Google LLC’s cloud division Google Cloud today announced big updates for Vertex AI Search that delivers artificial intelligence capabilities to empower information retrieval.
Using Vertex AI Search, which was made generally available in August, developers can quickly create conversational chatbot search applications that would have taken weeks in mere days or hours and use large language model AI capabilities to enable information discovery at large scale. Essentially it brings a better search experience to customers or internal employees by taking what would have been a simple keyword search and turning it into a conversational experience.
The new features include deeply customizable search, search tuning and new classification models such as include images and video. Google says they also give enterprise customers confidence that they’re getting accurate and grounded answers, with more options to prevent hallucinations.
Search can be a powerful but dry experience in ordinary circumstances, Lisa O’Malley, senior director of product management and AI at Google Cloud, told SiliconANGLE in an interview. “We think of search in combination with conversation because that’s how a lot of customers want to use it,” O’Malley said. “You’ll remember, if you go to a retailer, or a state agency website or any of these things, the search tends to be not great.”
With generative AI search capabilities, the system is now talking with the user in a humanlike manner, so a business will want that chat to understand and align with its audience. A healthcare provider, for instance, might have a chatbot on its system to provide search for medical records and it would be customized to speak to a physician. If placed on a patient-facing portal, it would be customized to use less medical language and speak plainly but with authority.
A retailer could have it focus on the fashion needs of a potential customer and be limited to an eighth-grade vocabulary. It also should maintain proper tone and brand focus during search conversations.
Vertex AI Search has already launched a service optimized for the healthcare and life sciences industry and focused on allowing physicians and clinicians to pull medical data from multiple sources. However, other organizations such as retail and hospitality businesses, government agencies and nonprofits would certainly have completely different needs. They each require ways to customize how the underlying AI produces its answers.
With search tuning, the underlying AI can be trained on what good search answers to a particular question look like so that when someone asks the same question, the large language model will bring back the most accurate answer in the future.
For example, a company might design its own contracts and internal employees constantly search for them. In order to tune the model on good answers, all that’s needed is to provide the question that might be passed to it and then give it 20 or 30 examples of contracts and show the model where the information sits in the knowledge base. That greatly increases the likelihood of good answer.
Developers can also now add semantic embeddings for search that include text, image and video, which can unlock the capability for multimodal search for content moderation, recommendations and semantic search.
A big concern for enterprise users is hallucinations, the possibility that an LLM might go haywire and output something totally inaccurate. Today Google Cloud added extra options for grounding, the capability to verify and lock results to known datasets. These include grounding to internal enterprise datasets for verification with citations and with selected third-party public datasets, such as Wikipedia.
“Obviously, we have years and years of experience building very powerful what we call information retrieval and deep retrieval tools,” O’Malley said. “But we also have, all of these new generative AI tools and large language models at our disposal. We believe that the combination of those two things is incredibly powerful. What we’ve done with search is almost abstract away the need to work with the large language models directly.”
Vertex AI has already seen interesting enterprise use with the recent announcement by Forbes, which launched its purpose-built news AI-powered news search tool Adelaide in beta test mode. The search tool generates personalized news recommendations and summaries from using Vertex AI’s foundation models and is tuned on Forbes’ own archive of news articles over the past 12 months.
“As we look to the future, we are enabling our audiences to better understand how AI can be a tool for good and enhance their lives,” said Vadim Supitskiy, chief digital and information officer of Forbes.
Your vote of support is important to us and it helps us keep the content FREE.
One click below supports our mission to provide free, deep, and relevant content.
Join the community that includes more than 15,000 #CubeAlumni experts, including Amazon.com CEO Andy Jassy, Dell Technologies founder and CEO Michael Dell, Intel CEO Pat Gelsinger, and many more luminaries and experts.