Google announced that Vertex AI will receive new models and updates to prevent apps and services from publishing inaccurate information.
The Vertex AI cloud development platform allows companies to build services using machine learning and large language models.
Google is best known for its consumer-focused chatbot Gemini, but it also offers enterprise solutions with its business-friendly artificial intelligence platform Vertex AI.
Google has made changes to its internal model, including moving Gemini 1.5 Flash from public preview to public release.
Gemini 1.5 Flash is the fastest Gemini model in Google API and a cost-effective alternative to the Gemini 1.5 Pro model.
Google has launched Imagen 3, a state-of-the-art image generation model aimed at Vertex AI clients. Highlights of this template include 40% faster creation, realistic collection creation, multi-language support, and built-in security features.
After launching Grounding with Google Search in May, Google announced that it would allow customers to use specialized third-party datasets to improve its service's AI results.
Using Google search, models can get information directly from the Internet.
Google said the service uses data from providers such as Moody's, MSCI, Thomson Reuters and ZoomInfo and may be linked to external datasets available in the third quarter of this year.
This is one of several new features that Google is developing to encourage companies to adopt business-friendly generative AI experiences by reducing the frequency with which models post misleading or inaccurate information.
Google also announced its “High Fidelity Grounding” program, which allows companies to extract insights from results generated by company datasets rather than aggregating knowledge via a Gemini model to ensure a high level of realism.
The high-resolution grounding mode runs on a special version of Gemini 1.5 Flash and is now available for preview via the Vertex AI Experiment Tool.
Google has also expanded its vector search feature to support hybrid searches, allowing users to find images by referring to similar graphics.
The update is available in public preview and allows vector searches to be combined with text keyword searches to improve accuracy.
Google Search Essentials will also soon offer a dynamic retrieval feature that automatically determines whether information should come from datasets created in Gemini or Google Search to find claims that may require constantly updated resources.
The ability to control the source of information for AI models could help improve the reputation that Google's AI-powered search capabilities have earned so far.
After Google struck a deal in February to access Reddit data to train artificial intelligence, Google's AI Showcase feature was mocked for making bizarre recommendations based on old Reddit posts.