Large language models (LLMs) that may yield astounding outcomes, such as writing summaries, drafting code, or recommending new ideas, are examples of artificial intelligence advancements. However, one major issue with these models is that they can occasionally give false or misleading information, a phenomenon known as “hallucination.”
To address this issue, Google released DataGemma, the first open models that use real-world statistical data to increase the accuracy of LLMs. These models are based on Google’s Data Commons, a vast knowledge base including trustworthy public data. DataGemma hopes that by including this trustworthy data into LLMs, it will eliminate hallucinations and improve the factual dependability of AI-generated outcomes.
What is Data Commons?
Data Commons is a publicly accessible knowledge graph containing over 240 billion data points from reputable sources including the United Nations, the World Health Organization (WHO), and the Centers for Disease Control and Prevention (CDC). This data covers a wide range of themes, including health, economy, demography, and the environment. It is a valuable resource for scholars, politicians, and organizations searching for accurate information.
Consider Data Commons to be a massive, ever-growing database that anybody may browse using a natural language interface driven by artificial intelligence. Users may, for example, query “Which African countries have improved their electricity access the most?” or “How does income affect diabetes rates in U.S. counties?” and get specific responses based on real-world data.
How DataGemma Addresses AI Hallucinations
As the use of generative AI grows, it becomes increasingly vital to anchor its outputs in true data. DataGemma is a series of lightweight models created with the same technology as Google’s Gemini models. These models relate to Data Commons, enhancing LLMs’ capacity to reason and communicate facts correctly.
DataGemma uses two main ways to improve the accuracy of AI replies. The first option is Retrieval-Interleaved Generation (RIG), which enhances the language model by proactively querying trusted sources to cross-check facts with Data Commons. When DataGemma is requested to generate a response, it recognizes sections of the question that require statistical data and pulls precise information from Data Commons, making it especially useful for fact-based searches. The second option is Retrieval-Augmented Generation (RAG), which allows the model to collect more context from external sources before generating a response. By utilizing Data Commons, the model absorbs important information and provides footnotes or references, ensuring that its outputs are both accurate and thorough. These strategies operate together to eliminate hallucinations and improve the accuracy of AI results, particularly for complicated or data-intensive inquiries.
Promising Early Results
Early research with the RIG and RAG approaches yielded promising results, notably in minimizing hallucinations when models were expected to handle numerical data. These findings indicate that consumers will benefit from more accurate and reliable information, whether for research, decision-making, or personal inquiries. A study paper describing these findings is available for further reading.
Google intends to continue improving these technologies and growing the DataGemma project, making it more accessible to developers and academics over time. DataGemma’s anchoring of LLMs in real data is a huge step forward in making AI more credible and usable for everyone.
For more details, check out Google’s research post and the accompanying quickstart notebooks for using DataGemma with RIG and RAG techniques.
Source: Google’s DataGemma Announcement, Sept. 12, 2024. You can check out the full article here.
Hi, I'm Voss Xolani, and I'm passionate about all things AI. With many years of experience in the tech industry, I specialize in explaining the functionality and benefits of AI-powered software for both businesses and individual users. My content explores the latest AI tools, offering practical insights on how they can streamline workflows, boost productivity, and drive innovation. I also review new software solutions to help readers understand their features and applications. Beyond that, I stay up-to-date with AI trends and experiment with emerging technologies to provide the most relevant information.