Gen AI Introductory Posts Series
Dr. Nisha Arora
I see many learners are overwhelmed by Gen AI related stuff. With my posts, I am explaining things in simple words one by one to educate a larger audience about the trending things.
Read my first post 📋📝, if that interest you👇
1. Generative AI: Introduction
A few years back, people were confused between ML, DL, and AI.
🤔 Now we have a new round of confusion and it goes like this: "Is ChatGPT the same as AI? Is LLM another word for Gen AI? Who even is Anthropic? What's Claude? Copilot that I see in Excel _ is it the one that's used by vibe coders?"
These terms are genuinely confusing because tech people (and the media) often use them loosely. 😇
Let's fix that 😎
AI (Artificial Intelligence) is the broad idea. Machines doing things that normally need human thinking. Planning a route, recognizing a face, filtering spam. That's AI.
Under AI, you have ML (Machine Learning), where machines learn from data instead of following hard-coded rules.
And under ML, you have DL (Deep Learning), which uses layered neural networks and handles things like image recognition or speech.
In short: DL sits inside ML, ML sits inside AI. Interested to know more about ML, my talk on Machine Learning Demystified can make it simple for you.
Now, Let's talk about generative AI... The trending thing!
Generative AI (Gen AI) is different. It generates new content such as text, images, audio, code, video. It doesn't just answer yes/no. It writes a paragraph, draws a picture, composes music.
Until a few years ago, this was not possible. That's the key shift. From AI that analyzes → to AI that creates.
More on how it works & Other related stuff, in next post!
Dr. Nisha Arora
Trainer | Author | Reviewer | Speaker ( Analytics, Data Science & AI )
📧 dr.aro...@gmail.com | 📍 Pune, India | 📞 +91-9654582757
Dr. Nisha Arora
Here's next post in that series 👇
🤔 AI, Gen AI, LLM, ChatGPT, Anthropic... are these all the same thing?
These terms are genuinely confusing because tech people (and the media) often use them loosely.
💬 ChatGPT, Claude, Gemini, Perplexity, NotebookLM, etc. are various products based on gen AI.
These interfaces you actually open and talk to.
There are many such interfaces and tools available for different tasks such as MidJourney or DALL·E for generating images, SlidesAI or Beautiful.ai for AI generated presentations, something else for AI generated videos etc.
Which of these have you used?
🤔 Now, let’s admit, there’s another layer of confusion due to ambiguity/similarity in names of products. At least I had this confusion, when I first heard about launch of various such products:
🙆♀️ Microsoft Copilot Vs GitHub Copilot
🙆♀️ Claude AI vs Claude Code
🙆♀️ NotebookLM _ Is it another Python notebook environment like Jupyter/Colab? Or is it LLM model name?
Wait! What’s LLM model?
We will discuss it very soon, stay tuned!🤝
These are genuine confusions which bothered me until I read about them and got clarity.
Then, LLM model’s name, parent company names, …yup, it’s too much confusion, if you have not read about those yet.
Write the question that bothers you about these terms & we will discuss!
Dr. Nisha Arora
Trainer | Author | Reviewer | Speaker ( Analytics, Data Science & AI )
📧 dr.aro...@gmail.com | 📍 Pune, India | 📞 +91-9654582757
Dr. Nisha Arora
Here's next post in that series 👇
💬 Let’s understand Gen AI Products!
As discussed in the previous post,
ChatGPT, Claude, Gemini are Gen AI products, you might have used some of those.
ChatGPT is made by a company called OpenAI. It's built on their LLM model (More on this later!). It's what most people encountered first, which is why "ChatGPT" became a default word for all AI chat tools like how people say "Google it" even when they're using Bing.
Claude is made
by Anthropic. Also, an LLM-based chat tool, known for handling long
documents well and being cautious in responses.
Gemini is Google's
version. Perplexity AI is developed by Perplexity, a startup.
So: ChatGPT ≠ all AI. It's one product, from one company, built on one model. There are several others.
🏢 OpenAI, Anthropic,
Google are the Companies/organizations building and running these models.
So now you understand some common
gen AI products and their parent companies.
Next, let’s talk about Microsoft Copilot and GitHub Copilot:
Microsoft Copilot is a productivity assistant built into apps like Word,
Excel, Outlook, and Windows, helping with writing, analysis, and everyday
tasks. While, GitHub Copilot is a coding assistant inside IDEs like VS
Code, focused on autocompleting, generating, and reviewing code to speed up
software development.
Before we discuss NotebookLM or Claude, we need to introduce the term LLM, keeping in mind that many of you might not have heard about deep learning at all.
Stay tuned!
