Generative AI vs. Large Language Models (LLMs): What's the Difference?

What are the differences between generative AI vs. large language models? How are these two buzzworthy technologies related? In this article, we’ll explore their connection. 

To help explain the concept, I asked ChatGPT to give me some analogies comparing generative AI to large language models (LLMs), and as the stand-in for generative AI, ChatGPT tried to take all the personality for itself. For example, it suggested, “Generative AI is the chatterbox at the cocktail party who keeps the conversation flowing with wild anecdotes, while LLMs are the meticulous librarians cataloging every word ever spoken at every party.” I mean, who sounds more fun? Well, joke’s on you, ChatGPT, because without LLMs, you wouldn’t exist. 

Fun aside, it’s important to understand the difference between the two:

• Generative AI is a form of artificial intelligence that creates new text, images, video, audio, or other content based on the vast amounts of data that the generative model was trained on.

• Large language models are a form of AI that focuses on understanding text inputs (using natural language processing) and creating human-like text based on a given input. LLMs are a subset of generative AI and focus primarily on language-related tasks.

Text-generating AI tools like ChatGPT and LLMs are inextricably connected. With the advent of multimodal LLMs—or LLMs that can take in non-text inputs like videos or images and output text or images—the distinctions may collapse. Let’s explore how generative AI and LLMs coincide, starting with three interesting facts about generative AI vs large language models. 

Here, you can explore another common question: Generative AI vs. machine learning.

Three major things stand out when you compare generative AI and LLMs. 

Generative AI is a broad category for a type of AI, referring to any artificial intelligence that can create original content. Generative AI tools are built on underlying AI models, such as a large language model (LLM). LLMs are the text-generating kind of generative AI. 

However, generative AI can also include other types of AI—such as models that work with video, text, or audio inputs and outputs. While many are built on LLMs, some generative AI tools are trained on things like a series of videos to create realistic-looking video content. LLMs focus mostly on text, but generative AI as a technology category is broader.

LLMs used to only be able to accept text inputs, as well. When OpenAI first released ChatGPT in 2022, it was built on a text-only LLM, GPT-3. But now, with the development of multimodal LLMs, these LLMs can accept audio, imagery, etc. as inputs. For example, OpenAI released the ability to use voice chats to understand human speech and read answers using voice text. 

Both generative AI and LLMs will revolutionize industries, but they will do so in different ways. Generative AI could change the way we do 3D modeling, generate video output, or create voice assistants and other audio. LLMs will focus more on text-based content creation but still have other significant uses (and may play a role in wider generative AI options like voice assistants).

LLMs have been around since the early 2010s, but they gained popularity when powerful generative AI tools like ChatGPT and Google’s Bard (now known as Gemini) launched. Everest Group notes that one reason 2023 saw such exponential growth is the expansion of parameters in large language models, with GPT-4 having more than 175 billion parameters.

They’re not only growing in terms of size—we’re seeing a proliferation of generative AI services on the market as well. Large, general purpose tools include chatbots like ChatGPT (powered by various models like GPT 4) from OpenAI, Anthropic’s Claude, Google Gemini, and Llama by Meta.

Generative AI tools specifically for creating new types of content have also hit the market. Some common models include Midjourney and the DALL-E image generator by OpenAI, which allows you to create images. There are also video generators like Runway ML and Synesthesia. The point is, there’s no shortage of generative AI solutions to choose from.

This chart outlines the key differences:

In summary, what’s the difference between LLMs and generative AI? An LLM is a type of AI model that uses machine learning built on billions of parameters to understand and produce text, while generative AI is a category that contains a myriad of tools built to use information from LLMs and other types of AI models using machine learning to generate new content.
So is ChatGPT right about the chatterbox vs. the librarian?

I hate to admit it, but perhaps it is.

So when should you apply each form of AI? 

LLMs can be used for a wide range of text-based tasks, such as language translation, content generation, and content personalization. They can also power customer service chatbots that respond to inquiries from humans, and they’re commonly used with copilots that act as virtual assistants. These include code-generation tools for building new apps or generating answers from critical business documents.  

Generative AI models can include these tasks, but can be broadened for other creative fields like image generation, music composition, or video creation. However, it's critical to note that generative AI still uses the LLM for their text-based tasks. 

While these two technologies have consumer uses, they also have numerous business applications that help enterprises across a range of tasks. Deploying generative AI in critical business processes can offload repetitive tasks from workers to AI, freeing them up for higher-value work and increasing work efficiency.

Imagine a financial services customer service team is overwhelmed keeping up with the daily deluge of customer queries. Their customer support systems get quickly overwhelmed by overflowing inboxes and lengthy support phone queues. Because generative AI can understand human language, it can offload some of these tasks and speed up the process. For instance, a genAI input prompt could extract critical information from an email and transfer it to a database without requiring manual user input from an overworked employee. The customer service rep can then have an email response automatically created using AI-aided content generation tools, which they can review before responding to customers. 

Important to note: Generative AI isn’t the only AI tool that can be used for customer service. Discriminative models also help. These tools can be used to determine the difference between classes of information, such as whether an incoming request is related to billing, technical support, or fraud prevention. AI is a broad field, so it’s critical to note where a specific type of AI, such as a discriminative model, may be used for something like content classification or where it may be better to use generative AI to create content from scratch.

A customer asks a case worker a question about their case. Rather than going through each email, document, and chat transcript to find an answer, the case worker asks a large language model to provide a summary of data related to the question. The LLM provides a content summarization of the key players, case highlights, and suggested next steps. In this scenario, the customer was also having technical trouble uploading documents to their case, so the case worker uses a generative AI-powered video creation tool to send them a video walkthrough of the process.

A marketer wants to create a synthetic audience persona via generative AI. They prompt an LLM with questions like “Where does my persona get their news?” or “How does my persona like to be communicated with?” and use the responses to craft a story about their persona. Once done, they take that information and prompt a generative AI tool to create images that represent that persona. 

An analyst takes a data file and uploads it to an LLM. They ask the tool to analyze the data and provide trends. The analyst vets the trends and uses their knowledge of the data’s context to select and edit only the trends that make sense. They then use a generative AI tool to create charts that display the trend data in their organization’s brand colors. 

As you can see, generative AI is a big, broad category that includes multiple models—LLM is one of those that’s gotten a lot of attention (and LLMs are certainly versatile), but they’re just one type of generative AI. So really, it’s not about generative AI vs large language models—it’s about combining the two and using the best tool for the job.