Simple and Constrained LLM Agents - Building Controlled AI Systems for Production

Explore a 46-minute talk by Dario Cannone from Faculty at the Data Science Festival, focusing on building simple and constrained LLM agents. Gain insights into the speaker's experience developing an LLM-based agent for a UK fintech company and its deployment in a complex production ecosystem. Learn about the risks associated with unpredictable interaction loops in LLM agents and discover a mitigation strategy involving simpler agents with pre-determined subtask chains. Examine the advantages and disadvantages of this approach, which prioritizes control over generalizability. Suitable for technical practitioners, this session was part of the Data Science Festival MayDay event 2024. Delve into cutting-edge ideas and real-world problem-solving in the data science field through this talk from the Data Science Festival community.

Simple and constrained LLM agents