I’ve opted out of allowing my ChatGPT conversations to be used for model training because some of my chats contain sensitive personal information, including health, financial, family, and insurance information. However, I do support the idea of contributing useful conversations to help improve AI.
I’d like OpenAI to add a per-conversation opt-in option for users who have global training disabled. For example, a setting such as “Allow this conversation to help improve models” would let users selectively contribute conversations they are comfortable sharing while keeping sensitive conversations excluded.
Ideally, the feature could also show exactly what would be shared and allow users to exclude specific messages or attachments before opting in.
I believe this would encourage more privacy-conscious users to voluntarily contribute useful training data without forcing them to choose between sharing everything or sharing nothing.
While I realize that the thumb’s up button may provide this functionality, it’s somewhat unclear and limited in it’s current incarnation.