What is a Large Language Model?
A Large Language Model, or LLM, is a neural network trained on massive amounts of text. As of August 2023, AIs have been trained on trillions of sequences of vocabulary across multiple languages. The amount of data these systems are trained on will keep growing.
“If you’ve used ChatGPT, you are using an LLM. If you use Bing Search, you are using an LLM,” said Eck. "While they are called language models, they’ve become what we call multimodal. So the same model can operate on text, video, images, music.”
For instance, AI is enabling more effective service and fraud detection in call center operations through improved real-time sentiment analysis of the caller’s voice, tone and language. In compliance and regulatory functions, automating time-intensive reporting such as audit reports, reconciliation and initial KYC analysis to understand if a potential customer will be verifiable will shape operations in years to come.
AI tools won’t necessarily take over these roles, but they are likely to make them a lot more efficient. “I like the word that Microsoft uses: they think of these tools as co-pilots. They’re not sitting in the pilot seat. You’re still in the pilot seat, but they’re a co-pilot. And quite frankly, they’re really good,” said Eck.
What does that look like in real life? With code writing, you need to know your objectives and the right questions to ask. It also helps to have a baseline understanding of the technology. Having a perfect string of Python code is useless if you don’t know how to deploy it. Similarly, a compliance officer can generate reports faster with AI, but they still need to understand the regulations in the jurisdiction in question.
Other use cases make it easier to use existing tools, increasing productivity and speed to innovation.
Client360 is our portal to services for Fiserv solutions, which has been infused with AI. This includes a client-facing chatbot, issue resolution and internal tools for Fiserv associates to support clients and their customers.
Another example is developer productivity, according to Eck. “I have a product I created called Developer Studio that’s kind of like our catalog of APIs and documentation. I took all the data from Developer Studio and trained a model … a developer can just type in ‘what API should I call to issue a card?’ and it might come back and say, ‘There are three options'.” From there, the developer could ask for next steps through conversational chat, ask how to call that API in Java or request more information. Put simply, it makes the developer’s job faster and easier.