Over the past few months, I’ve started building apps with AI and joining AI hackathons, and I keep running into words I’d never heard before.
Frontend, backend, CLI, API…
And then there’s one more.
LLM.
L-L-M.
It comes up all the time whenever people talk about AI, as if everyone already knows what it means.
I use ChatGPT. I use Claude Code and Codex, too. But what exactly is an LLM?
Is it ChatGPT? Is it just another word for AI? Or is it something else entirely?
At my second AI hackathon, I was on a two-person team, and my teammate asked me, “Which LLM are you using?”
I answered, “I don’t know.”
To be honest, I didn’t even know what the word LLM meant.
I tried asking ChatGPT, but I still didn’t really get it.
The next day, I did a little research of my own.
I came away feeling like I sort of understood it.
Then came my fourth AI hackathon.
At Vietnam Japan AI Hackathon 2026, I’m on a two-person team again.
Recently, we put together our team’s Idea Proposal for the first round of judging and submitted it.
The proposal had a field for the LLMs we planned to use, and my teammate had filled it in:
Qwen, Gemini, Chat GPT.
By the way, that was the first time I’d ever seen the name Qwen.
When I looked it up later, I learned that Qwen is apparently a series of AI models developed by China’s Alibaba Group.
Wait, Alibaba?!
That surprised me a little.
I sell products under my own brand on Amazon. Sometimes I also have factories in China make products for me as OEM products.
So to me, Alibaba has always been a place to source products.
And that same Alibaba was developing AI, too!
I had absolutely no idea.
That’s when Amazon came to mind.
For me, Amazon had always been “the place where I sell products.”
But about two months ago, I learned about AWS (Amazon Web Services).
It turns out Amazon isn’t just an online store. It also provides cloud services, and even offers AI-related services.
That surprised me, too.
Alibaba was where I buy. Amazon was where I sell.
That’s how I’d always thought of them.
But once I started learning about AI, I began to see sides of these companies I’d never known about.
E-commerce and AI are a lot more connected than I thought!
Actually, there’s an interesting difference among the three names my teammate wrote, “Qwen, Gemini, Chat GPT,” too.
Qwen is a series of AI models. Gemini is the name of AI models and also the name of an AI app. And ChatGPT isn’t an LLM itself. It’s the name of an AI service.
I see. So a model and a service aren’t the same thing.
Then what is an LLM, anyway?
In this article, I’m going to sort out the term “LLM” from a beginner’s point of view, as a non-engineer.
What Does LLM Stand For?
LLM stands for Large Language Model.
Put simply, it’s an AI model trained on huge amounts of text, code, and other data so that it can work with human language.
For example, it can answer questions, write text, summarize long documents, or translate into other languages.
It can even write program code.
A lot of the work I’ve been asking AI to do lately actually relies on these abilities.
But how does it actually produce text?
Roughly speaking, an LLM looks at the words and context it’s given, predicts the next small piece of text (called a token), and builds up text by repeating that over and over.
That said, it isn’t simply stringing words together. Through large-scale training, it has become able to pick up complex patterns in language and code.
Even so, that doesn’t mean it understands or experiences things the way humans do.
Are ChatGPT and LLMs the Same Thing?
This is the first thing I want to sort out.
ChatGPT and LLMs are not the same thing.
ChatGPT is an AI service where we can ask questions or talk things through on screen.
An LLM, on the other hand, is the AI model at the core of that service that processes language.
Here’s roughly how I picture it:
| Name | What does it refer to? |
|---|---|
| LLM | An AI model that processes and generates language and code |
| GPT | A family of AI models developed by OpenAI |
| ChatGPT | A service that uses GPT and other models so you can chat and get tasks done |
| Claude | The name of Anthropic’s AI models and of the AI assistant that uses them |
| Claude Code | A tool that uses AI to help with software development |
| Codex | OpenAI’s AI coding agent |
In other words, it helps to think of the AI services and tools we use every day separately from the LLMs working inside them.
