Prompt it, not Google it - Prompt Engineering for Data Scientists

“Prompt it”, not “Google it” :
Prompt Engineering for Stat
Programmers and Biostatisticians
Kevin Lee

Disclaimer
The views and opinions presented here represent those of
the speaker and should not be considered to represent
any companies or organizations.

Agenda
➢Prompt it, not Google it
➢What is Prompt Engineering?
➢How to Prompt Better
➢Prompt Engineering Technique
➢Use Cases in Biometrics
➢Discussion
3

Prompt it
is the new Google it
5

What is Prompt?
6
• Prompt – A tool to communicate with Gen AI (ChatGPT)
• “Prompt It” is the new “Google it”

7
How to
prompt
better
Prompt
Engineering

What is Prompt Engineering?
8
• Prompt Engineering – the art of creating effective
communication/ prompt that guide ChatGPT to generate
desired response.
• Bad Prompt – Bad Response
• Good Prompt – Good Response

How to work with the recent, smart graduate at work?
9

ChatGPT is a very smart, capable, recent
graduate
10
=?

How to prompt better
(How to communicate better)
- Get Specific
- Add Constraints
- Add more contexts
- Set the role
- Coach with feedback
- Probe further
- Step by step introduction

Get specific
12
Generic Prompt Specific Prompt
Write an essay about flowers. Write an essay about flowers at 9th
grade level
Write me a rap song Write me a rap song using
references to SpongeBob Square
Pants.
Write me a promotion request. Write me a promotion request for
Manager of Statistical Programmer.
• Generic inquiry will provide generic (and less impressive) answers

Add Parameter or Constraints
13
Generic Prompt Prompt with parameter
Write an essay about flowers. Write an essay about flowers at 9th
grade level within 200 words
Tell me the best places to visit in
Prague.
Prague within 4 days.
Prague.
Prague with $500 budget.
• Restriction and constraints breed a creativity
• Example of parameters
• Under 100 words
• Include word of “creative” / do not use a word of “creative”
• Make it under $100
• Make it work for 15 people

Add more context
14
Generic Prompt Prompt with more context
Write a greeting email. I have not seen my friend for a
long time. I want to know how he
is doing. Write a greeting email.
• If one adds more background, ChatGPT can provide more relevant,
customized answers.
• Provide the reason that I am using ChatGPT

Set the role
15
Generic Prompt Prompt with role
Write a greeting email. I am a salesperson who markets
new medicine. Write a greeting
email.
Review this proposal and provide
your feedback.
Pretend you are the head of IT.
Review this proposal and provide
your feedback.
How do you respond below
proposal?
You are Steve Jobs. How do you
respond below proposal?
• If one gives a role to ChatGPT and ask, ChatGPT play a role given and
provide the answers.
• Example of roles : boss, parent, salesperson,

Coach ChatGPT with feedback
16
Generic Prompt Prompt with feedback
Write a greeting email.
Make it in 50 words.
Make it more formal.
Make it more persuasive.
• After ChatGPT provide an answer, one can provide a feedback and ask
again.
• By iterating, one can get the better answer

Probe further
17
Generic Prompt Prompts with probing
Tell me more about paragraph 2.
Tell me more about option 1and 2.
Please explain the attraction in
paragraph 2.
• After receiving the answers, one could probe further in the current chat.
• One can find more detail information from the certain part of previous
response.

Provide a reference
18
Generic Prompt Prompt with feedback
Write a greeting email. I want you to use #my_style to
write a greeting email.
#my_style:
………..
• This technique is called ‘Few-shot Learning’
• By providing reference/ examples, one could receive customized response.
• It will help ChatGPT to mimic writing style and format.

Provide step by step instruction
19
Generic Prompt Prompt with step by step introduction
Perform the following actions:
1. Summarize the below text
2. Translate in French
3. Convert into JSON format
• One can give step by step instruction on how ChatGPT think the answer
“step by step”
• This will help ChatGPT to perform better.

