SlideShare a Scribd company logo
How to Encourage Children to Read Books?
LLMs for Education and Fun!
22/11/2024
Maciej Dzieżyc, Ewelina Kucal
Delivering outcomes
in software, cloud,
data and AI
3
& the challenge
4
5
6
7
8
9
10
Book Aligned Activates (BAAs) categories
Activity Category
Code
Activity Category
Name Activity Skill
GROW
Growing Positive
Emotions
Self Awareness
Self Management
Responsible
Decision-Making
Relationship Skills
Social Awareness
PLAY Play-based
Physical
Symbolic
Pretend
Play with Objects
Game with Rules
READ Reading Champion
Phonics
Phonemic Awareness
Vocabulary
Fluency
Reading
Comprehension
Writing
11
Solution
12
AWS Services 101
• Amazon Simple Service Storage (Amazon S3)
• Objects (files) stored in buckets
• S3 Events (“notifications”) about changes, e.g., new file upload
• AWS Lambda
• Serverless function running in the cloud
• Put Python code and just run it
• Amazon Bedrock
• Collection of APIs to Large Language Models (among other services)
• AWS Step Functions
• Workflow Orchestration – glue between elements of the solution
• Easier visibility into data being send between modules
13
Architecture
Model choice
14
• Amazon Bedrock API
• Pay-as-you-go
• Basically, no setup required
• Claude 3 Opus
• Why this model?
• The state-of-the-art model on Amazon Bedrock at
the time of delivery
• Low load of requests
• Multilingual input
• No fine-tuning necessary
How to find SOTA model?
15
• Leaderboard at lmarena.ai
• LM Arena asks users to blindly rate outputs of the
LLMs for the user provided prompt
• Elo score calculated like in chess or online
multiplayer games
• This method has its limitations, but it’s best quick
assessment
System message
16
Feature-specific set of instructions
GOAL:
o to direct and improve the quality and safety of a
model’s output.
BEST PRACTICES:
o Use clear and direct language
o Keep instructions concise
o Emphasize certain words by bolding **words**
o Address the AI system using first-person language
o Build robust instructions
o Repeat most important constraints at the end
17
Long and elaborate system message
Define task
Explanation of pre/post
Reader’s context
BAAs requirements
BAAs examples
(few-shot learning)
The structure of the output
(chain of thought)
18
Zero-shot vs Few-shot approaches
Zero-shot
Harry, Ron and Hermione walk away from their last year at Hogwarts to find and destroy the
remaining Horcruxes, putting an end to Voldemort's bid for immortality. But with Harry's
beloved Dumbledore dead and Voldemort's unscrupulous Death Eaters on the loose, the
world is more dangerous than ever.
Genre:
19
Zero-shot vs Few-shot approaches
Few-shot
Paul Atreides, a brilliant and gifted young man born into a great destiny beyond his
understanding, must travel to the most dangerous planet in the universe to ensure the future
of his family and his people. As malevolent forces explode into conflict over the planet's
exclusive supply of the most precious resource in existence-a commodity capable of
unlocking humanity's greatest potential-only those who can conquer their fear will survive.
['Action', 'Adventure', 'Science Fiction’]
A botched store robbery places Wonder Woman in a global battle against a powerful and
mysterious ancient force that puts her powers in jeopardy.
['Action', 'Adventure', 'Fantasy']
After the devastating events of Avengers: Infinity War, the universe is in ruins due to the
efforts of the Mad Titan, Thanos. With the help of remaining allies, the Avengers must
assemble once more in order to undo Thanos' actions and restore order to the universe once
and for all, no matter what consequences may be in store.
['Adventure', 'Science Fiction', 'Action']
Harry, Ron and Hermione walk away from their last year at Hogwarts to find and destroy the
remaining Horcruxes, putting an end to Voldemort's bid for immortality. But with Harry's
beloved Dumbledore dead and Voldemort's unscrupulous Death Eaters on the loose, the
world is more dangerous than ever.
20
Few-shot learning
Prompting techniques
• Few-shot learning and zero-shot learning are a
by-product of emerging properties of LLMs
• We used a few-shot learning in the first version of
the prompt
• Why few-shot over zero-shot?
• Activities should not only reflect values, requirements
etc. but also “vibe” and style of existing one
• Few-shot – more tokens – higher cost
21
Chain of thought
Prompting techniques
• Challenge: model was not complaint - Claude 3 Opus sometimes returned activities in the language other
than book’s e.g., Spanish instead of English
• Solution:
The output template:
``` “Thought” which we discard in post-processing
Detected language: [English/Swahili]
[ { "skill": "Vocabulary", "placement": "Pre/Post", "activity": "...", "title": "..." }, ... ]
```
22
The Challenge of Validation
How did we get to this prompt?
• There’s no ground truth like in
Retrieval Augmented
Generation Chatbots or
classical Machine Learning
• Solution? Just ask the client!
• Write a prompt
• Identify pitfalls in outputs
• Correct
• Generate outputs for the client
• Gather feedback
• Repeat
23
Outcomes
24
Heroes to Zero #4: More Clean For More Green
{
"skill": "Self Awareness",
"placement": "Pre",
"activity": "The main character in this story is good at
understanding what people need. Before reading, think
about what you're good at. What are your strengths?
Write about or discuss with someone a time when you
used one of your strengths to accomplish something.",
"title": "Knowing Your Strengths",
"category": "Grow"
}
25
Rini and Jojo Work It Out: The Super-Fast Apple Sorting Machine
{
"skill": "Physical",
"placement": "Post",
"activity": "Rini and Jojo made an apple sorting machine.
Gather some small balls or rolled up socks.
See if you can gently toss them into a basket or box.
Take a step back each time to make it more challenging.
How many can you get in a row?",
"title": "Toss and Sort",
"category": "Play“
}
Summary
26
• Activity generation time reduced from days to
seconds
• More creative, less repetitive activities
• Solution delivered within a month
• High value with out of the box tools
• Scaling to zero architecture
27
Key takeaways
• There are founds for AI4Good
• LLMs are not only user facing chatbots
• Close collaboration with a client was crucial as
there was no ground truth
• Start small and build around feedback and failure
Thank you
Questions?
How to reach us?
dziezyc.ai
LinkedIn: Ewelina Kucal
29
Disclaimer
• This publication contains proprietary information that is protected by copyright. SoftwareOne reserves all
rights thereto.
• SoftwareOne shall not be liable for possible errors in this document. Liability for damages directly and
indirectly associated with the supply or use of this document is excluded as far as legally permissible.
• The information presented herein is intended exclusively as a guide offered by SoftwareOne. The publisher's
product use rights, agreement terms and conditions and other definitions prevail over the information
provided herein. The content must not be copied, reproduced, passed to third parties or used for any other
purposes without written permission of SoftwareOne
• Copyright© by SoftwareOne. All Rights Reserved. SoftwareOne is a registered trademark of SoftwareOne.
All other trademarks, service marks or trade names appearing herein are the property of their respective
owners.

