+

Getting Started
With AI
WITH
TAYLO
June 6, 2024 MALM R
SHEIM
ER

Waterfall question: Where are you

joining us from today?
BROUGHT TO YOU BY

+
👋 Why I’m here

● COO & Head of Strategy at Section – leading our

internal AI usage and building our first AI-powered
learning experience

● 10+ years partnering with thought leaders like Scott

Galloway to extract extensible frameworks and
insights for others

● Former consultant for CMOs and Chief Digital Officers

(Nike, P&G, Unilever, etc.)

● Used building a startup as my own personal MBA –

and now doing the same with AI
Poll: How often do you use Gen AI?

A B C D

Everyday! Every week I’ve tried it Not at all

a few times
 Why you’re here
And want to
get here!

INTEGRATING AI
INTO MOST
WORKFLOWS

ADVANCED
PROMPTING
WOW!
You’re here
UPLOADING
FILES

PROVIDE
CURIOUS NOOB CONTEXT

PROMPTING
BASICS
WAIT, WHAT??
We’re at the start of a big change in
how we interact with computers.

2023

2007

1998

1984

1974

1946
What makes these new
models special?

● It feels like they understand us

● They can carry a natural conversation

● They remember what we’re talking about

● They are surprisingly adept at “thinking” tasks

● They seem to have personality and emotion

What is an LLM?

TEXT TEXT
(PROMPT) (RESPONSE)

(LLM)

A computer interaction that feels like a natural conversation.

LLMs just reflect our world back at us.
They are all about context.

ROYALTY CHESS POKER DRAG ROCK N ROLL

You have a new (very smart) teammate
US IQ Score by Percentile
120-130: IQ range for AI

2nd 16th 50th 84th 98th

55 70 85 100 115 130 145

Source: Google Bard, ChatGPT, Claude by Anthropic.

Imagine a
teammate who…
● Knows all the Internet’s information
● Learns incredibly quickly
● Is highly logical, data-driven, and
unemotional when making decisions
● Is analytical + highly creative
● Is infinitely patient and eager to please
● Doesn’t get tired or bored
● Can move across domains,
functions, or disciplines seamlessly
Your teammate
also has flaws

Imagine a teammate who…

● Occasionally makes surprising mistakes
● “Hallucinates” or makes up answers when
they see patterns that don’t make sense to
humans
● Is over confident – it’s tough to tell when
they’re making up an answer
● Biased – just like other humans
● Performance varies day-to-day
Early adopters get time back
How much time are you saving using AI?
April 2024, n=1,308 respondents

No time saved

30%
15%
20%+ time saved
20%

of respondents say they’d

feel “very disappointed”
1-5% time saved if they could no longer
17%
use AI

10-20% time saved 5-10% time saved

22% 26%

Source: Section AI Readiness survey, April 2024.

An AI class will emerge
THIS IS
STARTINYOU!
EARLY F G
THE WIN OR
.

THE AI GAP WILL BE

MORE DRAMATIC
THAN DIGITAL
DIVIDE

NOT USING AI USING AI

 STEP 1 IN AI MASTERY

Set up your “AI workspace”

3 steps to set 1. Choose your LLM
up your 2. Decide how you’ll handle sensitive
workspace information
3. Set up custom instructions
QUICK POLL

Q. Has your company given you access to an

LLM through work? If so, which one?
Consumer-ready generative AI models

THE MOST CAPABLE THE ECOSYSTEMS

OpenAI Anthropic Google Microsoft

ChatGPT4 Claude Gemini CoPilot

THE NICHE PLAYERS

Perplexity Quora Poe Inflection Pi Open Source

(For search) (For variety) (For therapy) (It’s complicated)
I’d start with GPT and Claude

ChatGPT-4/o Claude by Anthropic

Now free! (premium version $20/month) Start free

● Relatively easy to on board ● Very easy to use

● Includes Internet search, data ● Better analysis of PDFs and text
analysis, and image recognition documents
● Best for structured, analytical ● Best for thought partner work,
thinking; reviewing slides; giving you feedback, text analysis,
research, fact checking playing different personas
Three options to avoid the model training on
your sensitive data

1 2 3

Your company’s ChatGPT for ChatGPT Plus

secure instance Teams with data controls

Does not train the model on Does not train the model on Ability to turn off training the
your data your data model on your data

