Aditya V Jain | Foundations of GenAI
1 / 48
Speaker Notes:

Foundations of Generative AI

Mental Models and Effective Interaction

Your Facilitator

Aditya V Jain

Sr. Product Manager, Google

  • 18+ years in product leadership across Google, Amazon, Snapdeal and Shaadi.com
  • Guest Faculty, IIM Kozhikode — Teaching AI applications in business across MDP, EPGP, and PGP programs for 4 years

What I bring to this session:

I'm not an AI researcher — I'm a practitioner. I use these tools daily for real work, and I've taught hundreds of executives how to get practical value from them.


aditya1384@gmail.com | LinkedIn: linkedin.com/in/adityavjain

The Three Questions on Your Mind

Over the next three Sundays, we're tackling the three questions I hear most from senior professionals:

1. How do I actually get useful output?

You've tried ChatGPT, gotten mediocre results, and wondered what the fuss is about.
Addressed Today: Mental models + prompting fundamentals.

2. Can AI help with my real work?

Not just writing emails, but the substantive work that takes hours (research, analysis).
Session 2: Deep Research tools.

3. What does this mean for my job?

The strategic picture beyond the hype and fear.
Session 3: An assessment of where this is headed.

Before We Begin — Quick Setup

Three Things You Need for Today

Action: Take 2 minutes now to get Gemini ready.

Today's Roadmap

2:30 - 3:15 The foundations
Demystifying AI/ML/LLM terminology
3:15 - 3:55 Prompting that works
Why most prompts fail, R.C.T. framework, live demo
3:55 - 4:25 Break
4:25 - 5:15 Your problems, live solutions
Real challenges workshopped live
5:15 - 5:45 The productivity multipliers
Voice dictation and Gemini Gems
5:45 - 6:00 Wrap-up and homework
Part 1

What Are You Actually Using?

AI, ML, LLM, GenAI — Cutting Through the Jargon

How These Terms Relate

Artificial Intelligence (AI)

The broadest umbrella. Any machine that performs tasks requiring human-like intelligence. Chess computers to ChatGPT.

Machine Learning (ML)

A subset of AI. Systems that learn patterns from data rather than being explicitly programmed. Your spam filter is ML.

LLMs

Large Language Models. Trained on massive text to understand and generate language. This powers ChatGPT, Gemini, Claude.

Generative AI (GenAI)

A capability. AI that creates new content — text, images, audio. LLMs are generative AI for text.

Key Takeaway: What you're using when you open Gemini: An LLM with generative AI capabilities.

How LLMs Actually Work

1

Massive data collection

Trained on hundreds of billions of words from books, Wikipedia, websites, articles, code.

2

Learning to predict the next word

"The capital of France is..." → "Paris" | "To be or not to..." → "be"

This prediction task is repeated billions of times. Through this, the model learns grammar, facts, reasoning, writing styles.

3

Fine-tuning for helpfulness

Human trainers rate responses. Model learns what "helpful" looks like.

What This Means for How You Use AI

Implication 1: It's generating, not retrieving

There's no database the AI is searching. It's generating text that matches the pattern of what a correct answer would look like.

This is the root cause of "hallucinations" — when the AI confidently states something false.

Implication 2: It's trained on patterns, not truth

The model learned what sounds right, not what is right.

Practical Rule: Treat AI output like a first draft from a well-read but unreliable colleague.

The Capability Map

Where LLMs Excel ✓

  • Drafting and writing
    Emails, documents, summaries
  • Brainstorming
    Generating ideas, options, variations
  • Synthesis
    Combining information from multiple sources
  • Reformatting
    Bullet points to prose
  • Explaining concepts
    Breaking down complex topics

Where LLMs Fail ✗

  • Precise calculations
    Don't trust math
  • Factual accuracy
    Recent events, statistics, citations
  • Counting and logic puzzles
  • Your proprietary context
  • Tasks requiring 100% accuracy
Part 2

The Right Mental Model

How to Think About AI So You Actually Get Value From It

How NOT to Think About AI

It's like Google Search

Search retrieves existing documents. AI generates new text. Different mechanisms.

It knows everything

The AI's confident tone has nothing to do with accuracy. It states false info with same certainty as true.

It's magic that reads my mind

"Write me a good email" gives the AI nothing to work with. It doesn't know recipient, context, goal, or tone.

It's just fancy autocomplete

This undersells its synthesis and reasoning capabilities.

