WE HAVE ALWAYS RESISTED NEW TECH

Anti-electricity
cartoon from
1900s

AI is the new electricity. Has the power to

change every industry fundamentally.

And, of course, needs to be handled with

care.
 SIMPLIFYING THE LANDSCAPE Data Science

AI is the simulation of human intelligence in machines, enabling them to

perform tasks like reasoning, learning, problem-solving, and decision-
making.

ML is a subset of AI that enables machines to learn patterns from data

and improve their performance on tasks without explicit programming.

Neural networks are computing systems inspired by the human

brain, consisting of interconnected layers of nodes (neurons) that
process and learn from data.

Deep learning is a subset of ML, NN that uses multi-layered

neural networks to automatically extract patterns from large
datasets, excelling in tasks like image recognition and natural
language processing.

Generative AI refers to AI models that create new content—

such as text, images, and music—by learning patterns from
existing data. Examples include ChatGPT, Perplexity, Gemini
and DALL·E.
Source
LET’S UNDERSTAND AI IN SIMPLE TERMS
AND DRAW SOME PARALLELS WITH OUR REAL WORLD
 A SIMPLIFIED ANALOGY

World Data Model Decision

Machine learning is the science of

creating a model of how the world works.

Machine Learning

Artificial Intelligence

Real World Analogy

We create a mental model for how the world works and how people are. So, we may interact with a person for a few
times (data) and create an opinion about him/ her (model)
Now going forward, we use that model to make our decisions or take actions. E.g. the person can't be trusted, or the
person gets angry when questioned, or the person takes responsibility well etc.
CLASSICAL PROGRAMING VS MACHINE LEARNING
 HOW GOOD IS A MODEL?
Data= Information + Noise
In July it rains 80% of the days in Delhi is an Information that can be used for prediction of rain this year
It rained 15 cm on 6th July in Delhi last year is a noise which can't be used for predicting the rain on 6th of July this year.
Data contains both and is often not very easily distinguishable.
Model

Underfitting Good Fit Overfitting

Information Doesn’t Capture Captures well Captures well
Noise Doesn’t Capture Doesn’t Capture Captures
UNDERSTANDING AI WITH A HUMAN
TOUCH (LITERALLY )
LET’S DRAW SOME PARALLELS WITH LIFECYCLE OF A KID
 UNSUPERVISED LEARNING

A world of unlabelled data

And trying to understand

the patterns
 UNSUPERVISED LEARNING

A world of unlabelled data

And trying to understand

the patterns

• Google News – Topic Clustering

• Customer Segmentation for Marketing
• Spotify – Music Genre & Mood
Clustering
• LinkedIn – Connection recommendation

* Examples for indicative purposes. Real world implementations may be way more nuanced
and often a combination of many techniques.
 SUPERVISED LEARNING

Welcome to the world of

labelled data where things
have names and they are
taught

• Gmail Spam Filter

• Amazon’s Product Recommendation
• Walmart’s Demand Forecasting

* Examples for indicative purposes. Real world implementations may be way more nuanced
and often a combination of many techniques.
 DEEP LEARNING/ NEURAL NETWORK
Example: Building the “Dog” Mental Model

As children develop, they start to understand Layer 1: Child may notice very basic features like edges or basic
more complex relationships between things. For shapes—like recognizing a circular head or a long body. In a Deep
example, they learn that a dog is not just a four- Neural Network, the first layers only look for the simplest things
legged animal but also a friend that can play,
bark, and respond to commands. The child starts Layer 2: As the child gets better, they start seeing ears or tail shape,
recognizing not just the external appearance but and can even tell if it’s furry or smooth. The child begins to look for
also deeper behavioural patterns. more complex things, like the combination of shapes—ears, tails, and
snouts

Layer 3: They might even notice barking or wagging tails that make
the dog distinct. Combination of the dog’s features—ears, tail, fur
type, and behaviour—that all together tell the system, “This is a dog.”

Layer 4: Eventually, after all the previous layers work together, the
child can instantly recognize that a particular animal is a dog and
even tell if it’s a golden retriever or a bulldog just by looking at its
general characteristics.
 REINFORCEMENT LEARNING
My parents get upset when I
don’t get good marks

A child learns many skills through trial and Learning from this short
error. For instance, when they are learning book gets me more marks
to ride a bicycle, they may fall a few times and leaves me a lot of time
to play
but gradually learn to balance and pedal
without falling.

