Don’t miss OpenAI, Chevron, Nvidia, Kaiser Permanente, and Capital One leaders only at VentureBeat Transform 2024. Gain essential insights about GenAI and expand your network at this exclusive three day event. Learn More
As it seeks to win over enterprise customers, Google is trying to ensure that its AI platform minimizes hallucinations. It’s a big deal for organizations, especially those whose executives are already wary about the technology. Google is doubling down on model grounding to reassure them, turning to reputable third-party services Moody’s, MSCI, Thomson Reuters and Zoominfo. These four will be available within Vertex AI starting next quarter. They will offer developers qualified data to backstop their model outputs against to ensure responses are factually accurate.
What it means for enterprise developers is that they can now leverage the fine-tuned quality data and expertise of Subject Matter Experts to ensure it meets their standards. As Nick Reed, Moody’s chief product officer, explains during a press briefing this week, “What’s on offer here is our data being made available to other people so that they can use it to ground their own responses that they might use either internally or externally for their own interactions. We think the service offering from Google is a way of being able to leverage their models and our data to be able to put both of those things, as well as Google Search, into an assistant or an application that the customers of the grounding service would use.”
Google already offers Google Search as part of its grounding service, allowing businesses to augment Gemini outputs with “fresh and high-quality information” from the web. But there are now more trustworthy options for companies who want source-of-truth information and do not risk having inaccurate information from a blog that may have gamed SEO to be highly authoritative.
But that’s not all, as Google is also announcing high-fidelity grounding. Available through an experimental preview, it’s designed to help AI systems work better with a given set of specific information. It is useful when dealing with tasks like summarizing multiple documents simultaneously or extracting important data from financial reports. Grounded with high-fidelity is powered by Google’s Gemini 1.5 Flash.
Countdown to VB Transform 2024
Join enterprise leaders in San Francisco from July 9 to 11 for our flagship AI event. Connect with peers, explore the opportunities and challenges of Generative AI, and learn how to integrate AI applications into your industry. Register Now
When asked if the new grounding partnerships and the new grounding with high-fidelity signaled Google’s generative AI as being the “most reliable factual model out there compared to competitors,” Google Cloud Chief Executive Thomas Kurian cites three key differentiators: First, Google is using its reputation as having the “most trusted source of web data and real-world understanding” with Google Search to help ground AI against web data. Second, the company provides options on how data should be grounded in specific sub-verticals — “you control what you want to ground against.” And third, high-fidelity grounding directs AI to pay more attention to what is being prompted, not all the data that the model may have been trained on, improving the quality of the response.
“All three of these elements combined give you the highest degree of control on the quality of the answers that a model can generate,” he remarks. “And it’s our attempt at reducing hallucination and improving trust in models.
This announcement comes on the same day as other Google Cloud news, such as the public release of Gemini 1.5 Flash and 1.5 Pro, Gemma 2, and other Vertex AI updates.
Source: venturebeat.com