More Related Content

PPTX
James Coplien - Trygve - October 17, 2016
Foo Café Copenhagen
 
PDF
Peter Norvig - NYC Machine Learning 2013
Michael Scovetta
 
PDF
Artificial Intelligence and Intuition
Viktor Dörfler
 
PPTX
Software Craftsmanship and Agile Code Games
Mike Clement
 
PPT
2012 My Top 10 Tools
Howie DiBlasi
 
PPTX
Visual tools and innovation games - full day workshop - sp intersections - no...
Ruven Gotz
 
PPTX
Robin Gadd - The Future Of Technologies For Learning March 2011
robingadd
 
PDF
Design and Darkmatter, Connecting Storytelling with Business Outcomes
Trip O'Dell
 
James Coplien - Trygve - October 17, 2016
Foo Café Copenhagen
 
Peter Norvig - NYC Machine Learning 2013
Michael Scovetta
 
Artificial Intelligence and Intuition
Viktor Dörfler
 
Software Craftsmanship and Agile Code Games
Mike Clement
 
2012 My Top 10 Tools
Howie DiBlasi
 
Visual tools and innovation games - full day workshop - sp intersections - no...
Ruven Gotz
 
Robin Gadd - The Future Of Technologies For Learning March 2011
robingadd
 
Design and Darkmatter, Connecting Storytelling with Business Outcomes
Trip O'Dell
 

Similar to [DSC Europe 24] Ewelina Kucal & Maciej Dziezyc - How to Encourage Children to Read Books? LLMs for Education and Fun! (20)

