Beyond the Buzzword: What is an LLM?

3 August 2023 | Ella Hislop

A quick Google search of the terms “LLM" and “AI” brings up 79 million results. For a term that has only risen to prominence in the last six months, it’s generated an incredible amount of interest. Today, we’re taking a deep dive into what an LLM actually is, how they work, and how we apply Luminance’s LLM in the legal world.

What is an LLM?

A Large Language Model or ‘LLM’ is based on algorithms that use vast amounts of data to understand, summarise and produce content. This allows LLMs to perform a wide range of Natural Language Processing tasks (or in other words – text-based tasks) like providing conservational chatbot assistance, summarising data or translating content. Famous examples of LLMs include OpenAI’s GPT-3 and Google’s BERT.

How does an LLM work?

The important thing to know about LLMs is that data is absolute rocket fuel for them. To that end, they’re trained on vast amounts of data from across the internet as well as books, articles and even social media posts. We’re talking hundreds of billions of datapoints here! It’s during this training on huge datasets that the model learns the patterns, grammatical and semantic structures present across text. The more data that is inputted into the model, the better it becomes at predicting how words are connected with each other.

Once an LLM has been through all this training, it works in a remarkably similar way to the human brain. Much like we can connect different concepts, words, and phrases, LLMs can do the same. And this is all thanks to state-of-the-art techniques in Natural Language Processing called Transformers. Rather than just examining text word-by-word, Transformers form a very good understanding of what each word actually means in the context of a full sentence.

How does Luminance use an LLM?

Generalist LLMs, such as ChatGPT, undergo training using a vast dataset extracted from the entirety of the internet. However, this immense breadth of training data might not always align with the specialised knowledge and requirements demanded by certain industries.

So, what we’ve done at Luminance is develop our very own legal LLM through a process called fine-tuning. This is achieved through exposing the model to vast amounts of domain-specific training data. Luminance’s pioneering ‘legal-grade’ AI has been informed by 150+ million verified legal documents, creating a highly specialised LLM which avoids the potential pitfalls of generalist LLMs – not least ‘hallucinating’ fake precedent which, as displayed by recent headlines, risks immense reputational damage.

Luminance’s LLM underpins many features of our AI, including its ability to identify 1,000+ legal concepts out-of-the-box, and Ask Lumi, our first-of-its-kind chatbot that allows users to ask questions about their contract and receive instant responses. Lawyers can even redraft contracts with a click of a button using Ask Lumi.

What benefits can LLMs bring to me and my team?

LLMs bring a wealth of opportunities to teams across all industries. In a legal context, Luminance can ease the burden of labour-intensive tasks, such as large-scale contract reviews or negotiations, and therefore streamline much of the commercial process. When I run a Proof-of-Value with businesses around the world, they always highlight how Luminance is up and running within minutes – much of this is thanks to the LLM’s ability to read and understand legal documents out-of-the-box! Legacy technologies, meanwhile, often require weeks of dedicated training, creating further stress on limited resources.

The recent call to train AI with legal documents underscores the significance of developing specialised LLMs for specialised industries. We couldn’t agree more.

Interested to see our ‘legal-grade’ LLM in action? Request a demo today.