Google's DataGemma is the first large-scale Gen AI with RAG

Google's DataGemma is the first large-scale Gen AI with RAG - why it matters

Google has introduced DataGemma, a novel integration of its Gemma open-source large language models with the Data Commons public database. By leveraging retrieval-augmented generation, this system allows AI to fetch verified, publicly available statistics before generating responses. This approach aims to significantly reduce hallucinations and enhance factual accuracy by grounding the model’s outputs in real-world data from sources like the United Nations, rather than relying solely on internal training data. The implementation utilizes two distinct methodologies to improve response quality. The first method employs retrieval-interleaved generation for fact-checking specific queries, while the second uses full retrieval-augmented generation to provide comprehensive, source-cited answers. By combining these techniques with extensive context windows, the model can analyze large volumes of retrieved information simultaneously, ensuring that reasoning is supported by concrete evidence and improving utility for research and decision-making tasks. This development is highly relevant to the open_data community as it demonstrates a scalable framework for utilizing public datasets to enhance artificial intelligence transparency and reliability. It highlights a shift toward hybrid systems that combine open-source models with open data infrastructure, proving that publicly available information can effectively ground generative AI. This validates the potential of open data ecosystems not just as storage, but as active components in building trustworthy, fact-based AI applications that serve broader societal needs.

Source: zdnet.com
Published on 2024-09-17

From Fraud to Fairness: How Transparent Data Strengthens Democracy and Trust »

Open Data

Related news