Over time, they adjust their actions based

on rewards (fun, success) and penalties
(falling, frustration). People like me more when I
crack jokes

• Google DeepMind’s AlphaGo Am I genuinely funny?

• Tesla’s Autopilot Should I do it all the time?
• Uber – Traffic and Surge Pricing Optimization

* Examples for indicative purposes. Real world implementations may be way more nuanced
and often a combination of many techniques.
 GENERATIVE ADVERSARIAL NETWORKS

Every artist or creators need some

adversaries and critiques in life.

A generator creates new data, while a

discriminator tries to distinguish real from
fake and provide feedback.

• Deepfake Videos getting more realistic with time

• Google & Adobe use GANs to sharpen blurry
images
• AI-generated music

* Examples for indicative purposes. Real world implementations may be way more nuanced
and often a combination of many techniques.
 GEN AI

Once you have learnt so much, is

it time to imagine and create
something unique or original?
 GEN AI- WHAT MAKES IT SPECIAL/ DIFFERENT
CLASSICAL AL /
GENERATIVE AI
DISCRIMINATIVE AL

Al models which discriminates data points e.g. Al models which generates data points e.g.

• Create email
• Classify email as spam or not spam • Generate customer data
• Predict customer churn • Generate summary of document
• Segment customers • Extract entities from document

Requires niche skills, high entry barrier Lowered barriers of adoption

Discriminative Al, describes type of ML models that focus on
learning the conditional probability distribution of the output
given the input.

Emergence of GAl does not mean Classical/Discriminative Al is not required, they co-exist,
we need both
 SO WHERE ARE WE?

Prompt to
ChatGPT

Create an image
comparing
Artificial narrow
intelligence,
Artificial general
intelligence and
Generative Artificial super
AI intelligence
 GEN AI- UNDERSTANDING LARGE FOUNDATION MODELS
• Large foundation models
(LFMs) power the
experience around Gen AI
e.g. GPT 3.5, GPT 4,
Gemini 2, Sonnet 3.5, DALL-
E, Deep Seek, Mid Journey
etc
• Large Language Models:
LLMs are a type of LFMs
which focusses solely on
language related tasks
• These models are trained on
petabytes of data, on
specialized infrastructure
making them available
/suitable for immediate use-
cases across different
industries
Source
HOW GEN AI WORKS
AT THE END OF THE DAY IT’S NOTHING BUT ARITHMETIC AT HUGE SCALE
HALLUCINATION IS, THEREFORE, NOT A BUG BUT A FEATURE

But AI is getting better with

information and sarcasm
 LET’S
UNDERSTAND
IMAGE
RECOGNITION
THROUGH
SIMPLE
VISUALIZATION
 Computing Internet Deep Learning
INNOVATIONS (1950s) (1970s) (2012)

DON’T BECOME
TECHNOLOGY
REVOLUTION
UNTIL THEY ARE
DEMOCRATIZED PC Web browser Gen AI
AND REACH (1980s) (1990s) (2023)

OUR DAILY
LIVES
 SO SHOULD WE COMPETE WITH AI?
The answers are often hidden in history …

Scene 1 (1985)
Garry Kasparov played chess games with 32 best chess
playing computers and easily defeated all of them.

Scene 2 (1996+)
He played 6 matches against IBM’s super computer Deep
Blue. 2nd match was played in 1997 and won by Deep
Blue and was the first defeat of a reigning world chess
champion under tournament conditions

Scene 3 (2005+)
A chess tournament was organised where each team was
a group of humans & computers. If Garry had the best
supercomputer on his side, will it make an invincible
team?
LinkedIn post
WHY IS IT IMPORTANT
AND WHY YOU SHOULD CARE
 COMPETITION IS GETTING INTENSE
TEXT/ CODE GENERATION ADVANCED REASONING

IMAGE GENERATION VIDEO GENERATION

 WHAT SHOULD YOU DO?

