Every sizable corporation is asking the same question:
What are the primary applications for large language models (LLMs) like GPT-4 within our business, and where should we first allocate our investment?
While I can’t provide a custom-tailored solution for every individual scenario, I can shed some light on the top eight emerging use cases that I’ve identified from my interactions with colleagues and various organizations. These use cases are currently in the strategy and planning phase and are evolving at a rapid pace.
I’ve arranged these eight use cases into three distinct categories. In this discussion, we’ll dive into the first category that I’ve labeled ‘classic natural language processing use cases.’ There are five of these. These are the types of tasks you’d have tackled using traditional statistical methods or by adapting BERT-family NLP transformers, before the advent of GPT models. Although the underlying technology for these use cases may be identical or bear similarities, this categorization helps us understand their practical applications within a business process.
1. Text classification: This is about assigning categories to documents, or even to sections within a document. It’s the most straightforward and common NLP use case in the corporate environment.
2. Information extraction: This is where you pull out names, places, or specific sections from documents, like contract clauses. It’s often used for filling out a relational graph, like a 360-degree view of a customer.
3. Semantic search: This is about searching for information based on its meaning, which is a more powerful technique than the traditional keyword search.
4. Information summarization: This is an area where LLMs, such as GPT-4, outshine traditional BERT-style NLP models. The older techniques of abstractive or extractive summarization have become irrelevant because LLMs can summarize information in the way users need it.
5. Information comparison: This use case is about finding similar documents in a large pile of files. It’s commonly seen in organizations dealing with case files, like law enforcement agencies or intelligence bureaus.
These five use cases often form the starting point for most companies because they deal with natural language processing capabilities that they’re already quite familiar with. However, these organizations are facing a couple of challenges.
The first one is that applying LLMs to these classical NLP use cases comes with the same problems as earlier technologies: they usually provide only incremental value to a specific business process. To get a good return on investment, you need a capabilities-based strategy that can offer value across various business processes. For instance, classifying text might bring incremental value to a specific process, like managing claims or underwriting. So, to get a high ROI from a text classification use case, you need to create a text classification capability that can be applied across multiple business units. It’s not easy, it takes time, and it requires a substantial investment.
The second challenge is that many companies already have initiatives underway that are delivering good results, and it doesn’t make sense to disrupt them to replace them with a new technology that solves similar problems. Especially given that deploying LLMs comes with its own set of unique challenges, as I’ve mentioned before.
So, a lot of companies are shifting their focus to new use cases that can deliver greater business value in a shorter time. We’ll take a look at these in Part 2. Have a great day!
PROLEGO GREATEST HITS:
Prolego is an elite consulting team of AI engineers, strategists, and creative professionals guiding the world’s largest companies through the AI transformation. Founded in 2017 by technology veterans Kevin Dewalt and Russ Rands, Prolego has helped dozens of Fortune 1000 companies develop AI strategies, transform their workforce, and build state-of-the-art AI solutions.