Navigating the World of Generative AI

In the past several months, the AI experts at Verint have been fielding a slew of questions about everything racing through the news cycle concerning ChatGPT and other large language models. Some come to us curious about the rapid advancement of this technology. Others are a bit worried about how it could affect society at large. Then there’s a lot of folks wondering how these models can be put to use for their businesses.

At Verint’s Engage conference earlier this month, those experts—along with leading analysts—took to the stage to cut through the noise and give some history and insights about generative AI, and also gave a rundown of how this technology is already used within Verint Open CCaaS Platform.

Brian Yang, senior director of product strategy at Verint, started out by defining Generative Pretrained Transformers (GPT) models. The goal of these models, he said, is to synthesize and relay information with data used to determine how to respond, all of which comes from an extensive training process. Also, GPT models are defined as neural networks that learn based on relationships in data.

“GPT models are part of a larger family of AI models called Large Language Models. And LLMs aren’t actually new—they’ve been used since the 1950s,” said Yang. “But GPT models are very new—they’ve only been around since 2018.”

But even though GPT models have only existed for about five years, adoption has been remarkable, said Yang. As an example, he pointed to the fact that about 100 million people around the world had used ChatGPT within the first two months of its launch in January of this year.

There remain, however, a range of challenges when it comes to using these generative models for your business and to serve your customer, as Verint Director of Research, Ian Beaver, pointed out.

“These models provide back information in well-structured formats that relay what they have seen in response. But the problem with that is that GPTs will quote things as facts with convincing reasoning and citations that may not exist. Essentially, they can make things up,” said Beaver.

Obviously, when used in a business setting, a company can’t afford to provide “made up” information to customers. They need AI that intimately knows the company’s terminology, policies, brand, pricing, geography and more that can’t be immediately sourced from a pre-existing model like ChatGPT.

Beaver also covered how Verint is currently using generative models—alongside Verint’s own Da Vinci AI and Analytics—to help better the customer service experience and improve outcomes for contact center agents.

Currently, Verint is using GPT models within more than a dozen active projects. Right now, for example, GPT models are used to summarize call transcriptions so an agent can receive a succinct overview of a customer’s journey during an engagement. Also, Beaver said that Verint uses Microsoft Azure Open AI and ChatGPT, HuggingFace, AWS, and Google models to bolster its proprietary Da Vinci models.

To learn more about Verint AI and see how we’re providing better customer and employee outcomes through automation, take a look at our Complete Guide to Conversational AI for Your Business.