It's a clever idea: The authors instruct multiple LLM-simulated agents to assume different roles in a Socratic dialogue to figure out how to solve problems proposed by the user. Think of it as a team of LLM-simulated agents, each with a different set of goals and responsibilities (e.g., one may be creative, another may be analytical, another may be a proofreader, etc.).
It's a clever idea: The authors instruct multiple LLM-simulated agents to assume different roles in a Socratic dialogue to figure out how to solve problems proposed by the user. Think of it as a team of LLM-simulated agents, each with a different set of goals and responsibilities (e.g., one may be creative, another may be analytical, another may be a proofreader, etc.).
Demo (video): https://vimeo.com/821089266
Code: https://github.com/RunzheYang/SocraticAI