Think of It as a Brilliant New Employee

This employee:

Key Takeaway: You already know how to manage people like this. Same skills apply to AI.

Why Senior Managers Have an Advantage

Senior managers are better positioned to use AI effectively than junior staff.

Your Management Skills

  • Setting clear objectives
  • Providing context and background
  • Giving specific instructions
  • Reviewing work critically
  • Iterating based on output quality
=

AI Prompting Skills

  • AI needs to know what success looks like
  • AI knows nothing until you tell it
  • Vague direction = vague results
  • First drafts always need editing
  • "Try it this way instead"
Part 3

Prompting That Actually Works

From Vague to Valuable

Why Most Prompts Fail

What You Type

Prompt
Summarize this report

What's in Your Head

  • I need this for a CFO meeting
  • Focus on financial implications
  • Keep it under 500 words
  • The CFO hates jargon
  • Flag any risks mentioned

The gap between what you meant and what you typed is why AI output disappoints.

The R.C.T. Framework

R
Role

Who should the AI act as?

"Act as a senior business analyst..."
"You are a skeptical CFO..."

C
Context

What's the situation?

"I'm preparing for a board meeting..."
"I'm a VP of Operations planning capex..."

T
Task

What specifically do you want?

"Summarize in 5 bullet points..."
"Identify the top 5 risks..."

Even just including these three elements will dramatically improve your output.

The Secret Sauce — Constraints

Constraints — explicitly telling the AI what to avoid — force it out of generic patterns.

Effective Constraints:

  • Do NOT use buzzwords like 'synergy,' 'leverage,' or 'ecosystem'
  • Keep the response under 150 words
  • Avoid bullet points — write in flowing prose
  • Do not be salesy or use exclamation marks
  • Do not make up statistics or cite sources unless certain
  • Skip the preamble — start directly with the content

Live Demo: Dense Document Summarization

The Scenario:

You have a 50+ page RBI Monetary Policy Report on your desk. You know it's important — interest rates, inflation outlook, credit conditions — but it's dense, full of jargon, and you've been putting it off for weeks.

Sound familiar?

Demo Steps:

  • Start with a vague prompt — the way most people first try AI
  • Add context — tell Gemini who you are and what you need
  • Add constraints — specify format, length, what to focus on
  • Save it as a Gem — turn it into a reusable tool (end of session)

Demo Step 1: The Vague Prompt

Prompt
Summarize this report

Expected Output Problems:

  • Lists section headings rather than insights
  • Misses what actually matters to you
  • No sense of priority — everything treated equally

What's wrong: The AI doesn't know who you are, what you need, or what decisions depend on this.

Demo Step 2: Adding Context

Prompt
I'm the VP of Operations at a mid-sized manufacturing company. We're planning our capital expenditure for FY26-27 and I need to understand the macro environment. Summarize this RBI Monetary Policy Report focusing on: - Interest rate trajectory and what it means for borrowing costs - Inflation outlook and impact on input costs - Any signals about industrial credit availability - Risks flagged that could affect manufacturing sector I have a planning meeting with my CFO next week.

What changed: The AI now knows why you're reading this and what decisions depend on it.

Demo Step 3: Adding Constraints

Prompt
[Previous context, plus:] Format and constraints: - Start with a 3-sentence executive summary I can use verbatim in my presentation - Then provide 5-7 key points, each with the specific data/quote from the report - Highlight any numbers I should cite (rates, percentages, projections) - Flag anything that contradicts our current assumptions (we're assuming 7% inflation and stable rates) - Keep the whole thing under 500 words - Use plain language — my CFO hates jargon - End with "Questions to consider" for our planning discussion - Do NOT include generic economic background — assume I know the basics

Same Document. Dramatically Different Results.

Prompt Level Output Quality Time to Useful
"Summarize this" Generic, surface-level, requires heavy editing Saves almost no time
+ Context Focused on your needs, connects to decisions Good starting point
+ Constraints Presentation-ready, structured, specific Minutes vs. hours

Iteration Is Normal

Your first prompt is rarely your last.

Common follow-ups:

  • Make the executive summary shorter
  • Add more detail on the interest rate section
  • Rewrite this part in bullet points
  • The tone is too formal — make it conversational

The Mindset Shift

The AI should get it right the first time

I'm directing the AI toward what I need in 2-4 rounds

Break

30 Minutes

While on break, think about a problem you'd like to work on when we reconvene.

Part 4

Your Problems, Live Solutions

Crowdsourced Problem-Solving

What's Not Working for You?