PDF
UX STRAT Europe 2019: Zhaochang He, VMware
UX STRAT
 
PPTX
Howtostopsucking
Hugo Pinto
 
PPTX
How to stop sucking and be awesome instead
codinghorror
 
PPTX
Howtostopsuckingandbeawesomeinstead 120601013410-phpapp01
Hugo Pinto
 
PDF
How2research
kewuc
 
KEY
Bluffers guide to elitist jargon - Martijn Verburg, Richard Warburton, James ...
JAX London
 
PPTX
Tale of the Knowledge Organization In an Age of Wicked Problems
GomindSHIFT
 
PPTX
Memo oct2013
Stephen Abram
 
PPTX
Learning in the Age of Knowledge on Demand
Tim O'Reilly
 
PDF
CFP workshop
Amit Zur
 
PPTX
Building Social Software for the Anti-Social: Part I
codinghorror
 
PDF
BSidesLV 2013 - Using Machine Learning to Support Information Security
Alex Pinto
 
PDF
Visual Tools and Innovation Games - full day workshop - dev intersections - n...
Michelle Caldwell, PSM, SSGB
 
PDF
Get connected with python
Jan Kroon
 
PDF
Deep Learning Lightning Talk
Mateusz Buśkiewicz
 
PPT
SLoodle: Moodle and Second Life
gpotter
 
PPT
Augmented Reality in the Library
Eliot Boden
 
PDF
LRMI: Peek Under the Hood of Personalized Learning
AAP PreK-12 Learning Group
 
KEY
Bluffers guide to Terminology
Jim Gough
 
PDF
Future-Proof Your L&D With New Tech and Gamification
Aggregage
 
UX STRAT Europe 2019: Zhaochang He, VMware
UX STRAT
 
Howtostopsucking
Hugo Pinto
 
How to stop sucking and be awesome instead
codinghorror
 
Howtostopsuckingandbeawesomeinstead 120601013410-phpapp01
Hugo Pinto
 
How2research
kewuc
 
Bluffers guide to elitist jargon - Martijn Verburg, Richard Warburton, James ...
JAX London
 
Tale of the Knowledge Organization In an Age of Wicked Problems
GomindSHIFT
 
Memo oct2013
Stephen Abram
 
Learning in the Age of Knowledge on Demand
Tim O'Reilly
 
CFP workshop
Amit Zur
 
Building Social Software for the Anti-Social: Part I
codinghorror
 
BSidesLV 2013 - Using Machine Learning to Support Information Security
Alex Pinto
 
Visual Tools and Innovation Games - full day workshop - dev intersections - n...
Michelle Caldwell, PSM, SSGB
 
Get connected with python
Jan Kroon
 
Deep Learning Lightning Talk
Mateusz Buśkiewicz
 
SLoodle: Moodle and Second Life
gpotter
 
Augmented Reality in the Library
Eliot Boden
 
LRMI: Peek Under the Hood of Personalized Learning
AAP PreK-12 Learning Group
 
Bluffers guide to Terminology
Jim Gough
 
Future-Proof Your L&D With New Tech and Gamification
Aggregage
 
Ad

More from DataScienceConferenc1 (20)

PPTX
[DSC Europe 24] Anastasia Shapedko - How Alice, our intelligent personal assi...
DataScienceConferenc1
 
PPTX
[DSC Europe 24] Joy Chatterjee - Balancing Personalization and Experimentatio...
DataScienceConferenc1
 
PPTX
[DSC Europe 24] Pratul Chakravarty - Personalized Insights and Engagements us...
DataScienceConferenc1
 
PPTX
[DSC Europe 24] Domagoj Maric - Modern Web Data Extraction: Techniques, Tools...
DataScienceConferenc1
 
PPTX
[DSC Europe 24] Marcin Szymaniuk - The path to Effective Data Migration - Ove...
DataScienceConferenc1
 
PPTX
[DSC Europe 24] Fran Mikulicic - Building a Data-Driven Culture: What the C-S...
DataScienceConferenc1
 
PPTX
[DSC Europe 24] Sofija Pervulov - Building up the Bosch Semantic Data Lake
DataScienceConferenc1
 
PDF
[DSC Europe 24] Dani Ei-Ayyas - Overcoming Loneliness with LLM Dating Assistant
DataScienceConferenc1
 