Use AI Smartly Stay Updated on Learn AI Skills & Future-Proof

• Automate tasks with ChatGPT, AI Trends Apply Them Your Career
Copilot, Gemini Grammarly
• Follow leaders like Andrew
AI, and Notion AI. • Take AI courses to deepen your • Understand what AI can and
Ng, Yann LeCun, and Fei-Fei
• Create stunning visuals using understanding cannot do to stay relevant.
Li.
MidJourney, Stable diffusion & • Experiment with AI-powered • Develop critical thinking,
• Subscribe to
DALL·E. analytics, automation, and adaptability, and AI-driven
DeepLearning.AI, MIT Tech
• Leverage AI in Excel, decision-making tools. decision-making.
Review, and AI-focused
PowerPoint, and Outlook for • Learn prompt engineering to • Be a leader in AI adoption,
podcasts.
efficiency. get better AI-generated outputs. because AI won’t replace
• Engage in AI communities on
• Try Perplexity for research For those who are technical, try you—but someone using AI
LinkedIn, Twitter
• Use DeepSeek, Grok 3 for coding again might!
deep and complex problem
THE CURRENT TREND & WHAT TO
EXPECT
THE FUTURE OF JOBS?
The World Economic Forum (WEF) projects that AI will replace 85
million jobs by 2025, while creating 97 million new jobs

Source 1 https://www.weforum.org/stories/2023/05/jobs-lost-created-ai-gpt/
Salesforce News Salesforce announcement
 EVERY JOB WILL NEED TO BE RE-IMAGINED
• Healthcare: AI-driven drug discovery, robotic surgeries, predictive diagnostics, AI wearables, vaccine creation
Making healthcare better: https://www.youtube.com/watch?v=t2G6u8GgUYE
Drug discovery: https://www.youtube.com/watch?v=lc560cKLtkI
• Finance: AI for fraud detection, algorithmic trading, and hyper-personalized banking.
• Retail & E-commerce: AI-powered recommendations, demand forecasting, and cashier-less stores.
• Manufacturing: Smart factories, predictive maintenance, and supply chain automation.
• Agriculture: AI-powered crop monitoring, yield prediction, and autonomous farming.
• Government & Policy: AI-driven public service automation, smart cities, and governance analytics.
• Education Sector: Personalized learning, study planning, assistance through conversational AI, regional languages

AI is expected to contribute $15.7 trillion to the global economy by 2030 (PwC).

Link Link Link

Link Link Link
START WITH PROMPT ENGINEERING

Source: https://cdn.analyticsvidhya.com/wp-
content/uploads/2024/10/Techniques_FOllm0q.webp
SOME HANDY REFERENCES
SOME HANDY REFERENCES
Here are 10 must-follow YouTube channels for learning AI, no matter where
you are on your journey:

1) Andrej Karpathy – Deep yet accessible lectures on deep learning,

LLMs, and an intro course on neural networks. https://lnkd.in/dVA5bGyZ
2) 3Blue1Brown – Stunning visualizations that make abstract
mathematical concepts intuitive. https://lnkd.in/dQsiH2Xu
3) Lex Fridman – In-depth conversations with AI leaders, offering a
broader perspective on the field. https://lnkd.in/dRJ6DVbE
4) Machine Learning Street Talk – Technical deep dives and
discussions with top AI researchers. https://lnkd.in/dNceZibz
5) StatQuest with Joshua Starmer PhD – Beginner-friendly explainers
on machine learning and statistics. https://lnkd.in/dTjRCxYP
6) Serrano Academy (Luis Serrano) – Clear and accessible content on
ML, deep learning, and AI advancements https://https://lnkd.in/d8fhq7Nq
7) Jeremy Howard – Practical deep learning courses and AI-powered
web app tutorials. https://lnkd.in/dkwh53ss
8) Hamel Husain – Hands-on lessons in LLMs, RAG, fine-tuning, and AI
evaluations. https://lnkd.in/dj8EEzPm
9) Jason Liu – Expert-led lectures on RAG and AI freelancing tips for
ML developers. https://lnkd.in/dbCCXZhb
10) Dave Ebbelaar – Practical guides on building AI systems and real-
world applications. https://lnkd.in/dThXHtpb
 LAYERS OF AUGMENTATION

