Google’s DataGemma combines Gemma LLMs and Data Commons to prevent inaccuracies in generative AI outputs by grounding them in factual, publicly available data.
RAG enables grounding LLMs in data, with DataGemma being the first implementation of this technique at the scale of cloud-based generative AI.
Google's approach allows for fact-checking of prompts using Data Commons data, providing reliable statistics in responses while enabling extensive knowledge retrieval.
Data Commons gathers information from institutions like the United Nations, enhancing the factual accuracy of generative AI with real-world data integration.
[
Collection
]
[
|
...
]