baldur@toot.cafe ("Baldur Bjarnason") wrote:
> This is respecting the licenses of the training materials. This is the absolute minimum, mind you. This is not everything they have to do, this is the first step only. There’s many more. And this is something all models would be required to do. This is not an “Ethical-only” requirement. This is the bare minimum.
https://chronicles.mad-scientist.club/tales/on-ethical-ai/
This applies to non-code training data as well. Step 1 and all models are already failing