PPTX
[DSC Europe 24] Nikola Milosevic - VerifAI: Biomedical Generative Question-An...
DataScienceConferenc1
 
PPTX
[DSC Europe 24] Josip Saban - Buidling cloud data platforms in enterprises
DataScienceConferenc1
 
PPTX
[DSC Europe 24] Sray Agarwal - 2025: year of Ai dilemma - ethics, regulations...
DataScienceConferenc1
 
PDF
[DSC Europe 24] Peter Kertys & Maros Buban - Application of AI technologies i...
DataScienceConferenc1
 
PPTX
[DSC Europe 24] Orsalia Andreou - Fostering Trust in AI-Driven Finance
DataScienceConferenc1
 
PPTX
[DSC Europe 24] Arnault Ioualalen - AI Trustworthiness – A Path Toward Mass A...
DataScienceConferenc1
 
PDF
[DSC Europe 24] Nathan Coyle - Open Data for Everybody: Social Action, Peace ...
DataScienceConferenc1
 
PPTX
[DSC Europe 24] Miodrag Vladic - Revolutionizing Information Access: All Worl...
DataScienceConferenc1
 
PPTX
[DSC Europe 24] Katherine Munro - Where there’s a will, there’s a way: The ma...
DataScienceConferenc1
 
PPTX
[DSC Europe 24] Ana Stojkovic Knezevic - How to effectively manage AI/ML proj...
DataScienceConferenc1
 
PPTX
[DSC Europe 24] Simun Sunjic & Lovro Matosevic - Empowering Sales with Intell...
DataScienceConferenc1
 
PPTX
[DSC Europe 24] Igor Sevo - Intelligent Interfaces and operating systems
DataScienceConferenc1
 
[DSC Europe 24] Anastasia Shapedko - How Alice, our intelligent personal assi...
DataScienceConferenc1
 
[DSC Europe 24] Joy Chatterjee - Balancing Personalization and Experimentatio...
DataScienceConferenc1
 
[DSC Europe 24] Pratul Chakravarty - Personalized Insights and Engagements us...
DataScienceConferenc1
 
[DSC Europe 24] Domagoj Maric - Modern Web Data Extraction: Techniques, Tools...
DataScienceConferenc1
 
[DSC Europe 24] Marcin Szymaniuk - The path to Effective Data Migration - Ove...
DataScienceConferenc1
 
[DSC Europe 24] Fran Mikulicic - Building a Data-Driven Culture: What the C-S...
DataScienceConferenc1
 
[DSC Europe 24] Sofija Pervulov - Building up the Bosch Semantic Data Lake
DataScienceConferenc1
 
[DSC Europe 24] Dani Ei-Ayyas - Overcoming Loneliness with LLM Dating Assistant
DataScienceConferenc1
 
[DSC Europe 24] Nikola Milosevic - VerifAI: Biomedical Generative Question-An...
DataScienceConferenc1
 
[DSC Europe 24] Josip Saban - Buidling cloud data platforms in enterprises
DataScienceConferenc1
 
[DSC Europe 24] Sray Agarwal - 2025: year of Ai dilemma - ethics, regulations...
DataScienceConferenc1
 
[DSC Europe 24] Peter Kertys & Maros Buban - Application of AI technologies i...
DataScienceConferenc1
 
[DSC Europe 24] Orsalia Andreou - Fostering Trust in AI-Driven Finance
DataScienceConferenc1
 
[DSC Europe 24] Arnault Ioualalen - AI Trustworthiness – A Path Toward Mass A...
DataScienceConferenc1
 
[DSC Europe 24] Nathan Coyle - Open Data for Everybody: Social Action, Peace ...
DataScienceConferenc1
 
[DSC Europe 24] Miodrag Vladic - Revolutionizing Information Access: All Worl...
DataScienceConferenc1
 
[DSC Europe 24] Katherine Munro - Where there’s a will, there’s a way: The ma...
DataScienceConferenc1
 
[DSC Europe 24] Ana Stojkovic Knezevic - How to effectively manage AI/ML proj...
DataScienceConferenc1
 
[DSC Europe 24] Simun Sunjic & Lovro Matosevic - Empowering Sales with Intell...
DataScienceConferenc1
 
[DSC Europe 24] Igor Sevo - Intelligent Interfaces and operating systems
DataScienceConferenc1
 
Ad

Recently uploaded (20)

