Truth-Seeker: Using LLM Agents to Build and Verify Knowledge Bases

Explore an innovative approach to building and verifying knowledge bases using LLM agents in this 32-minute conference talk from DevConf.US 2024. Discover how Truth-seeker leverages open-source LLMs to create agents that process source documents, break them down into evaluable statements, and construct a robust knowledge base. Learn about the scoring system that assesses statements based on their sourcing, consistency with existing knowledge, and classification as facts, opinions, or biases. Gain insights into how this tool aims to enhance the quality of training data by enabling filtering of undesirable information and enriching valuable content with additional sources. Presented by Jeremy Peterson, this talk offers a deep dive into cutting-edge techniques for improving the foundation of quality LLMs through advanced data curation and verification methods.

Truth-seeker: Using LLM agents to build and verify knowledge bases - DevConf.US 2024