Large Language Models: 3 Examples of Problems They Can Solve

Large language models (LLMs) are all the rage, fueled by the release of OpenAI's ChatGPT in late 2022, initially powered by the LLM GPT-3. Aside from the news hype, what can LLMs actually, getting-down-to-brass-tacks, nitty-gritty do for your business? Here, we’ll look at three examples of problems they can solve. But first, a quick definition of LLMs. 

According to Gartner®, “A large language model (LLM) is a specialized type of artificial intelligence (AI) that has been trained on vast amounts of text to understand existing content and generate original content.”¹

Based on how you expand an LLM by pairing it with other technology, you can produce a wide variety of content types, such as audio, images, designs, and likely soon an inexhaustible list of other kinds of content. But currently, businesses are discovering how helpful LLMs alone can be for text-based artifact creation. 

According to a Gartner® report, “Artifact creation — The acceleration of large language models (LLMs), and foundation models in general, has seen the contribution of AI techniques reaching a new scale across a wide variety of content (such as text, audio/visual, programming and data assets, designs and learning methods).”² 

Here are three examples of artifacts that can help your business solve important problems.

Many organizations have a constant influx of customer support messages that they need to respond to. Particularly when worked into a process, an LLM is a perfect co-pilot to assist customer support representatives in generating replies to these messages. In Appian, an AI platform for process automation, an application developer simply has to place a connector in the application to the Azure OpenAI platform plug-in. This plug-in uses the customer’s email copy as a prompt and generates an initial response for the employee to edit as needed. The application developer can set up requirements for tone, length, language, type, and more to quickly generate strong email communications both internally and externally. 

Like the other artifacts produced by LLMs, humans should remain in control. For example, experts should check for AI “hallucinations” and make sure the tone is in line with the company’s standards. 

Any organization with a large dataset could benefit from an LLM for knowledge management. Because LLMs can understand natural language queries, employees could theoretically type in something like, “What is the minimum order required for free shipping?” and receive an answer. But the key here is that the LLM must be trained on your organization’s dataset rather than a public model like ChatGPT, which pulls from public datasets. This concept of private AI is constantly being developed and enhanced but is promising for organizations who want to reap the benefits of AI without compromising their data.

Developers can use LLMs to write code. Development is traditionally a lengthy, expensive process, and these productivity-boosting LLMs offer a way for developers to move more quickly. And it’s not just common languages like HTML or JavaScript that LLMs can write. Even platforms like Appian use generative AI in some instances to build application designs using SAIL, our low-code design system and user interface framework. 

All these examples of LLMs solving problems show that they can do a lot for productivity, and it’s only the early days of this revolution. But to protect you down the road, here are two suggestions to help you save your organization from costly problems later on.

AI can carry risk. Heed our advice to stay safe:

• Keep your data private. Beware of AI models that won’t keep your data private or will use your data to train its own dataset. The benefits may seem worth it now, but there are private AI options that will protect your data, and your customers’ data, too, so that you aren’t risking anything for the future.
• Adopt AI for efficiency’s sake, not its own. Rather than pursuing AI for its own sake (unless, of course, that’s what your business does), adopt AI in the context of efficiency. This will help you get closer to truly operationalizing AI rather than going down dead-end streets. What might this look like? Choose to adopt software that’s already incorporating AI into everything it does.