Prompt Engineering
Technique
- Zero Shot
- Few Shot
- Reflection
- Chain of Thought
- Tree of Thought

Prompt Technique : Zero-Shot
21
• Purpose : One time Query
• Method : Straight-up Query
• Examples
• Prompt : What is 2 + 8?
• Output : 2 + 8 equals 10.

Prompt Technique : Few-Shot
22
• Purpose : Teaching AI models to follow specific patterns
• Method : Feeding model with examples before asking the desired questions
• Examples
• Prompt :
2@4 = 8
3@3 = 9
4@2 = 8
5@3 = 15
4@2 =
• Output :
It seems like you have a pattern where you're using "@" as an operator. Based on
the examples you provided:
2@4 = 8 (2 multiplied by 4)
So, following this pattern, 4@2 would be equal to 4 multiplied by 2, which is 8.

Prompt Technique : Chain of Thought Example
23
• Prompt :
Classify the safety issues of the text.
Follow these steps in the classification.
1. Identify adverse event in text.
2. Identify patient information in text
3. Identify medication information in text
If text answers to “Yes” on all the steps,
classify as ‘Safety Issues’ Otherwise ‘No
Safety Issues’
Let’s think step by step for classification of
the text.
Text: Patient 001 has adverse events after
taking the medication.
• Output :
Let's follow the steps you provided to classify the safety issues in
the text:
Step 1: Identify adverse event in text
- The text mentions "Patient 001 has adverse events after taking
the medication." This indicates that there is an adverse event
mentioned in the text.
Step 2: Identify patient information in text
- The text mentions "Patient 001," which is patient information.
Step 3: Identify medication information in text
- The text mentions "taking the medication," which is medication
information.
Now, let's evaluate the results of each step:
- Step 1: Adverse event identified
- Step 2: Patient information identified
- Step 3: Medication information identified
Since all three steps have been answered "Yes," we can classify
this text as 'Safety Issues' because it contains information about a
patient experiencing adverse events after taking medication.

ChatGPT Use Cases in
Biometrics
- Simple Inquiry
- Content Development
- Coding
- Comparison
- Data Analysis
- Patient Profiles
- Safety Cases
- Image Analysis
- Art Generation

25
Simple Inquiry
Prompt :
What is
schizophrenia?
Tell me in 200
words.
Output :
Schizophrenia is a complex and chronic mental disorder that profoundly affects a person's thinking,
emotions, and behavior. It typically emerges in late adolescence or early adulthood, and its exact
cause remains unclear, though it is thought to involve a combination of genetic, environmental, and
neurobiological factors.
Individuals with schizophrenia often experience a range of symptoms, including hallucinations (false
sensory perceptions), delusions (false beliefs), disorganized thinking, and reduced emotional
expression. They may also struggle with cognitive impairments, such as difficulty with memory,
attention, and decision-making.
Schizophrenia can have a profound impact on a person's daily life and functioning, making it
challenging to maintain relationships, hold a job, or pursue personal goals. It can lead to significant
distress and impairment in social, occupational, and educational domains.
Treatment for schizophrenia typically involves a combination of antipsychotic medications,
psychotherapy, and psychosocial support. While these interventions can help manage symptoms
and improve the quality of life for individuals with schizophrenia, the condition is often chronic and
requires ongoing care and support.
Early intervention and a comprehensive, individualized treatment plan are crucial for managing
schizophrenia and helping affected individuals lead fulfilling lives while minimizing the impact of their
symptoms.

26
Content Development
Prompt :
I am salesperson.
I need to contact the doctors
about new medicine, which is
3 times more effective to lower
blood pressure.
write marketing emails within
100 words.
Output :
Subject: Revolutionize Blood Pressure Management with Our New Medicine!
Dear Dr. [Doctor's Last Name],
I hope this message finds you well. I am excited to introduce our breakthrough
medication, [Medicine Name], designed to transform blood pressure management.
With its cutting-edge formula, it is proven to be three times more effective in lowering
blood pressure compared to existing solutions.
[Medicine Name] offers a unique opportunity for your patients to achieve better
health outcomes. Its advanced technology ensures quicker results and fewer side
effects, leading to enhanced patient compliance.
We would be honored to provide you with more information and samples for your
evaluation. Together, we can make a significant impact on improving patients' lives.
Let's revolutionize hypertension treatment together. Please reply to this email, and I'll
be happy to arrange a meeting at your convenience.
Best Regards,
[Your Name]
[Your Title]
[Your Contact Information]

