Google Cloud’s Vertex AI platform, which enables enterprise customers to build their own services using machine learning and Google’s large language models, has new capabilities aimed at preventing the spread of misinformation.
Fonte da imagem: cloud.google.com
Vertex AI Grounding with Google Search, a feature that gives AI models access to real-time data from the web, was rolled out to all customers in May. Now, customers of the cloud platform will be able to further enhance the results of their AI services by using specialized third-party data sets. The service will use data from providers such as Moody’s, MSCI, Thomson Reuters, and ZoomInfo, Google said, and will be available to companies “in the third quarter of this year.” It’s just one of a number of features Google is developing to encourage organizations to start deploying their own enterprise AI models on its infrastructure, all aimed at reducing the frequency of bad data in system responses.
Another feature is a “high-fidelity mode,” which allows organizations to plug their own datasets into models instead of Google’s extensive Gemini knowledge base. This mode is based on a special version of the Gemini 1.5 Flash model and can be enabled through the Vertex AI experimental features section. Enterprise customers will also be able to perform hybrid searches, using sample images and text explanations as queries. And Grounding with Google Search will soon feature dynamic search, which will automatically select sources of information: installed Gemini datasets or Google Search if the resources are expected to be updated frequently.
Controlling the datasets that power Google’s AI models could help the company improve its reputation, which has suffered since the introduction of AI into its search engine. While consumers have taken advice to eat rocks and put glue on pizza in good humor, such mistakes could prove costly for corporate clients.