Grounding
Domain: LLM Concepts
🎯 Motivation
Grounding means anchoring the model’s response to verifiable facts (documents, databases, APIs) rather than letting it rely on potentially inaccurate memory.
📋 Concrete Example
Without grounding: ‘The Eiffel Tower was built in 1888.’ (Wrong - it was 1889.) With grounding: RAG retrieves a Wikipedia article, model reads it, says ‘1889’ - correct.
🔗 Analogy
Citing sources in an essay - instead of making claims from memory (which might be wrong), you look up the fact in a book and cite it. Grounding makes claims verifiable and trustworthy.