27
Codes conversion
Prompt :
Convert below SAS codes to R codes:
data dm;
merge demo rand;
by usubjid ;
run;
Output :
You can convert the SAS code to R using the `merge` function in R to
perform the equivalent operation. Assuming you have two data frames
`demo` and `rand` that you want to merge by the `usubjid` variable,
here's the equivalent R code:
```
# Assuming you have loaded your data into two data frames: demo
and rand
# Merge the data frames by usubjid
dm <- merge(demo, rand, by="usubjid")
```
In this code:
- `demo` and `rand` are your input data frames.
- We use the `merge` function to merge the data frames.
- The `by` parameter specifies the key variable by which to merge the
data frames, which in this case is `usubjid`.
After running this code, the merged data frame `dm` will contain the
result of the merge operation, similar to what you would achieve with
the SAS code you provided.

28
Comparison between SAS and R
Prompt :
I want you to compare SAS and R
programming. Provide me with
differences, present in a table format
Output :

29
Patient Profile Development
• Prompt :
Use below data to provide patient profiles:
{ 'patient' : '01-001', 'sex' : 'male', 'age' : 25, 'race' : 'White'}
• Output :
Based on the provided data, here is a patient profile for '01-001':
- Patient ID: 01-001
- Sex: Male
- Age: 25
- Race: White
This patient is a 25-year-old male of White race. Please let me know if you need any
additional information or if you would like to create profiles for more patients.

30
Data Analysis (1)
• Prompt :
Select patients whose sex is male. Use below "Patient Data". Provide me the answers only.
“Patient Data”:
{ 'patient' : '01-001', 'sex' : 'male', 'age' : 25, 'race' : 'White’},
{ 'patient' : '01-002', 'sex' : ‘female', 'age’ : 28, 'race' : ‘Asian’},
{ 'patient' : ‘02-001', 'sex' : ‘female', 'age’ : 50, 'race' : ‘White’},
{ 'patient' : ‘02-002', 'sex' : 'male', 'age’ : 40, 'race' : ‘Black’}
• Output :
The selected patients whose sex is male are:
1. {'patient': '01-001', 'sex': 'male', 'age': 25, 'race': 'White'}
2. {'patient': '02-002', 'sex': 'male', 'age': 40, 'race': 'Black'}

31
Data Analysis (2)
• Prompt :
Count below “Patient Data” by race
“Patient Data”:
{ 'patient' : '01-001', 'sex' : 'male', 'age' : 25, 'race' : 'White’},
{ 'patient' : '01-002', 'sex' : ‘female', 'age’ : 28, 'race' : ‘Asian’},
{ 'patient' : ‘02-001', 'sex' : ‘female', 'age’ : 50, 'race' : ‘White’},
{ 'patient' : ‘02-002', 'sex' : 'male', 'age’ : 40, 'race' : ‘Black’}
• Output :
Certainly! Let's count the number of patients by race in the given "Patient Data":
- White: 2
- Asian: 1
- Black: 1

Boston Consulting : ChatGPT Users vs Non-users
• Results of consultants
using ChatGPT
• finished 12.2% more
tasks on average
• completed tasks
25.1% more quickly
• produced 40%
higher quality results
32

33
What can we ( Stat Programmers and
Biostatisticans) when we are fully utilizing with
prompt engineering?

Prompt it, not Google it - Prompt Engineering for Data Scientists

More Related Content

Similar to Prompt it, not Google it - Prompt Engineering for Data Scientists (20)

More from Kevin Lee (20)

Recently uploaded (20)

Prompt it, not Google it - Prompt Engineering for Data Scientists