Compare ungrounded vs. grounded (RAG) LLM responses. Hallucinated content is highlighted in red; grounded, cited content in green.
Hallucinated (unverified)
Grounded (cited source)
Ungrounded LLMTrust: ?
Select a question above to see responses
Grounded LLM (RAG)Trust: ?
Select a question above to see responses
How This Works: The "Ungrounded" panel simulates a general LLM responding from memory (training data). It may be outdated, wrong, or fabricated — but sounds confident. The "Grounded" panel simulates a RAG-augmented LLM that first retrieves relevant documents from a curated, verified knowledge base before generating a response, and cites its sources.