Generative AI Explained!
- Leoza Kabir Barker
- Jun 26
- 4 min read
Updated: 4 days ago
Generative AI is everywhere, but let’s be honest, it can be confusing. When I first heard that it "predicts the next word," I didn’t understand what that actually meant. It sounded far too simple to explain how these models write long essays, answer complex questions, or generate realistic images.
If you’ve been nodding along in meetings or scrolling through posts without fully grasping how it works, this article is for you. My goal is to break it down with practical, plain-language examples that make everything click.
What Is Generative AI?
Generative AI is a form of artificial intelligence that goes beyond analyzing data. It creates something new. This could be a story, an image, a poem, a piece of code, or even a product prototype.
It works by learning patterns from massive datasets, millions of examples of human-written text, images, or code, and then uses that training to generate original content that follows similar patterns.
Unlike traditional AI, which focuses on classification or prediction (e.g., "Is this a cat or dog?"), generative AI learns the structure and style of the data it’s trained on and creates entirely new content that reflects that knowledge.
As Microsoft describes it:
"Like a chef who cooks unique dishes from familiar ingredients, generative AI creates something new by remixing what it has learned."
How does generative AI work
At the heart of most generative AI systems, like ChatGPT, is a model called a large language model. These models generate content by predicting the next word in a sequence, over and over again, based on patterns learned during training.
You give it a starting prompt, and it builds a response one word at a time, selecting the most likely next word with each step.
At its core, this complex process comes down to one simple question:
"Given everything I’ve seen so far, what is the most likely next word?"
The model does not pull responses from a database or rely on pre-written scripts. Instead, it generates content in real time by continuously predicting and generating one word after another.
Let’s walk through two examples.
Example 1: Dolphin Fun Fact
Prompt: "Tell me a fun fact about dolphins."
Step-by-Step Breakdown:
Step 1: Read the Prompt
The model analyzes the prompt and identifies the intention. It has seen many examples of questions asking for fun facts about animals.
Step 2: Predict the First Word
It considers several options: "Dolphins," "They," "Did," or "One."
Let’s say it chooses "Dolphins."
Step 3: Predict the Next Word
It now asks, what follows "Dolphins"?
It may consider "are," "can," "have," or "use." Suppose it picks "are."
Now the sentence is: "Dolphins are."
Step 4: Continue Word by Word
It continues building:
"Dolphins are known"
"Dolphins are known to be"
"Dolphins are known to be intelligent"
"Dolphins are known to be intelligent animals that use tools..."
Final Output:
"Dolphins are known to be intelligent animals that can use tools, communicate with each other through whistles and clicks, and even recognize themselves in mirrors."
While it seems like the model "knows" these facts, it's simply predicting words based on the patterns it learned from a wide range of content about dolphins.
Example 2: Noisy Neighbour
Prompt: "My neighbour is very sensitive to noise. What measures can we take to keep the noise level down?"
Step-by-Step Breakdown:
Step 1: Understand the Intent
The model identifies the topic as noise sensitivity and recognizes that you’re seeking practical, polite advice.
Step 2: Predict the First Word
It considers possible beginnings like "There," "You," or "Some." It chooses: "There."
Step 3: Predict the Next Words
"There are"
"There are a few"
"There are a few steps"
"There are a few steps you can take to reduce noise"
Step 4: Continue Building the Response
"There are a few steps you can take to reduce noise. First, consider using rugs or carpets to absorb sound. You can also install door sweeps or weatherstripping to minimize sound leakage. If possible, keep louder activities to certain hours of the day and let your neighbor know in advance. Soft music or white noise machines can also help mask sounds. Communication and small adjustments often go a long way."
The model constructs tone, phrasing, and sentence structure based on similar examples it has seen before.
How It Works Behind the Scenes
According to resources from MIT, Microsoft, and the University of Toronto, here’s what happens under the hood:
Your input is tokenized (broken into chunks like words or subwords)
A neural network model called a transformer processes these tokens
It uses an attention mechanism to weigh the context of each word
It then predicts the next token based on that context
This continues until a complete response is formed
All of this happens in milliseconds, backed by billions of parameters and trained on large, diverse datasets.
Why It Feels So Smart
Generative AI feels intelligent because it:
Understands tone, structure, and context
Maintains coherent and relevant conversation flow
Adjusts output based on the prompt style
But it’s not thinking. It’s simply following patterns it has seen and producing the most probable output.
The Limits
Despite its impressive output, generative AI has well-documented limitations:
Hallucinations: It can generate incorrect or misleading information
Bias: It reflects biases present in its training data
Copyright: It may unintentionally replicate protected content
Opacity: It’s difficult to explain exactly why it chose a specific response
What Can You Use It For?
Generative AI is useful across a variety of applications:
Writing emails, blogs, and social content
Brainstorming and ideation
Summarizing long documents
Generating code
Creating visuals and graphics
Writing music, poetry, or dialogue
The possibilities are rapidly expanding.
Sources That Helped Me Understand
MIT News: Explained - Generative AI
Microsoft AI 101
University of Toronto Library Guide
If this post helped clarify how generative AI works, feel free to share it with someone who’s been curious but unsure. Understanding the basics is the first step toward using it effectively.
About Me
I'm Leoza Kabir Barker, a Functional Architect at Alithya with a focus on the Power Platform. Through my expertise, I aim to streamline processes, optimize operations, and maximize productivity.
Connect with Me
Blog: www.leozakabirbarker.com
Comments