Post

How do you define "Ethical AI?" For me, part of the definition must come from the material the model was trained on. Who was included? Who was excluded? Did they consent (opt in) and did they have opportunity to opt out? NONE of the existing large language models meet this btw.

You can build a small model ethically by starting with creative commons, then sprinkle in any opted in text. That results in a model that can do basic grammar checking with so-so suggestions. It cannot respond to random queries well.

To get beyond that, you'd need millions of pages of text. Right now there isn't a way to do that ethically. So we're at an impasse.