Warm regards,
Dr. Nisha Arora
Dr. Nisha Arora
🧠 Let’s talk about LLM, the Engine Behind Most Gen AI You've Seen
*LLM stands for Large Language Model*. It's a very large mathematical system, specifically one trained on massive amounts of text to understand and generate language.
"Large" here means genuinely massive. Trained on billions of documents, internet, books, articles, websites, code.
As a result, LLM model that can answer questions, summarize reports, write emails, translate languages, explain complex topics, and even write code.
Your HR team using AI to draft job descriptions? Likely an LLM underneath. A finance professional summarizing a 50-page report in 2 minutes? LLM. A faculty member getting instant feedback on a lesson plan? LLM.
*Not all Gen AI is an LLM though*. Image generators like DALL·E or Midjourney are Gen AI, but they're not language models, they work with images, not text.
There are different types of LLM Models available. Are you familiar with some?
Dr. Nisha Arora
🗂️ You might have heard of Bert, GPT, Llama, Gemini, Claude Opus, and Claude Sonnet, etc.
Did you ever wonder, why so many different models and do they serve the same purpose?
Let me tell you: Not All LLMs Are the Same, Here's How to Tell Them Apart
In the last post, we cleared up the confusion between AI, Gen AI, LLMs, and all those company names. If you missed it, go read that first, this post builds on it.
Let's break down the main ways to classify LLMs. 👇
1️⃣ By Architecture — How Are They Built?
2️⃣ By Availability — Who Can Access It?
3️⃣ By Task — What Are They Built For?
4️⃣ By Modality — What Kind of Data Can It Handle?
By Architecture means based on the design philosophy of the model. There are three main types:
🔍 Autoencoding Models, ✍️ Autoregressive Models and 🔄 Seq2Seq Models
Wanna go little bit into what these 3 types of model do?
Dr. Nisha Arora
🧠 Let’s talk about LLM, the Engine Behind Most Gen AI You've Seen
"Large" here means genuinely massive. Trained on billions of documents, internet, books, articles, websites, code.
As a result, LLM model that can answer questions, summarize reports, write emails, translate languages, explain complex topics, and even write code.
Your HR team using AI to draft job descriptions? Likely an LLM underneath. A finance professional summarizing a 50-page report in 2 minutes? LLM. A faculty member getting instant feedback on a lesson plan? LLM.
There are different types of LLM Models available. Are you familiar with some?
Dr. Nisha Arora
Trainer | Author | Reviewer | Speaker ( Analytics, Data Science & AI )
📧 dr.aro...@gmail.com | 📍 Pune, India | 📞 +91-9654582757
Dr. Nisha Arora
🗂️ Not All LLMs Are the Same, Here's How to Tell Them Apart?
In the last post, we cleared up the confusion between AI, Gen AI, LLMs, and all those company names. If you missed it, go read that first, this post builds on it.
Let's break down the main ways to classify LLMs. 👇
1️⃣ By Architecture — How Are They Built?
2️⃣ By Availability — Who Can Access It?
3️⃣ By Task — What Are They Built For?
4️⃣ By Modality — What Kind of Data Can It Handle?
By Architecture means based on the design philosophy of the model. There are three main types:
🔍 Autoencoding Models, ✍️ Autoregressive Models and 🔄 Seq2Seq Models
Dr. Nisha Arora
Trainer | Author | Reviewer | Speaker ( Analytics, Data Science & AI )
📧 dr.aro...@gmail.com | 📍 Pune, India | 📞 +91-9654582757
Dr. Nisha Arora
Dear All,
Let's Understand Autoencoders 👇
Whether you’re working with images, text, or other forms of data, autoencoders are a versatile tool for tasks like dimensionality reduction, noise reduction, and anomaly detection. At a high level, autoencoders are a type of artificial neural network used primarily for unsupervised learning.
Autoencoder’s main goal is to learn a compressed, or “encoded,” representation of data and then reconstruct the original data from that compressed version.
1. Encoder: The encoder takes the input data (e.g., an image, a sound clip, or a text document) and compresses it into a lower-dimensional representation, known as the latent space or encoded representation.
2. Decoder: The decoder takes the compressed representation and tries to reconstruct the original input. The goal is to minimize the difference between the original and reconstructed data.
If we talk about autoencoders for text data, these models are trained to read and understand text really well. They look at a sentence from both directions, left to right AND right to left, to understand full context.
They're trained by randomly hiding words in a sentence and learning to predict the missing word using everything around it.
Use-cases include searching documents, or sentiment analysis ("Is this customer review positive or negative?"). Useful for a legal team running contract analysis, an HR system screening resumes or a bank flagging suspicious transaction descriptions.