PDF
A Systems Thinking Approach to Algorithmic Fairness.pdf
Epistamai
 
PDF
Mastering Query Optimization Techniques for Modern Data Engineers
Accentfuture
 
PPT
Chapter 3 METAL JOINING.pptnnnnnnnnnnnnn
JanakiRaman206018
 
PDF
Data Science Trends & Career Guide---ppt
jisajoy3061
 
PDF
Master Databricks SQL with AccentFuture – The Future of Data Warehousing
Accentfuture
 
PPTX
Moving the Public Sector (Government) to a Digital Adoption
PaulYoung221210
 
PDF
AI Lect 2 Identifying AI systems, branches of AI, etc.pdf
mswindow00
 
PDF
TIC ACTIVIDAD 1geeeeeeeeeeeeeeeeeeeeeeeeeeeeeer3.pdf
Thais Ruiz
 
PPTX
Global journeys: estimating international migration
Office for National Statistics
 
PDF
Company Presentation pada Perusahaan ADB.pdf
didikfahmi
 
PPTX
Understanding Prototyping in Design and Development
SadiaJanjua2
 
PPTX
Economic Sector Performance Recovery.pptx
yulisbaso2020
 
PPTX
Major-Components-ofNKJNNKNKNKNKronment.pptx
dushyantsharma1221
 
PPTX
Azure Data management Engineer project.pptx
sumitmundhe77
 
PPTX
Bharatiya Antariksh Hackathon 2025 Idea Submission PPT.pptx
abhinavmemories2026
 
PPTX
Introduction to Biostatistics Presentation.pptx
AtemJoshua
 
PDF
Linux OS guide to know, operate. Linux Filesystem, command, users and system
Kiran Maharjan
 
PDF
CH2-MODEL-SETUP-v2017.1-JC-APR27-2017.pdf
jcc00023con
 
PPTX
Purple and Violet Modern Marketing Presentation (1).pptx
SanthoshKumar229321
 
PPTX
Data Security Breach: Immediate Action Plan
varmabhuvan266
 
A Systems Thinking Approach to Algorithmic Fairness.pdf
Epistamai
 
Mastering Query Optimization Techniques for Modern Data Engineers
Accentfuture
 
Chapter 3 METAL JOINING.pptnnnnnnnnnnnnn
JanakiRaman206018
 
Data Science Trends & Career Guide---ppt
jisajoy3061
 
Master Databricks SQL with AccentFuture – The Future of Data Warehousing
Accentfuture
 
Moving the Public Sector (Government) to a Digital Adoption
PaulYoung221210
 
AI Lect 2 Identifying AI systems, branches of AI, etc.pdf
mswindow00
 
TIC ACTIVIDAD 1geeeeeeeeeeeeeeeeeeeeeeeeeeeeeer3.pdf
Thais Ruiz
 
Global journeys: estimating international migration
Office for National Statistics
 
Company Presentation pada Perusahaan ADB.pdf
didikfahmi
 
Understanding Prototyping in Design and Development
SadiaJanjua2
 
Economic Sector Performance Recovery.pptx
yulisbaso2020
 
Major-Components-ofNKJNNKNKNKNKronment.pptx
dushyantsharma1221
 
Azure Data management Engineer project.pptx
sumitmundhe77
 
Bharatiya Antariksh Hackathon 2025 Idea Submission PPT.pptx
abhinavmemories2026
 
Introduction to Biostatistics Presentation.pptx
AtemJoshua
 
Linux OS guide to know, operate. Linux Filesystem, command, users and system
Kiran Maharjan
 
CH2-MODEL-SETUP-v2017.1-JC-APR27-2017.pdf
jcc00023con
 
Purple and Violet Modern Marketing Presentation (1).pptx
SanthoshKumar229321
 
Data Security Breach: Immediate Action Plan
varmabhuvan266
 

[DSC Europe 24] Ewelina Kucal & Maciej Dziezyc - How to Encourage Children to Read Books? LLMs for Education and Fun!