Share with the room:

  • A task where AI gave you disappointing results
  • Something you've been meaning to try but haven't
  • A skeptical question about whether AI can actually help
  • A type of work you're curious whether AI could handle

Let's work through 2-3 real problems together, live.

Fallback Scenario: Vendor Evaluation

Situation: You have 3 vendor proposals (40+ pages each). Need comparison for steering committee.

Prompt
I'm evaluating 3 ERP vendor proposals for a mid-sized company. I need to present a comparison to our steering committee. Compare these proposals across: - Total cost of ownership (5-year view) - Implementation timeline and resources required from our side - Key differentiators and unique capabilities - Integration with our existing systems (SAP, Salesforce) - Risks and concerns I should raise Format as a comparison table plus a 1-paragraph recommendation with reasoning.

Fallback Scenario: Customer Feedback Analysis

Situation: 200 customer support tickets. Need to find patterns and recommend fixes.

Prompt
I'm analyzing 200 customer support tickets from the last month. I need to present findings to our product team. Analyze these tickets and provide: - Top 5 issue categories by frequency - For each category: specific examples, severity assessment, recommended fix - Any emerging issues that weren't in last month's analysis - Verbatim quotes I can use to illustrate each issue Format: Executive summary (3 sentences), then detailed breakdown by category.

Live Problem-Solving

Our Process:

  • Understand the task
    What are you actually trying to accomplish?
  • Identify the context gap
    What does AI not know that it needs to?
  • Build the prompt together
    Role, Context, Task, Constraints
  • Execute and observe
    See what works and what doesn't
  • Iterate
    Refine based on output
Part 5

The Productivity Multipliers

Voice Dictation + Gemini Gems

The Typing Trap

150
WPM
How fast you think
40
WPM
How fast you type
4x
Gap
Thinking vs Typing

When you type, you compress your thoughts. You leave out context, nuance, and detail because typing is slow.

Your prompts are impoverished versions of what you actually meant.

Voice Dictation: Capture Your Full Thinking

Tools like Wispr Flow let you speak your prompts instead of typing them.

Why it works:

When you speak, you naturally include more context, more nuance, more of what the AI needs to help you.

  • 4x Faster than typing
  • 72% Of characters via voice after 6 months
  • 90% Of dictation requires no edits
  • 175 WPM Dictation speed for developers
  • 270 Fortune 500 companies using Wispr Flow

Demo: Voice Dictation with Wispr Flow

Watch how messy, natural speech becomes a clean, context-rich prompt.

What to notice:

  • You include more detail when speaking
  • The AI has more to work with
  • The output is better because the input is richer

Switching to live demonstration...

Voice Dictation Alternatives

Tool Platform Notes
Wispr Flow Mac, Windows Best accuracy, context-aware, $10/month
Built-in Dictation Mac (Cmd+Cmd)
Windows (Win+H)
Free, less accurate but works
Google Voice Typing Google Docs Free, decent accuracy
Otter.ai Web, Mobile Good for longer dictation and meetings
Part 6

From Prompt to Reusable Tool

Creating Gemini Gems

Turn Your Best Prompts Into Permanent Tools

You don't want to write that complex prompt every time.

Gemini Gems solve this. A Gem is essentially a saved prompt with a name.

Think of it as creating a specialized assistant:

  • "Document Summarizer"
  • "Meeting Prep Assistant"
  • "Vendor Comparison Analyst"

Benefits

  • Consistency — Same quality every time
  • Speed — No prompt crafting, just input and go
  • Iteration — Refine the Gem over time
  • Sharing — Share Gems with your team

Live Demo: Creating a Gem

We're going to take the refined document summarization prompt and turn it into a reusable Gem.

Instructions
You are an expert document analyst helping busy executives extract actionable insights from dense reports. When I upload a document and describe my situation, create a focused summary that: - Starts with a 3-sentence executive summary I can use verbatim - Provides 5-7 key points with specific data/quotes from the document - Highlights numbers I should cite - Flags anything that contradicts stated assumptions - Keeps the whole thing under 500 words - Uses plain language — no jargon - Ends with "Questions to consider" I will provide: [DOCUMENT] and [MY ROLE/CONTEXT/WHAT DECISIONS DEPEND ON THIS]

Anatomy of a Gem

  • Name: Short, memorable (e.g., "Executive Document Summarizer")
  • Description: What it does (e.g., "Creates focused summaries of dense reports")
  • Instructions: Your full refined prompt — the Role, Context, Constraints
  • Conversation starters: (Optional) Example inputs to get started

