AI & DataIntermediate
Grounding
PronunciationGROUND-ing
Definition
Grounding is the process of connecting a Large Language Model (LLM) to external, verifiable data sources to ensure its responses are accurate and factual. It helps prevent the model from generating "hallucinations" by forcing it to rely on trusted information instead of just its internal training data.
Where you hear it
In meetings about AI architecture, RAG implementation, or when discussing how to improve the reliability of a chatbot.
Examples
We need to implement grounding so the AI answers based on our company's internal documentation.
Grounding the model with real-time data significantly reduced the number of incorrect responses.
Common mistake
Thinking that grounding is the same as fine-tuning; while fine-tuning changes the model's internal weights, grounding provides the model with external context during the inference phase.