How to Use LLMs for Fact Checking - Advanced Techniques and Comparisons

Explore advanced techniques for using Large Language Models (LLMs) in fact-checking applications in this 41-minute video tutorial from Trelis Research. Learn about citation verification, enforcing structured responses with regex and JSON, and context focusing. Compare long context and retrieval approaches to fact-checking using a modified Irish Constitution as a test document. Discover how to set up structured responses with Gemini and OpenAI GPT-4, and explore fact-checking methods using BM25 and cosine similarity retrieval. Gain insights into the effectiveness of different approaches and access valuable resources for implementing these techniques in your own projects.

Intro - fact checking with LLMs
Video Overview
#1 Citation Verification Technique
#2 Enforcing Structured Responses regex / json with OpenAI
#3 Context Focusing
Comparing Long Context vs Retrieval approaches to fact-checking
Test Document Preparation adding errors to the Irish Constitution
Long Context Fact Checking whole document as context
Long Context Fact Checking with Structured Responses
Setting up structured responses with Gemini and Openai GPT-4o
Hacking structured responses with Claude
Fact checking with bm25 and cosine similarity retrieval
Results: Long Context vs Structured vs Retrieval Approaches
Video Resources