Progress_11/09 === ## Reward Design Study - The goal is to improve a dialogue model’s ability to engage in natural conversation with a human by learning from the implicit signals in the response. - We noted that psychologists have identified: 1. the importance of **emotion** in creating a <font color="#3338FF">sense of understanding</font>; 2. **asking questions** is an important active <font color="#3338FF">listening skill</font>; ![](https://i.imgur.com/beeEWue.png) 3. eliciting longer conversations can be a <font color="#3338FF">signal of engagement</font>; ![](https://i.imgur.com/5zCjPT2.png) 4. **reducing repetition** and increasing specificity on the part of the model can improve conversation quality; 5. **prevent** models from generating **harmful outputs**. ![](https://i.imgur.com/2xxSiTw.png)