$25/month $20/month
(min. 2 seats)
 STEP 2 IN AI MASTERY

Prompt with purpose

How to construct a great prompt
Prompt Building Blocks
You can build effective prompts by combining 2-6 of the
building blocks below. Prompting Tips

● Not all use cases require all

TASK six building blocks
ROLE BOUNDARIES
DESCRIPTION
● Parameters of your desired
output determine which
blocks you need

● Don’t overthink your initial

prompt
SPECIFIC
CONTEXT REASONING
REQUIREMENTS
Basic (but good) Prompts
Task
Context
Description

EXAMPLE #1
USE CASE
“Generate a catchy tagline for a new sports drink that
Simple, straight forward tasks with emphasizes its refreshing quality.”
almost no requirements

EXAMPLE #2

“Create an email subject line

for a 30% off seasonal sale campaign.”
STARTER TEMPLATE

“[task description] for [context]”

EXAMPLE #3

“Develop a meal plan for a vegetarian family that

accounts for a tomato allergy.”
Focused/Advanced Prompts
Task Specific
Context Role Requirements
Description

EXAMPLE #1
USE CASE
"As a project manager, create a risk assessment
More specific tasks where the for the upcoming software deployment.
output would vary depending on the
role AI is holding, and have specific Please adhere to the SWOT analysis format.”
requirements like format, style,
scope, etc.

EXAMPLE #2

STARTER TEMPLATE “As a content marketer specializing in digital marketing,

write the introduction for a blog post about the importance of
"As a [Role], [Task Description] for
[Context]. Adhere to [List Specific social media marketing in today's business landscape for an
Requirements]." audience of small business owners. Adhere to an informative
but engaging tone, and keep it between 5-6 sentences”
Comprehensive Prompts TASK
DESCRIPTION
CONTEXT ROLE
SPECIFIC
REQUIREMENTS
BOUNDARIES REASONING

EXAMPLE
USE CASE

Outputs with a complicated set of “As a coach roleplaying as my manager, respond and react
specifics, where fact checking and realistically based on my arguments for increasing my
accuracy are crucial. compensation based on my performance.

Please adhere to just roleplaying and only provide me

constructive feedback on how I did at the end. Focus on my
STARTER TEMPLATE negotiation tactics, responses, and tone. Be thoughtful,
"As a [Role], [Task Description] for constructive and blunt.
[Context]. Please adhere to [List
Specific Requirements]. Avoid Avoid departing from a calm professional tone. Explain your
[Boundaries]. Explain your thinking behind your responses and feedback.”
reasoning for [xx]?"
DEMO

Plan an offsite in Chicago

 The final polishing will come from you
AI has its superpowers, but so do humans

FINAL POLISHING CHECKLIST

❏ Additional Context: Infuse your deeper, more nuanced understanding of the context
into the output. Ensure content is appropriate & tailored for your particular audience.

❏ Quality Assurance: Double-check for errors, credibility, and accuracy. Look for
hallucinations, cut out irrelevant or unnecessary content, and ensure ethical
compliance.

❏ Format for Use: Make final edits to output so that the output is usable and applicable

❏ Cultural Sensitivity: Give a critical review for cultural and emotional sensitivity – check
for language or ideas that may be considered offensive or exclusionary
 STEP 3 IN AI MASTERY

Audit your workflows

for tasks AI can do
AI can augment two types of tasks – high
manpower and high brain power

💪 🧠
Manual Muscle Mental Muscle

Highly important
Highly repetitive
tasks that require
tasks that you perform
high quality thinking
consistently and follow
(and can benefit from
a similar pattern
a second opinion)
How to do this: Review your to-do list
Taylor’s to-do list 󰙣 = Manual Muscle

🧠
= Mental Muscle

��🏽
��🏽
��🏽
��🏽
��
��🏽
��🏽
��
��🏽
��
��
��
��🏽
 STEP 4 IN AI MASTERY

Map your workflows

to AI’s superpowers
AI has three personas to increase your
productivity

ASSISTANT STRATEGIST CREATOR

Cut out drudge Be your thought Build your first
work partner draft
Persona #1: The assistant

ASSISTANT USE CASES

ASSISTANT ● Summarize technical or academic research papers
● Review contracts or legal contracts
● Translate documents into various languages
Cut out
● Summarize key points of discussion and action items
drudge work
● Describe what this code does
● Analyze survey responses
● Digitize your handwritten notes
Demo: Contract review

75+ faculty contracts at Section

Contract template has changed 3x over the

last 3 years (and some faculty negotiate)

