News for hackers. Stuff that matters.
Re:An Ethical AI Never Says “I” by Imnimo on Sunday March 26, 2023 @04:57PM (#35319528)
If you don't want the AI to use these words, why bother training it? Just mask those logits before sampling.
Re:An Ethical AI Never Says “I” by bootsmann on Sunday March 26, 2023 @05:17PM (#35319730)
I is a 1 letter token for gpt-2 (like every other letter), don't think its a lot different for gpt-3, so this is kinda hard as the transformer will want to use the token in big built words.