Example: BERT model (by Google)
Types/applications of autoencoders 🤝
• Standard Autoencoding Language Model: Learns to compress and reconstruct input data
Applications: Text generation, sentiment analysis, summarization
• Denoising Autoencoding Language Model: Learns to remove noise from input data
Applications: Noise removal, overfitting prevention, image generation
• Variational Autoencoding Language Model (VAE): Learns to encode important features from input data
Applications: Image generation, anomaly detection, representation learning, natural language processing
• Sparse Autoencoding Language Model: Learns compact and efficient representations
Applications: Data compression, denoising, feature learning
• Contractive Autoencoding Language Model: Learns robust and stable representations
Applications: Feature learning, dimensionality reduction, denoising, data generation
For image data, image denoising (enhances overall image
clarity, making pictures sharper and more professional) example can be
explained by this visual
In the next post, we will talk about Autoregressive models. Stay Tuned!
Dr. Nisha Arora
Let's Understand Autoencoders 👇
Whether you’re working with images, text, or other forms of data, autoencoders are a versatile tool for tasks like dimensionality reduction, noise reduction, and anomaly detection. At a high level, autoencoders are a type of artificial neural network used primarily for unsupervised learning.
Autoencoder’s main goal is to learn a compressed, or “encoded,” representation of data and then reconstruct the original data from that compressed version.
1. Encoder: The encoder takes the input data (e.g., an image, a sound clip, or a text document) and compresses it into a lower-dimensional representation, known as the latent space or encoded representation.
2. Decoder: The decoder takes the compressed representation and tries to reconstruct the original input. The goal is to minimize the difference between the original and reconstructed data.
If we talk about autoencoders for text data, these models are trained to read and understand text really well. They look at a sentence from both directions, left to right AND right to left, to understand full context.
They're trained by randomly hiding words in a sentence and learning to predict the missing word using everything around it.
Use-cases include searching documents, or sentiment analysis ("Is this customer review positive or negative?"). Useful for a legal team running contract analysis, an HR system screening resumes or a bank flagging suspicious transaction descriptions.
Example: BERT model (by Google)
Types/applications of autoencoders 🤝
• Standard Autoencoding Language Model: Learns to compress and reconstruct input data
Applications: Text generation, sentiment analysis, summarization
• Denoising Autoencoding Language Model: Learns to remove noise from input data
Applications: Noise removal, overfitting prevention, image generation
• Variational Autoencoding Language Model (VAE): Learns to encode important features from input data
Applications: Image generation, anomaly detection, representation learning, natural language processing
• Sparse Autoencoding Language Model: Learns compact and efficient representations
Applications: Data compression, denoising, feature learning
• Contractive Autoencoding Language Model: Learns robust and stable representations
Applications: Feature learning, dimensionality reduction, denoising, data generation
Dr. Nisha Arora
Trainer | Author | Reviewer | Speaker ( Analytics, Data Science & AI )
📧 dr.aro...@gmail.com | 📍 Pune, India | 📞 +91-9654582757
Dr. Nisha Arora
Let's Understand Autoregressive Models
Autoregressive Models are the LLM
models that generates text. They predict what comes next, one word at a
time, based on everything written before. These models are trained by learning to
predict the next word in billions of sentences, over and over, until they
get really good at it.
It predicts the next word in a sequence based on the preceding words. Unlike the autoencoding task, where the model can see context from both sides of the missing word, the autoregressive model can only consider context from words preceding the missing word. It predicts the missing word based on the words that come before it in the sequence.
Use cases include writing emails, answering questions, and generating code. For example, a marketing team drafting campaign copy, a faculty member generating quiz questions from a chapter or a developer asking it to write a function. Large language models are either autoencoding, autoregressive or a combination of both.
Dr. Nisha Arora
Trainer | Author | Reviewer | Speaker ( Analytics, Data Science & AI )
📧 dr.aro...@gmail.com | 📍 Pune, India | 📞 +91-9654582757
Dr. Nisha Arora
Let's talk about Seq2Seq (Sequence-to-Sequence) Models:
Seq2Seq (Sequence-to-Sequence) is also encoder-decoder architectures (like autoencoder), but they serve entirely different purposes and handle data differently.
The primary difference is their goal: Seq2Seq translates one sequence into a different sequence, while an Autoencoder reconstructs its exact input data to learn efficient representations.
Basically, in Seq2Seq you give it something and it gives you something transformed back. It has two parts: one that reads the input, one that generates the output. Designed for transformation tasks. These models exclusively handles sequential data (text, audio, time-series)
Use cases include Language translation, converting speech to text, reformatting data.
Examples: T5, BART, older Google Translate
Next, we will talk about classification of LLM Models based on availability.
Dr. Nisha Arora
Trainer | Author | Reviewer | Speaker ( Analytics, Data Science & AI )
📧 dr.aro...@gmail.com | 📍 Pune, India | 📞 +91-9654582757