How do LLMs Work?

This is a very basic primer for non-technical folks!

LLMs are basically really really good at predicting words. They’re specifically trained to predict the next word in any reflection for you.

It’s important to understand how they work as we use them for our work.

When I began looking into AI seriously, it was at the end of last year, upon the spectacular launch of Chat GPT and the news that Microsoft had invested $10B in OpenAI, the group that developed it.

My news feeds exploded with plenty of information about AI technology and tools, and the language surrounding them. There are a few that we’ll explore here, purely from a business perspective.

What is a Large Language Model?

I asked my favorite AI tool du jour (and Large Language Model itself, Claude, from Anthropic, how to explain a Large Language model, using the following prompt - “Define Large Language model for me, my audience are non-technical professionals in business:”

Here is how I would explain large language models to non-technical professionals:

A large language model is a type of artificial intelligence or AI that has been trained on massive amounts of text data. This includes things like books, articles, websites, social media posts, and more. By exposing the model to such a huge pool of text data - often billions or trillions of words - the model learns the patterns and statistics of how humans use language.

The key benefit of large language models is that they can understand and generate very human-like text. This gives them powerful abilities like summarizing long documents, answering questions, translating between languages, and holding natural conversations.

For example, a large language model could read thousands of customer support tickets and learn to understand common questions and provide useful answers. Or it could be given millions of sales call transcripts and learn to have natural sounding sales conversations.

The technology is still early but improving rapidly. Large language models allow AI to communicate with people using normal, conversational language instead of just simple commands. This makes interacting with AI much more flexible and broadly applicable. It's an important step towards what is known as artificial general intelligence.

So in summary, large language models are AIs that have learned nuances of language from huge datasets, allowing more human-like communication abilities. This makes them very useful for a wide variety of applications.

How Do LLMs Work?

One of my first, best resource blog posts that gave me a visual ‘aha’ moment was an “Understanding ChatGPT” post by Jeff Prosise at Atmosera. Atmosera is a Microsoft business partner doing incredible work leveraging AI tools and the Azure cloud services. The post itself is intended for a technical audience, but he made a great picture of how these models work.

In short, this is how a LLM would translate the English phrase, “hello world” into “salut le monde” in French:

I found myself going back to the visual explanation over and over again, as I deepen my understanding of AI tools, how they work, and how they will continue to assist us in years to come. I hope it sticks for you, too!

7.31.23 Update - if you’d like to dig in further and learn more about how the models work, including word vectors and transformers, there’s a great article in Ars Technica about just that, here.

Resources from AIGG on your AI Journey of Understanding

Speaking of stickiness, have a look at our Resources section where you’ll find free drop-in HR Handbook policies for your team to review, augment and approve.

We have attorneys, anthropologists and business leaders to support you as you develop your Strategic AI Use Statements that can guide your organization’s use of the tools available to you. And we have bespoke educational workshops available to you as you explore your options.

Connect with us for more information, to get your free Legal and Operational Issues List, HR Handbook policy, or to schedule a workshop to learn more about how to make AI work safely for you. We are here for you.