What is an LLM? The ‘Explain It Like I’m Five’ Version

Okay, so you’re here looking for ‘What is an LLM?’. You’ve already made the first step, but I know how confusing and complicated it can get very quickly. It’s like you miss one little detail and suddenly your understanding of LLMs is shattered. So this post aims to explain Large Language Models (LLMs) in a way that a five year old could understand. Don’t worry if you still don’t get it… this five year old would have to be a super genious.

What is and LLM?

If you’re looking into LLMs then I am sure you will have used ChatGPT or Google Gemini (if you haven’t then you should go try them right now). Both their chat AIs, the ones you can chat with through your keyboard, are examples of LLMs. It’s effectivly an AI that can take text as an input and generate text as an output. Hopefully, with the right ‘training’, a text output that makes contextual sense based on your input. In five year old terms, it’s a robot that you give a sentence, and it reads it and gives on back.

The “Auto Correct” on steroroids

LLMs in fact, act very similar to the auto-correct, or ‘predict next word’ when typing on your phone, except to the max, and instead of predicting what the next word in that text to your mum is going to be, it’s predicting the next word in it’s very own sentence.

That’s right. LLMs don’t even know what they’re going to say while they’re saying it. LLMs are simply massive prediciton moddels, that just happen to have read every book, blog, and article in the word for ‘training’, and they simply predict one word after another.

Which leads us neatly on to the real quesiton that people are asking when they ask ‘What is an LLM?’.

How does and LLM work?

So now we know that LLMs are simply just huge prediciton models that have been trained on the majority of mankinds knowledge, how do they work? You know… under the hood? To understand this, we need to understand a few key principals of LLMs.

Your characters are input parameters

When it comes to LLMs, every letter, space, and comma you type acts like a little nudge to the AI (under the hood this is all numbers and super complex math, but since we’re keeping this for a five year old, it’s just a nudge for now). If say for example, you type ‘The cat sat on the…‘, the model looks at those specific characters, groups them into words that all have numerical values, and then passes those through it’s prediction model to start it’s response.

Because it’s been ‘trained’ on billions of sentences, it knows that the word ‘mat’ has a very high probability of coming next, while the word ‘fridge‘ has a very low one. It’s not thinking about a cat; it’s just doing math to find the most likely next piece of the puzzle.

It predicts a word — then the next — and the next

This is the truly ‘magic’ part. Once it predicts the word ‘mat’, it doesn’t just stop. It then takes your original sentence PLUS its own new word and feeds them both back into the prediction machine that we’ve been learning about. So for example;

1. Input: ‘The cat sat on the…’ → Predicts: ‘mat’
2. New Input: ‘The cat sat on the mat…’ → Predicts: ‘and’
3. New Input: ‘The cat sat on the mat and…’ → Predicts: ‘fell’

It does this thousands of times a second until it finishes a full thought. It’s basically the world’s most advanced game of ‘connect the dots’, where the dots are words and the final picture is an answer to your question. You might be able to see why LLMs use so much compute power now.

Why does this matter for your business?

Because LLMs are just giant prediction engines, the ‘training’ we mentioned earlier is everything. If you train a model on your company’s specific data, it stops predicting ‘generic’ internet answers and starts predicting answers that are specific to your customers, your products, and your workflows.

It’s a robot that has read your entire company handbook and is ready to chat about it. So weather it’s an AI chatbot on your website that answers customer queries and improves sales, or it’s an internal AI tool that digs out answers through thousands of internal company documents, AI is an invaluable tool in the modern business landscape.

Want to talk about AI?

I’m Douglas, I’m the fouder of HelloHorizon here in Norwich, and I’ve seen how Norfolk has been a completly underserved area when it comes to AI integrations for businesses, with many business owners I know choosing to avoid it completly instead of understand it and unlock the benfit. So I started AI Consulting Norwich to help local business owners like you, understand, strategize, and implement AI into their businesses. Send me an email today at [email protected] or leave a message on the contact form of this website. I’m happy to help 🙂

You can also checkout the page on HelloHoriozn about Automation & AI, which gives a few ways that AI can be practically implemented in to a business.