ConceptGrounding
Data Grounding
In the context of large models, it refers to constraining model outputs to strictly rely on a given trusted data source rather than freely generating from training knowledge, in order to reduce hallucination risk and improve verifiability.
Timeline (last 90 days)
Oct 14
Data grounding means making the large model's answers strictly anchored to a given trusted data source, avoiding free generation that relies on training memory, thereby reducing hallucinations.
Unverified50%