Breaking Down Generative AI at Data Summit 2023

The global hype surrounding ChatGPT, large language models (LLMs), and generative AI has launched a plethora of conversations about its efficacy in the business sphere; yet understanding how to best leverage these tools remains rather elusive.

At Data Summit 2023, David Seuss, CEO of Northern Light, led the session, “Investigating Neural Networks and AI,” focusing on what ChatGPT, LLMs, and generative AI are, as well as how they can provide a competitive edge for a wide variety of organizations seeking to drive positive business outcomes.

The annual Data Summit conference returned to Boston, May 10-11, 2023, with pre-conference workshops on May 9.

Seuss quoted Nicola Morini Bianzino, CTO at Ernst Young: “The ability to quickly retrieve, contextualize, and easily interpret knowledge may be the most powerful business application of large language models. A natural language interface combined with a powerful AI algorithm will help humans in coming up, more quickly, with a larger number of ideas and solutions that they subsequently can experiment with to eventually reveal more and better creative output.”

This robust combination of AI and LLMs finds itself within a business context as question-answering, according to Seuss.

Connecting business users to accurate answers through a variety of queries—including “What is Microsoft’s strategy in cloud computing?” “What factors affected Pfizer’s earnings?” and “What acquisitions has IBM made?”—is exactly how enterprises will gain valuable research to drive business success.

Despite this seemingly abundant treasure trove of value available at an organization’s fingertips, confusion about generative AI solidifies this gap between business and question-answering tools.

Seuss began to shed that layer of confusion by explaining that ChatGPT and LLMs—like GPT-3—are not the same thing. Rather, generative pre-trained models (or “GPT” models) are LLMs trained to generate human-like text by predicting what text is supposed to follow other text in a given context.

On the other hand, ChatGPT is a specific implementation of the GPT-3 architecture that has been trained to generate interactive human-like conversational responses that are appropriate and engaging for a given conversation.

“Despite the technical difference, public discourse usually equates the two and it is the term ‘ChatGPT’ that everyone recognizes even though the discussion might be about LLMs such as GPT-3 or GPT-4,” said Seuss.

Seuss dove further into this clarification by expanding on the topic of training your own LLM as opposed to using LLM APIs from OpenAI and Microsoft; he explained how, though it is tempting to want to tune an LLM to your application, the cost of training the LLM is massive. The high-end GPUs needed to train the LLM alone is a large aspect of the wildly unaffordable cost.

According to Rowan Curran, a Forrester analyst, “Analysts and technologists estimate that the critical process of training a large language model such as OpenAI’s GPT-3 could cost more than $4 million. More advanced language models could cost over ‘the high-single-digit millions’ to train.”

Seuss advised to test the APIs available before attempting the journey of modeling your own LLMs; “like virtually every other machine learning application, you just need to try it and see if it works rather than overthinking it,” he said.

Using an LLM API does not necessarily mean it won’t be adjustable to an enterprise’s needs, however. By submitting trusted content to an LLM API, as opposed to relying on the model’s internet sources, the quality of search results is improved based on the quality of searched material; this also helps avoid AI hallucination. 

Another critical aspect of leveraging an LLM is citations and links, where providing citations allows users to vet the source of their generated answer. This helps not only copyright compliance, but for driving the user to search the documents in-depth—generated answers and executive summary is a powerful beginning to the research process, not the end, Seuss explained.

“The future belongs to those organizations that can apply the power of the new search possibilities created by generative AI to the crucial questions being answered—one way or another—every day throughout their strategic functions such as product management, marketing, product development, and research,” concluded Seuss.

Many Data Summit 2023 presentations are available for review at