What is AI?
Artificial Intelligence (AI) is a broad field in Computer Science that focuses on creating
machines or software that can perform tasks that normally require human intelligence.
These tasks include:
Understanding language (like chatbots)
Recognizing images
Making decisions
Learning from data
In simple terms: AI = machines that can “think” or act smart
Types of AI (Based on how they work)
1. Generative AI (Gen AI)
Generative AI creates new content.
It doesn’t just analyze data—it produces something new based on what it has learned.
Examples:
Writing text (like ChatGPT)
Creating images (like DALL·E)
Generating music, videos, or code
How it works:
It learns patterns from large datasets and then generates similar—but new—outputs.
Simple idea: Gen AI = “creator AI”
2. Non-Generative AI (Traditional AI)
This type of AI does NOT create new content. Instead, it focuses on:
Classifying
Predicting
Detecting patterns
Making decisions
Examples:
Spam email filters
Face recognition systems
Recommendation systems (like Netflix suggestions)
Fraud detection in banks
How it works:
It uses models trained to analyze and give answers, not create new things.
Simple idea: Non-Gen AI = “decision-making AI”
Quick Comparison
Feature Generative AI Non-Generative AI
Main function Creates new content Analyzes & predicts
Output Text, images, audio Labels, decisions, scores
Example use Chatbots, art creation Spam detection, scoring
Easy Analogy
Generative AI = a writer or artist
Non-Generative AI = a judge or analyst
What is Generative AI?
Definition: AI systems that create new content (text, images, audio, code)
Examples:
Text: ChatGPT
Images: DALL·E
Code: GitHub Copilot
Difference between:
Traditional AI (classification, prediction)
Generative AI (creation)
How does it work?
Generative AI is usually built using advanced machine learning models, especially:
Neural Networks
Transformers
These models are trained on huge amounts of data and learn patterns in language, images,
or sound. Then, when you give a prompt, they predict and generate new content based
on what they learned.
Simple example
If you type:
“Write a story about a robot in space”
A generative AI system will create a completely new story, not copy one from
somewhere.
Types of Generative AI Models
Generative Adversarial Networks (GANs) are a powerful type of generative AI model
used to create highly realistic data—especially images.
Simple idea
A GAN works like a competition between two neural networks:
1. Generator → tries to create fake data (like images)
2. Discriminator → tries to detect whether the data is real or fake
They keep improving by competing with each other.
Key concept
GANs are based on:
Neural Networks
Adversarial Learning
The word “adversarial” means the two models are fighting against each other to get
better.
How it works (step-by-step)
1. The Generator creates fake data (e.g., a fake face)
2. The Discriminator checks it against real data
3. The Discriminator says: “real or fake?”
4. Both models adjust based on mistakes
5. Repeat thousands/millions of times
Over time, the Generator becomes so good that the fake data looks almost real
Easy analogy
Imagine:
A forger (Generator) trying to make fake paintings
A detective (Discriminator) trying to spot fakes
As both improve, the forgeries become extremely convincing.
What are GANs used for?
GANs are used in many areas:
Creating realistic human faces (that don’t exist)
Image enhancement (e.g., sharpening blurry photos)
Deepfakes and video editing
Art and design
Medical imaging
A
diffusion
model is
a type of
generative
AI model
used to
create
data (like
images,
audio, or
text) by
gradually
adding
and then
removing
noise. It’s
widely
used in
modern
AI
systems
like
DALL·E
and
Stable
Diffusion.
Core Idea (Simple Explanation)
Think of it like this:
1. Forward process (adding noise): Start with a clear image → slowly add random
noise step by step → until it becomes pure static.
2. Reverse process (learning to denoise): The model learns how to reverse that
process → starting from noise → gradually reconstruct a meaningful image.
How It Works (Step-by-step)
During training:
The model takes real data (like images)
Adds noise in many small steps
Learns how to predict and remove that noise at each step
During generation:
Start with random noise
Apply the learned denoising steps
End up with a realistic output (image, sound, etc.)
Why It’s Powerful
Produces high-quality, realistic outputs
More stable than older methods like GANs (Generative Adversarial Networks)
Can be guided with text prompts (e.g., “a cat wearing sunglasses”)
Types of Diffusion Models
DDPM (Denoising Diffusion Probabilistic Models) – the basic version
Latent Diffusion Models (LDMs) – faster and more efficient (used in Stable
Diffusion)
Score-based models – based on estimating probability gradients
Where You See Them
AI image generation (e.g., Midjourney)
Video generation
Audio synthesis
Drug discovery (molecule generation)
Intuition in One Line
A diffusion model learns how to turn pure noise into structured data by reversing a
gradual corruption process.
Encoder Decoder Models
Encoder–Decoder models are a type of neural network architecture used in generative AI
to transform one sequence into another.
In simple terms: They read input → understand it → generate a new output
Core idea
An encoder–decoder model has two main parts:
1. Encoder
Takes the input (text, audio, etc.)
Converts it into a meaningful internal representation (a vector)
2. Decoder
Uses that representation
Generates the output step-by-step
Key concepts behind it
These models rely on:
Neural Networks
Sequence-to-Sequence Learning
Transformers (modern versions)
How it works (example)
Task: Translate English → French
1. Encoder reads: “Hello, how are you?”
2. Converts it into a compressed meaning representation
3. Decoder generates: “Bonjour, comment ça va ?”
Simple analogy
Think of it like:
Encoder = listener (understands the sentence)
Decoder = speaker (rephrases it in another language)
Where are encoder–decoder models used?
They are widely used in:
Machine translation (Google Translate)
Text summarization
Chatbots and conversational AI
Speech-to-text and text-to-speech
Image captioning
Types of encoder–decoder models
1. RNN-based (older)
2. LSTM/GRU-based (improved memory)
3. Transformer-based (modern and powerful)
Example:
BERT (encoder-focused)
GPT (decoder-focused)
Important feature: Attention
Modern encoder–decoder models use:
Attention Mechanism
This allows the model to focus on important parts of the input while generating output.
In one line
Encoder–Decoder = understand input → generate output in a new form
# 2. Basic Concepts (Non-Technical)
Explain simply—avoid deep math:
What is a model
What is training data
What are tokens
Input → Processing → Output flow
Idea of probability (predicting next word/token)
3. Key Technologies Behind GenAI
Introduce at a high level:
Large Language Models (LLMs)
Transformer model
Neural networks
Multimodal AI (text + image + audio)
4. What Can GenAI Do?
Give practical use cases:
Content writing (blogs, emails)
Image generation
Coding help
Data summarization
Chatbots & virtual assistants
Business automation
✍ 5. Prompting Basics (Very Important)
What is a prompt
Good vs bad prompts
Techniques:
Clear instructions
Role prompting (“Act as a teacher…”)
Step-by-step prompting
Iteration (refining prompts)
6. Limitations of GenAI
Be honest about weaknesses:
Hallucinations (wrong answers)
Bias in outputs
Lack of real-time knowledge (sometimes)
Dependency on training data