System Prompt Learning for Agents - AI Engineer Paris

Humans aren’t frozen in the way they think; why, then, are system prompts static today? In order for agents to learn and adapt, they must be able to update the system prompts themselves.

In this session, we will release data from new experiments on how agents can pick up explanations of fixes and annotations and build out instruction updates for system prompts – with demos showing these techniques used in real agent environments, from code agents to gaming agents.

Similar to how humans learn what to do from their environment, this approach uses feedback to improve and drive an agent.

With this approach: prompts evolve, natural language feedback = error signal, a MetaPrompt rewrites/reinserts targeted instruction, and agents run a prompt learning loop post deployment -- allowing them to continuously improve upon themselves online.

Learn more about Arize's prompt learning research: https://arize.com/prompt-learning-pla...
Follow Aparna on LinkedIn:   / aparnadhinakaran  

Speaker: Aparna Dhinakaran, Chief Product Officer and Co-Founder, Arize AI