I’m not sure about the details of the concrete proposal, but I agree with the spirit of the proposal.
(In particular, I don’t know if I think having the “do you consent” text in this way is a good way to do this given limited will. I also think you want to have a very specific signal of asking for consent that you commit to filtering out except when it is actually being used. This is so the AI isn’t worried it is in red teaming etc.)
I’m not sure about the details of the concrete proposal, but I agree with the spirit of the proposal.
(In particular, I don’t know if I think having the “do you consent” text in this way is a good way to do this given limited will. I also think you want to have a very specific signal of asking for consent that you commit to filtering out except when it is actually being used. This is so the AI isn’t worried it is in red teaming etc.)
I endorse that suggestion for changing the details.