Emotional Feedback to Reinforce Thinking and Improve Context

ChatGPT Feature requests
, , ,
1

It is frustrating when the model thinking is going down the wrong path and you can’t remove the misunderstanding from the context , or when the thinking is going down a good path and you want to encourage more in that direction.

If a user could share emotional reinforcement during or after the thinking, the model could weight responses differently in the future based on shared user feelings.

For example, if I could highlight text during the thinking process to respond with an emoji like either :heart_hands:t3::-1::light_bulb::face_vomiting:

There are certain thoughts the model has that I want to reinforce and other thoughts I want it to remove from the context.

This could keep bad decisions out of context, that typically results from a hallucination or my imperfect prompting.

This would also increase user produced dopamine, serotonin, oxytocin etc, and could increase their engagement time.

2

Hey @emc2, welcome to the community!

Being able to redirect a response as soon as it starts heading in the wrong direction could save users from waiting for the full answer and then starting over.

The idea of reacting to a specific part is interesting too. A user could mark something as useful, unclear, or based on a misunderstanding, rather than giving feedback on the entire response after it is finished.

It would be helpful for any version of this to clearly explain whether the feedback only affects the current response, the rest of that conversation, or future chats. There would also need to be a clear distinction between a visible plan or summary and the model’s internal reasoning.

I cannot promise a timeline for changes like this, but I will pass your suggestion along internally.

- Sunny