artificial intelligence
Google Cloud has expanded Vertex AI’s grounding capabilities, significantly enhancing the platform’s ability to generate more accurate and reliable AI responses. These advancements are intended to mitigate AI hallucinations and improve the overall quality of generative AI applications and agents.
One key addition is the introduction of dynamic search for grounding with Google Search, now generally available. This innovative feature enables Gemini, Google’s advanced large-scale language model, to intelligently decide whether to ground a user query with Google Search or rely on its own knowledge. This approach helps balance response quality and cost-effectiveness, as grounding with Google Search incurs additional processing costs. Gemini makes this decision by understanding whether the requested information is static, slowly changing, or rapidly evolving.
For example, if you ask about the latest movies, Gemini will use Google search to get the latest information. Conversely, for general questions like “What is the capital of France?”, it will provide answers from its existing knowledge base without any external evidence. This dynamic approach not only improves the accuracy of responses, but also optimizes resource usage.
Google Cloud is also introducing an experimental “high fidelity” grounding mode that is targeted at industries such as healthcare and financial services where accuracy and reliability are paramount.
Additionally, Google will soon enable grounded models with third-party datasets, expected to be released in Q3. Google is working with specialized data providers, including Moody’s, MSCI, Thomson Reuters, and Zoominfo, to provide access to their datasets via Vertex AI. This capability will enable businesses to integrate highly specific and reliable information into their AI models, further increasing the accuracy and relevance of the responses generated.
For enterprises looking to incorporate AI models on private data, Google Cloud offers a suite of APIs for Vertex AI Search and Retrieval Augmented Generation (RAG). These tools help enterprises create custom RAG workflows, build semantic search engines, and enhance existing search capabilities. The APIs, now generally available, provide capabilities for document parsing, embedding generation, semantic ranking, grounded answer generation, and a fact-checking service called Check Grounding.
These enhancements are part of Google Cloud’s broader strategy to make generative AI more trustworthy and suitable for enterprise use. By connecting AI models to diverse, trusted sources of information like web data, enterprise documents, operational databases, and enterprise applications, Google aims to root AI in what it calls “enterprise truth.”
The focus on grounding heightens industry concerns about AI hallucinations: as AI models become more complex, the risk of them producing erroneous or unreliable outputs increases. Grounding techniques such as RAG mitigate this risk by feeding models facts from external knowledge sources, improving the accuracy and reliability of their responses.
By enabling businesses to leverage both public and private data sources, Google is enabling the development of more robust and reliable AI applications across industries.