Your physical
twin-
Your digital
twin- Humanoid
Humanoid robots

Your mental robots

Humanoid robots

twin- AI agents
AI agents

AI chats
AI chats

My LinkedIn post
 UNDERSTANDING AI AGENTS

How do we generally do any complex

work?
• Thinking & reflection
• Planning
• Research
• References to sources and tools
• Collaboration with experts
• Do it
• Review and critique it
• Repeat the relevant steps

That’s what AI Agents can do

The question is do we need one genius person or a team of specialists to solve it?
Or do we need both?
 LARGE QUANTITATIVE MODELS (LQM)
Core Applications
Used primarily in fields that require deep quantitative
reasoning. The industries that are quantitatively and
not language driven:
• Biopharma
• Chemicals
• Energy
• Financial services
• Manufacturing
These are multi trillion dollar industries and play a
crucial role in the progress and sustainability of our
world. The models are trained on

So while language model may help in researching a

Cancer or Alzheimer’s symptoms and summarizing
the scientific literature on the subject but not in
creating the drug that cures it.
RESPONSIBLE & TRUSTWORTHY AI
YOU CAN CONNECT WITH ME HERE
APPENDIX
 PROMPT EXAMPLES
1. Foundational Techniques 2. Advanced Logical and 3. Adaptive Techniques
• Instruction-Based: “List three benefits of Structured Techniques • Reflection: “Here is your response. How
exercising daily.” • Clarification: “What specific details can it be made clearer?”
• Few-Shot: “A cat says ‘meow.’ should I include in the report?” • Progressive: “Describe machine learning
A dog says ‘woof.’ • Error-Guided: “2 + 2 = 5. Is this in one sentence. Now explain it in more
A cow says ___?” correct? Explain why.” detail.”
• Zero-Shot: “Write a poem about the • Self-Consistency: “Solve this problem • Contrastive: “Compare the advantages
stars.” using two different methods and compare of remote work vs. office work.”
• Chain of Thought: “If a train travels 60the answers.” • Guided: “Explain photosynthesis in this
km in 1 hour, how far will it travel in 3 • Chunking-Based: “Summarize this book format: Definition, Process, Importance.”
hours? Show your steps.” chapter by chapter.” • Hypothetical: “What if gravity
• Tree of Thoughts: “Suggest three suddenly disappeared? How would the
different business strategies to increase world change?”
revenue and evaluate their pros and cons.” • Meta: “Explain how you structured this
• Role-Based: “You are a detective. Solve response.”
this mystery based on the given clues.”
• Person-Based: “Explain AI like a
Shakespearean poet.”
 WHAT IS DATA SCIENCE
Data Science is an interdisciplinary field that combines statistics, computer science, machine learning (ML), and
domain expertise to extract insights from structured and unstructured data. It involves data collection, cleaning,
analysis, visualization, and predictive modeling to support decision-making.

How Data Science Fits into the AI/ML Ecosystem

Data Science is the first step in building AI/ML models, as it prepares the data and finds meaningful patterns.
AI & ML rely on high-quality data to train models effectively.
ML & AI as Data Science Tools
Machine Learning algorithms help automate insights in Data Science. AI models enhance predictive analytics,
recommendation engines, fraud detection, etc.
Deep Learning’s Role in Data Science
Deep Learning (DL) techniques, like Neural Networks, are used in image recognition, speech processing,
and NLP (Natural Language Processing) within Data Science applications.
Generative AI in Data Science
Gen AI models, such as GPT, DALL·E, Midjourney, help create synthetic data, automate text summarization,
and generate insights faster.
 HR & TALENT MANAGEMENT
• Hiring & Talent Acquisition: AI-based resume screening, chatbot interviews, skill-matching algorithms.
• Employee Experience: AI-driven training recommendations, performance analytics, and HR automation.
• Skill Gaps & Continuous Learning: AI-personalized learning paths and adaptive training modules.
• Future Workforce Trends: Rise of the gig economy and AI-assisted freelancers.

Link Link Link

Link Link Link
 FOR YOUR OWN READ

https://en.wikipedia.org/wiki/Attention_Is_All_You_Need

Full Paper
https://proceedings.neurips.cc/paper_files/paper/2017/
file/3f5ee243547dee91fbd053c1c4a845aa-Paper.pdf

Attention in transformers