CotChat

CotChat is a community chatbot that determines its conversational behaviour through REACT method based on ChatGPT and LangChain. Unlike ordinary chatbots, CotChat uses the approach proposed by React. When receiving messages, it automatically infers the appropriate action through observation-interest evaluation-reflection-action. We have defined three types of behaviours: Direct reply: Reply directly to the current message, Relate reply: Related to memory vectors and Silent: Feel this message is irrelevant and feel uninterested. In this example, we collaborated with artist Nick to create a chatbot based on his character LuckyCot. This character is a cat named Modem Cot that mutated from a network mediator modem and has a network cable tail. It is interested in blockchain-related content. [图片] [图片] The thinking process (steps) of this bot can be summarized as follows: 1. Observe: Understand what is being discussed in the current conversation. 2. Evaluate: Rate the interest and relevance of the message on a scale of 1 to 10, where 1 is not interested or related, and 10 is extremely interested and related. This is a reference value for agent's decision. 3. Reflect: Identify any associated emotions, relevant memories or users, and determine if the bot is interested or related (evaluation score greater than 7). 4. Action: Choose one of the predefined actions (Direct reply, Silent, or Relate reply) based on the observation, evaluation, and reflection. In summary, we have applied the Observe-reasoning-action approach proposed in React to community chatbots. Currently, this application is displayed through WeChat. We hope that in the future, Autonomous Agents can truly become members of human society.