Fact Checking Rocks: how to build a fact-checking system
summary
A baseline fact‑checking web app checks user statements about rock music against a Wikipedia corpus using Haystack + FAISS for vector storage, SentenceTransformers to embed passages and a Natural Language Inference model to score entailment/contradiction; scores are aggregated as a weighted average and the pipeline breaks early on strong contradiction. Examples include “Elvis Presley is alive” (wiki contradiction) and a case with entailment 0.52 vs neutral 0.48; finds Wikipedia useful but incomplete. Dense retrieval, Natural Language Inference.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 1:50Agenda overview
- 2:13Deepset framework basics
- 5:22Project idea explained
- 8:46Retriever and entailment checker
- 10:48Implementation steps
- 17:27Testing the system
- 18:29Live demo
- 21:32Limitations and challenges
- 23:42Improvement suggestions
- 25:13Large language models role
- 26:46Agents and tools
- 35:57Conclusion and Q&A
related talks
32mTruth-seeker: Using LLM agents to build and verify knowledge bases
43mHarnessing AI to strengthen trustworthy information
19mDrowzee: Metamorphic Testing for Fact-conflicting Hallucination Detection in Large(…)
22mValidating AI Outputs at Live Events
31mEnhancing RAG-based apps by constructing and leveraging knowledge graphs with open-source LLMs
41mTruth Checker: Generative Large Language Models and Hallucinations
29mNavigating A World Where AI-Generated Misinformation Can Be Created At Scale
34mBuilding Trust with LLMs