Google is enhancing its efforts to improve the factual grounding of enterprise AI chatbots by integrating real-world financial data from business and financial services.
The plans aim to reduce inaccuracies in AI-generated information and improve overall reliability, known as ‘hallucinations,’ by grounding responses with verified data.
The news is part of Google’s plans to include third-party data sources, with initial partners comprising Moody’s, Thomson Reuters and ZoomInfo, on top of web search and internal company data.
Google wants enterprise AI to be more factually correct
Fairly late to the party, enterprises had initially waited for more secure and private systems to protect sensitive company data, however with more enterprises deploying AI solutions, the focus has shifted onto accuracy.
Speaking about the changes, Google Cloud CEO Thomas Kurian told Axios: You can actually trust the model to do a task on your behalf because you have a basis for trusting it.”
To further enhance reliability, Google is also introducing a confidence score which provides a numeric indicator of the AI model’s certainty in its answer. Enterprise users will also be able to direct the AI chatbot to prioritize information from specific documents or data included in a prompt, rather than its broder training data.
Kurian added: “We’ve taught the model how to guarantee that when it responds, it takes what’s in the input prompt as the primary information it needs to pay attention to.”
Moreover, Google is expanding Vector Search to support hybrid searches, combining vector-based images searches with text-based keyword searches, for improved accuracy. The upgrade is currently available in public preview.
Google Cloud’s announcement, authored by Burak Gokturk, VP & GM for Cloud AI & Industry Solutions, concludes: “As these technologies become even more capable, we are committed to helping businesses realize the full potential of grounded generative AI in the real world.”