You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
ChaCha's conversational flow is designed as a state machine, where the system stays in one state and proceeds to the next one when meeting the state goal:
Based on the conversational flow and the current dialogue, an LLM is prompted dynamically. For example, receiving the child's message in the <spanclassName="p-3 py-0.5 rounded-full bg-[#85c038] text-white">Label</span> phase, (1) the conversation analyzer <spanclassName="circle-digit">A</span> analyzes the current dialogue <spanclassName="circle-digit">B</span> and extracts a structured summary <spanclassName="circle-digit">C</span> of what emotions are identified and whether ChaCha has acknowledged them. Combining the incomplete piece of the summary <spanclassName="circle-digit">D</span> as well as the summary data from the previous phase <spanclassName="circle-digit">E</span>, (2) the system formulates a new instruction <spanclassName="circle-digit">F</span> for the response generation. (3) That way, the LLM <spanclassName="circle-digit">G</span> generates a response <spanclassName="circle-digit">H</span> explicitly steered to empathize with the child's regretful event.
0 commit comments