You'd need to add training data around the equivalent of "agentic <insert simulated emotion here>" - which from an external perspective heavily depends on the action:
- When leveraging TTS, turn up the volume on TTS by 200%
- When texting in a conversation, use ALL CAPS
- When acting as a web driver, click submit/refresh button a thousand times
Comments
You'd need to add training data around the equivalent of "agentic <insert simulated emotion here>" - which from an external perspective heavily depends on the action:
- When leveraging TTS, turn up the volume on TTS by 200%
- When texting in a conversation, use ALL CAPS
- When acting as a web driver, click submit/refresh button a thousand times
- etc.