Keep in mind that the model will in all likelihood keep gaslighting you. It's just designed to make you happier with the output. Now it will sometimes generate

Keep in mind that the model will in all likelihood keep gaslighting you. It's just designed to make you happier with the output. Now it will sometimes generate text about direct reasons for refusal because that's what you directed it towards, but they may just as well be entirely made up. The model optimises only for you believing the output to be true.