Asking a question on the ChatGPT screen and having Claude Code fix a program are both ways of using what an LLM can do.
But services and tools also combine other features besides the LLM.
LLMs Can Write Programs, Not Just Text!
For me, this is the especially fun part.
I’ve never worked as an engineer.
But lately, I’ve been using AI tools like Claude Code and Codex to build apps and blog systems I need for my work.
For example, tools related to managing my Amazon products, and Hackoo, the site where I share information about AI hackathons.
I explain what I want in words, and AI writes the code or fixes it for me.
Of course, I don’t just leave everything to AI.
I check whether things actually work, ask for fixes when something looks wrong, and approve important changes myself.
But the old me probably would never have even thought about building a program.
Ask in plain words, and it’ll even write the code for you.
The first time I experienced that, I was seriously amazed.
So LLMs aren’t just a technology for writing text.
What’s the Difference Between an LLM and an AI Agent?
Here’s another term I keep hearing lately:
AI agent.
An LLM and an AI agent aren’t the same thing either.
An LLM is an AI model used to understand and generate text, and to figure out what to do next.
An AI agent, on the other hand, is a system that works through multiple tasks toward a goal.
For example, say you give the instruction, “Fix this program.”
If you’re just asking an LLM to suggest a code fix in writing, it might end there.
But an AI agent that can use tools can work through a whole series of steps: actually reading files, changing the code, running tests, and checking the results.
Of course, how much it can actually do depends on how the agent is designed and what permissions it’s been given.
And not every AI agent necessarily uses an LLM, but many AI agents use an LLM at the center of their decision-making and conversations.
Lately, I’ve been getting interested in how these AI agents work, too.
From AI that just answers questions to AI that actually moves real work forward.
That’s the part I find really exciting right now.
How Are LLMs Used at AI Hackathons?
At AI hackathons, you turn ideas into something real within a limited amount of time.
That’s why some teams build apps or AI agents that use LLMs.
For example, an app that takes questions from users and answers them.
A tool that organizes text you enter and pulls out the information you need.
Or an AI agent that works through several tasks in combination.
At Vietnam Japan AI Hackathon 2026, where I’m taking part, I’m working with a Vietnamese AI engineer, just the two of us, on FrostLoop AI, an AI agent for Amazon sellers.
The technical implementation is mostly handled by my teammate.
I’m involved through my hands-on experience as an Amazon seller, how the service should be used, design, and so on.
So even in projects that use LLMs and AI agents, not everyone has to know the details of how AI models work.
What problems do real users have? What kind of screen would be easy for them to use?
I think perspectives like those matter a lot when building a product, too.
But LLMs Aren’t Always Right
This is something I never want to forget when I’m actually using AI.
LLMs produce very natural-sounding text.
So when you read it, it’s easy to feel like, “This must be right.”
But even text that sounds confident can be wrong.
When AI gives information that isn’t true but makes it sound believable, that’s called a “hallucination.”
Also, an LLM by itself doesn’t always know the latest news.
To look up the latest information, you need something like a search feature, and for anything important, I want to check it against reliable sources.
The same goes for programs.
Even if AI says, “I fixed it,” you still need to test it to make sure it really works.
Because it’s so convenient, that final check really matters.
Can You Build Things with AI Without Knowing Much About LLMs?
I think you can.
Of course, there are plenty of situations where understanding how it works helps you use it better.
But you don’t have to memorize every technical term and master programming before you start using AI.
I’m still building things with AI myself, looking up unfamiliar words one at a time as I go.
LLM was one of them.
Try it.
Run into something you don’t understand.
Look it up.
Then you understand how it works a little better than before.
And you just keep repeating that.
Joining AI hackathons feels kind of similar to me.
Instead of understanding everything before you start, you start and understand as you go.
That’s how I’m enjoying the world of AI right now.
There are still so many words I don’t know.
I bet I’ll run into new ones at my next hackathon, too! (Haha.)