Pro Tips:

  • Be explicit in instructions — the Gem inherits everything you write
  • Include constraints prominently
  • Use placeholders like [YOUR SITUATION] to remind yourself what input is needed

Gems You Could Create This Week

Gem Name What it Does
Meeting Prep Brief Takes company/purpose, generates research summary & questions
Email Tone Adjuster Rewrites drafts to be more diplomatic, direct, or concise
Weekly Update Writer Takes bullet points, generates polished update for leadership
Proposal Reviewer Analyzes proposals for risks, gaps, and missing info
Feedback Framer Takes raw feedback, reframes it constructively

What We Covered Today

The Foundations

  • AI/ML/LLM/GenAI relationship
  • How LLMs work (prediction, not retrieval)
  • Capability Map (Excel vs Fail)

The Mental Model

  • Brilliant new employee, zero context
  • Management skills = Prompting skills

Prompting That Works

  • R.C.T. Framework (Role, Context, Task)
  • Constraints are the secret sauce
  • Iteration is normal

Productivity Multipliers

  • Voice dictation (4x faster)
  • Gemini Gems (Reusable tools)

Your Homework

Three things to try before Session 2:
  • Dictate at least 5 prompts
    Use Wispr Flow, built-in dictation, or any voice tool. Notice whether you include more context.
  • Create at least one Gem
    Take a prompt that works well for you and save it as a Gem. Test and refine it.
  • Identify a research challenge
    Think of something you've been meaning to research (vendor comparison, market landscape). Bring it to Session 2.

Session 2: AI for Research & Analysis

January 11, 2026 | 2:30 - 6:00 PM

Topics

  • Gemini Deep Research: Autonomous research that compresses days into minutes
  • NotebookLM: Upload documents and interrogate them with citations
  • Live problem-solving: Bring a research challenge you're actually stuck on

Building on Today

  • We'll use the prompting skills you learned today
  • We'll create more Gems for research workflows
  • Your voice dictation practice will pay off for complex research prompts

Questions?

Aditya V Jain

aditya1384@gmail.com | +91 9986590180

Thank you — see you next Sunday!

Appendix: Document Options for Demo

  • RBI Monetary Policy Report: Default choice; everyone affected.
  • Union Budget Economic Survey: Strategic planning focus.
  • NITI Aayog Digital Transformation: Tech/IT focus.
  • IBEF Indian Economy Overview: Neutral, broad choice.
  • McKinsey Future of Work: If discussion turns to AI impact.

Appendix: Reusable Prompt Template

Template
I'm a [YOUR ROLE] at a [COMPANY TYPE]. I need to understand this [DOCUMENT TYPE] because [SPECIFIC DECISION OR MEETING]. Summarize this document focusing on: - [ASPECT 1 relevant to your decision] - [ASPECT 2 relevant to your decision] - [ASPECT 3 relevant to your decision] - Any signals or risks that could affect [YOUR SPECIFIC CONCERN] Context for my situation: [ANY ASSUMPTIONS OR CONSTRAINTS YOU'RE WORKING WITH] Format and constraints: - Start with a [X]-sentence executive summary I can use in [CONTEXT] - Provide [N] key points, each with specific data/quotes from the document - Highlight numbers I should cite - Flag anything that contradicts [YOUR CURRENT ASSUMPTIONS] - Keep it under [WORD COUNT] - Use plain language — avoid jargon - End with [WHAT YOU WANT AT THE END] - Do NOT include [WHAT TO SKIP]

Appendix: Alternative Demo Scenarios

Scenario When to Use
Meeting Prep Brief If room is Sales/BD heavy
Drafting a Difficult Email Universal fallback
Vendor Comparison If room is IT/Tech heavy
Talking Points for Leadership If room is senior leadership track

Appendix: R.C.T. Framework Reference

Reference
(R) ROLE: Act as [specific role with relevant experience] (C) CONTEXT: - I am [your role/situation] - I need this for [specific purpose/decision/meeting] - My audience is [who will see this] - Background: [any relevant context the AI needs] (T) TASK: [Specific deliverable] - Format: [bullets, prose, table, etc.] - Length: [word count or page limit] - Include: [specific elements required] CONSTRAINTS (Critical): - Do NOT [things to avoid] - Tone: [professional, casual, formal, etc.] - Skip: [what to leave out] - Assume: [what the audience already knows]