Day 11: Plan Reflection
Plan Reflection
Once a plan is developed, it is essential for LLM (Large Language Model) agents, on which AI Assistants are based, to evaluate its efficacy. Assistants relying on LLMs employ internal feedback systems that leverage established models to enhance their strategies. They have the capability to engage with humans to adapt their plans according to feedback and preferences. Moreover, agents collect insights from their surroundings, whether physical or digital, utilising results and observations to continuously improve their plans.
There are two effective methods for incorporating feedback in planning - ReAct and Reflexion.
ReAct helps an LLM solve complex tasks by cycling through a sequence of thought, action, and observation, repeating these steps as needed. It takes in feedback from the environment—including observations and input from humans or other models—allowing the LLM to adjust its approach based on real-time feedback.
Reflexion focuses on self-assessment, enabling the agent to evaluate its previous actions against desired outcomes, thereby fostering continuous improvement.
Comments