What is Generative AI? A Gentle Introduction to Understanding the Differences, Mechanisms, and Capabilities of LLMs
“I use ChatGPT every day. But what exactly is an LLM?”
Surprisingly, not many people can answer this question clearly.
In fact, 58.8% of people in Japan have already used generative AI
(according to the 2026 Information and Communications White Paper published by the Ministry of Internal Affairs and Communications on July 24, 2026. This is more than double the 26.7% from the previous survey just a year ago, with about 30% of respondents saying they use it “almost every day”).
More people are using it. But few understand what is inside.
This gap is exactly what creates the difference in proficiency.
By reading this article, you will understand the following three things:
The difference between generative AI and LLMs — including their relationship with AI, machine learning, and deep learning
Why they make mistakes — understanding the mechanism makes the cause intuitive
Tasks you can delegate and tasks you shouldn't — drawing the line in practical work
This article is for non-engineers who use AI tools at work. The goal is not to provide a technical explanation, but to help you judge “how much you can trust AI.”
Conclusion: Generative AI is “AI that creates,” and LLMs are “the engine inside that handles language.”
Generative AI is a general term for AI that creates new text, images, audio, etc.
LLMs (Large Language Models) are AI that can handle language like a human by learning from vast amounts of text. They are the “engine inside” tools like ChatGPT.
LLMs are just predicting the next word. That is why they can return plausible-sounding text even if it is not factual.
The third point is the most important part of this article.
※ This article contains promotions (affiliate links). If you want to learn about this field comprehensively, a book for non-engineers is the shortcut.
The relationship between Generative AI, LLMs, and AI
First, let's look at the big picture. These three are not parallel; they are nested.
The outermost layer is AI. This includes shogi AI, image recognition, translation AI, autonomous driving, and more.
Generative AI is a type of AI, but it is not all of AI.
Inside that is machine learning, which learns patterns from data.
Inside that is deep learning, which learns using mechanisms modeled after the human brain.
Born from that is generative AI.
Among generative AI, the type that specializes in language is the LLM (Large Language Model).
It is an AI that has learned a massive amount of text and can read and write like a human.
Is ChatGPT a generative AI? Or is it an LLM?
This is a point of confusion, so let's clarify it.
ChatGPT, Claude, Gemini, and Copilot are the names of 'services'.
The engine running inside them is the LLM.
In terms of smartphones, ChatGPT is the iPhone, and the model inside is the iOS.
Even on the same device, the OS gets updated, and when it is updated, the performance changes.
News saying 'ChatGPT has become smarter' usually means the model inside has been updated.
The AI chat you use daily should also have a screen for selecting models.
Think of the names listed there as the generations and types of LLMs.
Why do they tell plausible lies?
This is the main point. What an LLM is doing is only predicting the 'next word'..
When 'The capital of Japan is' is entered, it selects a word with a high probability of coming next. Then it predicts the next word again. By repeating this, a sentence is formed.
In other words, an LLM does not choose based on whether it is a fact, but based on whether it is plausible.
That is why it may confidently write titles of books that do not exist or legal clauses that do not exist. It is not trying to lie; it just so happens that the result of lining up plausible words is different from the facts.
This phenomenon is called hallucination (more details on this in the next topic). For now, just remember it as 'something that cannot be eliminated due to the mechanism'.
Knowing this one point determines your distance from AI.
You can trust the form of the text. Do not trust the factual content.
Tasks to delegate vs. tasks to do yourself
Once you understand the mechanism, the line between them becomes naturally visible.
The boundary is simple.
You can delegate tasks where there is no single correct answer.
You should verify tasks that have only one correct answer yourself.
Drafting, summarizing, paraphrasing, and brainstorming are the former. Numbers, proper nouns, the latest events, and calculations are the latter. Think of it not as 'AI is useless,' but as a question of 'how much should I hand over'.
I personally use AI every day for work.
Even so, I always verify numbers and proper nouns myself.
I continue to do this without exception. As long as an LLM is only predicting words, it will not work unless a human takes responsibility for this part.
It has become a 'given' even in companies
According to the same white paper, the percentage of companies using generative AI for some kind of business operation is 86.4% (previously 55.2%). It is no longer an exception.
On the other hand, 27.0% of companies in Japan report having "no organizational initiatives" to change how work is done. This is significantly higher compared to 1.4% in the U.S., 4.9% in Germany, and 2.6% in China.
The tools have been introduced, but the methods haven't changed. Many workplaces are currently at this stage.
Frequently Asked Questions
In simple terms, what is generative AI?
It is a general term for AI that creates new text, images, audio, and more. Its defining feature is that it creates content on the spot rather than selecting from existing answers.
What is the difference between generative AI and LLMs?
Generative AI is the broad category, while LLMs (Large Language Models) are the subset that handles language. Image generation AI is also generative AI, but it is not an LLM.
Is ChatGPT a generative AI or an LLM?
ChatGPT is the name of the service, and the LLM is the engine running inside it. Claude and Gemini share the same relationship.
What is the relationship between AI, machine learning, deep learning, and generative AI?
AI is the broadest category, which contains machine learning, which in turn contains deep learning, and generative AI is a nested concept born from that.
Why does generative AI return incorrect answers?
It is because it selects words based on how plausible they are as the next word, rather than whether they are factual. Due to its mechanism, this cannot be completely eliminated.
What is generative AI good at, and what is it bad at?
It excels at drafting, summarizing, paraphrasing, and brainstorming, but struggles with accurate numbers, the latest events, calculations, and individual judgments.
Is programming required to use generative AI?
It is not required. You can use it simply by typing in the chat box.
Can I use it for free?
Major services offer a range of free usage. High-performance models and usage limits vary depending on the plan.
Summary
Generative AI is AI that creates things. An LLM is the engine inside that handles language.
And finally, one more time.LLMs are only predicting the next word.
They only seem smart because they are good at choosing plausible words. They do not guarantee facts.
The way you use AI changes completely depending on whether you know this or not.
This is because you will be able to know when to be skeptical.
The next time you get an answer from AI,check only the numbers and proper nouns, and take a moment to verify them. Just by doing that, AI will instantly become a reliable tool.
Next time, we will cover that very "plausible lie"—known as "hallucination."
For those who want to learn more (2 books)
If you want to start reading: An introductory book on the IT Passport You can organize basic terms like generative AI, machine learning, cloud, and API in one book for non-engineers (items related to generative AI have also been added to the IT Passport exam syllabus/IPA).
If you want to go a step further: An introductory book explaining the mechanism of generative AI Once you delve into "why it can predict words," you will be able to read AI news on your own. An introductory book without mathematical formulas is sufficient.
※ This article contains promotions (affiliate links).
Read Next
What is a Prompt? A Simple Introduction to Writing Instructions for AI, Tips, and Templates (Part 3)
What are Tokens? A Simple Introduction to AI Pricing, Limits, and Context (Part 2)