New features are being added to Vertex AI, the Google Cloud development platform that enables businesses to create services utilizing Google’s massive language models and machine learning to help stop apps and services from spreading false information. Following the May general release of Vertex AI’s Grounding with Google Search feature, which allows models to pull real-time data from the web, Google has now revealed that users will also be able to enhance the AI outcomes of their services by utilizing specialized third-party datasets.
According to Google, the service would make use of data from suppliers such as Thomson Reuters, Moody’s, MSCI, and ZoomInfo. It is expected that third-party dataset grounding will be ready in “Q3 of this year.” This is just one of many new features that Google is working on to make its “enterprise-ready” generative AI experiences more appealing to businesses by lowering the frequency with which models produce false or incorrect outcomes.
An further option is “high-fidelity mode,” which lets businesses use their own corporate datasets as a source of information for created outputs rather than using Gemini’s larger knowledge library. Powered by a customized version of Gemini 1.5 Flash, high-fidelity mode can be accessed in preview right now through Vertex AI’s Experiments tool.
Hybrid search will also be supported by Vector Search, which lets users find photos by referencing related graphics. To increase accuracy, the update-which is now in public preview-allows text-based keyword searches to be combined with vector-based searches. Soon, Grounding with Google Search will also offer a “dynamic retrieval” option that will decide automatically whether to use Google Search or Gemini’s pre-existing datasets for prompts that might ask for regularly updated resources.
The unimpressive reputation Google’s AI-powered search features have so far earned may be improved with greater control over the information sources used by Google’s AI models. In February, after agreeing to use Reddit’s data for AI training, Google’s AI Overviews tool came under fire for suggesting odd things like putting Elmer’s glue on pizza, based on previous Reddit postings.