Google Launches DataGemma to Improve AI Accuracy with Real-World Data

The new open-source models use Google's Data Commons to reduce hallucinations in large language models.

DataGemma introduces two new instruction-tuned models aimed at mitigating AI hallucinations in statistical queries.
The models leverage Google's Data Commons, a vast repository with over 240 billion data points from trusted sources.
Two methods, Retrieval-Interleaved Generation (RIG) and Retrieval-Augmented Generation (RAG), are employed to enhance factual accuracy.
Early tests show DataGemma models improve factual accuracy from 5-17% to 58% using RIG and 24-29% using RAG.
Google plans to further refine these models and integrate them into broader AI applications, starting with limited-access phases.