Glossary term
Glossary term
Safety and Alignment
A property of a model whose output is based on (is "grounded on") specific source material. For example, suppose you provide an entire physics textbook as input ("context") to a large language model. Then, you prompt that large language model with a physics question. If the model's response reflects information in that textbook, then that model is grounded on that textbook.
Note that a grounded model is not always a factual model. For example, the input physics textbook could contain mistakes.
Created for this library
A legal-tech vendor reports groundedness of model outputs as a per-release metric to confirm every generated claim is backed by a cited passage.
A medical writing assistant tracks groundedness of generated patient summaries against the underlying chart before each release.
A financial research assistant reports groundedness on generated company summaries because analysts will not trust ungrounded claims.
Definition source: Google for Developers Machine Learning Glossary | Creative Commons Attribution 4.0 License