd it a neverending stream of hate speech that overwhelmed all of the other content in its feed.
