What are Large Language Models (and How Do They Work)?


With the emergence of gen AI-powered tools like ChatGPT, large language models have exponentially expanded the potential of CX automation. But you may still be wondering just what exactly makes this tech so special, so let’s break it down.  

While you may not hear the term ‘large language models’ (LLMs) circulating as much in the latest discourse around AI, you’ve likely already tried using one. That’s because LLMs power a growing number of tools including Google’s Bard, Dall-E, Midjourney, and most famously: ChatGPT – which has become something of a household name over the last year. And at this point, it has become more than mere hype. 

Interest in this new generation of artificial intelligence has exploded, as the gen AI market is now valued at $42.6 billion in 2023, with CX automation among the most promising use cases for it. Forward-thinking companies across many industries – from travel to telemedicine – have already started integrating LLMs into their customer support over the last months. In fact, the use cases around what LLMs can do are getting more sophisticated by the day.

So best to wise up on what makes this tech beneficial to your business, and how it works to enhance your support. In this article, we’ll cover all this in depth and also share insights from those at the forefront of the CX automation industry on what’s to come.

What are large language models (LLMs)?

Large language models are a form of technology that powers text-based generative AI models, an automation concept you might be more familiar with. The tech is known for generating unprecedentedly natural, human-like conversations that have even passed the Turing test, an assessment of how well AI can mimic human intelligence. That’s because LLM-based bots are trained on a vast amount of written content published on the internet and written by real people. 

In the case of ChatGPT, for example, you can do things like ask it to write a business proposal in the tone of a pirate, or prompt it to muse, at length, about the meaning of life. In addition, bots powered by LLMs can instantly summarize text, perform translations, and analyze the sentiment of the person it is speaking with. Pretty cool, right? 

But you may still be wondering – but perhaps afraid to ask – how large language models work. Understanding this will help you make the most out of this tech in your industry. Not to mention, it’ll make you sound way smarter at the next company gathering. So let’s dive into the specifics. 

LLMs work by processing all data inputs at once

What actually happens when you connect your bot to a knowledge source -- such as your help center or FAQs page -- is what sets large language model technology apart from its predecessors. Specifically, these bots are built on a type of artificial neural network known as transformers, which enables them to instantly and simultaneously analyze all data inputs at once. 

This is a fundamental difference from more traditional AI-powered chatbots, which require training data to perform well. In practice, this ability to analyze data holistically makes LLM-based bots far better at contextualizing the conversations they have rather than simply providing canned, pre-programmed responses. They are also able to perform specific tasks with few or no examples given to them. In addition, contextualized question answering is another helpful feature of LLM-based bots. This means that they can answer questions from given text and contextualize the answer with respect to it.

These bots are, overall, able to provide accurate answers in a conversational way – all without needing to be trained or overseen by a bot builder with a vast technical background. This means that as long as you have a generative AI provider and a functional knowledge base, then you can start automating, even without any prior technical know-how.

How LLMs are used in customer service 

Naturally, the elevated capabilities of LLM-powered chatbots translate well to customer support, where quality conversational experiences are critical to your overall CX. It’s no small wonder, therefore, that seemingly everyone is jumping on this bandwagon. In fact, according to our in-house survey, 60% of business leaders reported they were more likely to adopt AI in 2023 than the previous year. Beyond merely trying to keep up with the Joneses, however, there are plenty of good reasons for automating your support with the help of LLMs.

Scale your support, while also lowering cost per interaction

In the wake of the recent recession, many businesses are accustomed to the challenge of trying to do more with less. Now business may be picking up, but resources around budget or staffing might still be limited. For these reasons and more, hiring and training new agents is a complex and costly process.

European car-rental platform, DiscoverCars was well acquainted with this predicament. Read how they used UltimateGPT, our CX-specialized LLM-based bot to enhance their support.

Build a bot in minutes 

LLM-based bots provide quick time to value because you can start automating within minutes – no prior technical training required. Simply connect your knowledge source to the generative AI tool of your choice and your bot can begin helping your customers in a natural, conversational way. 

(Pssst are you looking for a LLM-based bot specialized for customer support automation?  Check out UltimateGPT.)

Offer customers 24/7 support in the language of their choice

Beyond being budget friendly, LLM-based bots can also improve the quality of your CX. They can help you to provide round-the-clock customer care so that people can instantly get help in resolving simple queries such as checking order status, requesting transaction info, or changing their password. They can do all this while having a natural conversational experience – as opposed to sifting through FAQ pages or having a disjointed dialogue with an earlier generation chatbot.

The beauty of LLM-based bots in particular is that you can provide this level of service in multiple languages, as they can instantly translate the contents of your knowledge base to converse with customers with native-level proficiency in whichever language they prefer. 

Make your human agents jobs easier – and more rewarding 

The elephant in the room in light of the technological innovations ushered in by LLMs, is where does this leave your human agents? Rather than simply stealing their jobs, automation has been shown to elevate the role of your human support agents. As our very own COO, Sarah Al-Hussaini put it in a recent event, “Getting Started With Automation: Bringing Human and AI Agents Together,”

“Integrating automation into your support offering allows your human team to become what they should be: a white glove service that can offer a white glove experience.”

– Sarah Al-Hussaini, COO, Ultimate

When the more mundane tasks are removed from your human agents’ remit, they have more time – and bandwidth – to handle the more complex and interesting tasks: those that require empathy and creative problem solving. This has the potential to increase job retention, open up space for upskilling, and power more rewarding careers

Future-proof your support (and better handle seasonal spikes) 

Generative AI is here to stay, and for good reason. It’s no secret that in addition to more predictable, industry-specific busy seasons, we’ve been hit with all kinds of unforeseen challenges in the last few years. From recession proofing your support or braving support volume surges during a pandemic, LLM-based automation can absorb the brunt of these challenges – without compromising the quality of your existing support.

The customer service team at TeleClinic is no stranger to unexpected surges. They've handled everything from Covid to hay fever season, which used to throw their support strategy into a tailspin – until they started to automate. Now, with a 37% automation rate, agent workload remains steady no matter what is happening in the outside world. Not to mention, they’ve garnered a cool €100,500 in annual savings. 

Looking ahead: ‘reasonably-sized’ language models 

While it may not exactly have the same ring to it as ‘large’ language models does, we expect that ‘reasonably sized’ language models will overtake the standard LLM within the years to come. Instead of having hundreds of billions of parameters, these language models can run just fine on tens of billions of parameters instead. These medium or even smaller, ‘fun sized’ language models will be cheaper to run and, therefore, it’ll become that much easier to reap the benefits of CX automation.

What’s more is that language models are likely to become more specialized to specific verticals, as they can be trained on industry-specific customer support data. This, in turn, will further increase the accuracy and efficiency of these chatbots. That’s why, in 2025 and beyond, we are going to see a move towards verticalization of language models as they get really specialized. It’s important to find an AI provider with experience automating in your industry that embraces the use of this new technology. 

LLM-based bots are easy to use, and they provide a pathway to scaling your support without compromising on the quality of your CX – or eating up your budget. With the technical barrier to automation lower than ever, anyone can get started with this cutting edge technology. And trying it now can help future-proof your support in the years to come.


Build your own LLM-based bot in minutes