Manual, time-consuming process to review

Time savings: 80%

● 5 hours to review manually
● 1 hour to write prompt and review
outputs from ChatGPT for all contracts
Persona #2: The strategist

STRATEGIST USE CASES

STRATEGIST ● Critique your sales pitch
● Role play a conversation with a client or boss
STRATEGIST
● Critique your presentation and talking points
Be Be
youryour
thought ● Prioritize and recommend marketing channels
partner
thought partner ● Review your product roadmap
● Refine your resume or cover letter
Case study: AI as board member

“I am the CEO of Section. This is the board

meeting deck. Pretend to be a board member
and tell me how you think a board member
will react or what questions they might ask.”
AI went toe-to-toe with our board - April 2024
Great insights from trusted group.
Human Board A- 100% But took hours to set up, run, & follow up.

Appropriately ambitious. Macro market understanding.

Claude A- 91% Third-order thinking.

Some tailored advice and pushback, some generic.

ChatGPT 4 B+ 89% Thorough questions. Most points of view offered.

Mostly summarized and reflected our deck.

Gemini C+ 78% Limited pushback, insights, or advice.

Mostly summarized and reflected our deck.

Copilot C- 72% General feedback only focused on quantitative data.
Persona #3: The creator

CREATOR USE CASES

CREATOR ● Generate content in your voice (blog posts, social media,
etc.)
STRATEGIST ● Draft product briefs or specs
Be your thought
Build your ● Review a product landing page and write a press release
partner about it
first draft
● Generate code
● Draft your employee reviews
● Write an outline of just about anything
Case Study

Ask AI to write the first draft of your

employee review.
= start your review 25% of the way there
Case Study: Using AI to write a review
Rating of AIs after uploading a 1:1 agenda document and asking AI to draft an employee review

My rating Insights

● Effectively extracted accomplishments and things to work on from 1:1 agenda (no
hallucinations)
Claude from Anthropic A
● Identified 4 out of the 5 accomplishments I would have identified (highly strategic)
● Identified unique opportunities based on priorities and employee’s tone in 1:1 agenda

● Effectively extracted accomplishments and things to work on from 1:1 agenda (no
hallucinations)
ChatGPT-4 B
● Selected more tactical, rather than strategic accomplishments (more literal)
● Areas of opportunity were vague (could have been true of anyone)

● Effectively extracted accomplishments and things to work on from 1:1 agenda (no
hallucinations)
Gemini Advanced B-
● Seemed to extract and select more tactical accomplishments like GPT-4 but did highlight but
did highlight 1 out of 5 strategic accomplishments

● Could not upload PDF

Microsoft Copilot N/A
Reality check: How to think about hallucinations

ASSISTANT STRATEGIST CREATOR

Cut out drudge work Be your thought partner Build your first draft

● Perform spot checks and ● Ask more than one LLM ● Remember the output is a
request source links and compare results draft – proofread and fact
check
● Conduct a deeper ● Run past a trusted human
investigations when you thought partner ● Ask AI to proof read and
find inaccuracies edit itself
● Utilize Claude or GPT-4
● QA data analysis and (more business minded ● Get specific - the more
other outputs as you tools) info you provide, the less
would an interns AI will hallucinate to fill in
gaps
QUICK POLL

Q. Which persona (assistant, strategist,

creator) are you most excited about using in
your work?
It’s another THE INTERNET RESEARCH REPORTS

tool in your
tool kit

YOUR NETWORK AI

Your colleagues Your mentor

Your manager Your customers
 Climb your learning curve

DAILY
USAGE

ADVANCED
PROMPTING
WOW!

UPLOADING
FILES

PROVIDE
CURIOUS NOOB CONTEXT

PROMPTING
BASICS
WAIT, WHAT??
30 minutes a
day. Make it a
habit.
If you could save 3 weeks a year, would you?

1,800
HOURS WORKED
X 10%
PRODUCTIVITY GAINS
= 22
DAYS SAVED
PER YEAR / TIME SAVED PER YEAR
My advice to you

Pick 1-2 specific use cases or prompts to start with –

allocate 30 minutes every day.

Don’t get discouraged by bad output – these tools will

continue to get better.

You need a habit – look for cues or triggers to remind

you to use AI as part of the process.

Acknowledge where you used AI to your stakeholders –

and where you didn’t.
AI as our co-pilot, not our master…

INPUT OUTPUT

INSTIGATOR AI JUDGE
Questions?