  • 1. How to Encourage Children to Read Books? LLMs for Education and Fun! 22/11/2024 Maciej Dzieżyc, Ewelina Kucal
  • 2. Delivering outcomes in software, cloud, data and AI
  • 4. 4
  • 5. 5
  • 6. 6
  • 7. 7
  • 8. 8
  • 9. 9
  • 10. 10 Book Aligned Activates (BAAs) categories Activity Category Code Activity Category Name Activity Skill GROW Growing Positive Emotions Self Awareness Self Management Responsible Decision-Making Relationship Skills Social Awareness PLAY Play-based Physical Symbolic Pretend Play with Objects Game with Rules READ Reading Champion Phonics Phonemic Awareness Vocabulary Fluency Reading Comprehension Writing
  • 12. 12 AWS Services 101 • Amazon Simple Service Storage (Amazon S3) • Objects (files) stored in buckets • S3 Events (“notifications”) about changes, e.g., new file upload • AWS Lambda • Serverless function running in the cloud • Put Python code and just run it • Amazon Bedrock • Collection of APIs to Large Language Models (among other services) • AWS Step Functions • Workflow Orchestration – glue between elements of the solution • Easier visibility into data being send between modules
  • 14. Model choice 14 • Amazon Bedrock API • Pay-as-you-go • Basically, no setup required • Claude 3 Opus • Why this model? • The state-of-the-art model on Amazon Bedrock at the time of delivery • Low load of requests • Multilingual input • No fine-tuning necessary
  • 15. How to find SOTA model? 15 • Leaderboard at lmarena.ai • LM Arena asks users to blindly rate outputs of the LLMs for the user provided prompt • Elo score calculated like in chess or online multiplayer games • This method has its limitations, but it’s best quick assessment
  • 16. System message 16 Feature-specific set of instructions GOAL: o to direct and improve the quality and safety of a model’s output. BEST PRACTICES: o Use clear and direct language o Keep instructions concise o Emphasize certain words by bolding **words** o Address the AI system using first-person language o Build robust instructions o Repeat most important constraints at the end
  • 17. 17 Long and elaborate system message Define task Explanation of pre/post Reader’s context BAAs requirements BAAs examples (few-shot learning) The structure of the output (chain of thought)
  • 18. 18 Zero-shot vs Few-shot approaches Zero-shot Harry, Ron and Hermione walk away from their last year at Hogwarts to find and destroy the remaining Horcruxes, putting an end to Voldemort's bid for immortality. But with Harry's beloved Dumbledore dead and Voldemort's unscrupulous Death Eaters on the loose, the world is more dangerous than ever. Genre:
  • 19. 19 Zero-shot vs Few-shot approaches Few-shot Paul Atreides, a brilliant and gifted young man born into a great destiny beyond his understanding, must travel to the most dangerous planet in the universe to ensure the future of his family and his people. As malevolent forces explode into conflict over the planet's exclusive supply of the most precious resource in existence-a commodity capable of unlocking humanity's greatest potential-only those who can conquer their fear will survive. ['Action', 'Adventure', 'Science Fiction’] A botched store robbery places Wonder Woman in a global battle against a powerful and mysterious ancient force that puts her powers in jeopardy. ['Action', 'Adventure', 'Fantasy'] After the devastating events of Avengers: Infinity War, the universe is in ruins due to the efforts of the Mad Titan, Thanos. With the help of remaining allies, the Avengers must assemble once more in order to undo Thanos' actions and restore order to the universe once and for all, no matter what consequences may be in store. ['Adventure', 'Science Fiction', 'Action'] Harry, Ron and Hermione walk away from their last year at Hogwarts to find and destroy the remaining Horcruxes, putting an end to Voldemort's bid for immortality. But with Harry's beloved Dumbledore dead and Voldemort's unscrupulous Death Eaters on the loose, the world is more dangerous than ever.
  • 20. 20 Few-shot learning Prompting techniques • Few-shot learning and zero-shot learning are a by-product of emerging properties of LLMs • We used a few-shot learning in the first version of the prompt • Why few-shot over zero-shot? • Activities should not only reflect values, requirements etc. but also “vibe” and style of existing one • Few-shot – more tokens – higher cost
  • 21. 21 Chain of thought Prompting techniques • Challenge: model was not complaint - Claude 3 Opus sometimes returned activities in the language other than book’s e.g., Spanish instead of English • Solution: The output template: ``` “Thought” which we discard in post-processing Detected language: [English/Swahili] [ { "skill": "Vocabulary", "placement": "Pre/Post", "activity": "...", "title": "..." }, ... ] ```
  • 22. 22 The Challenge of Validation How did we get to this prompt? • There’s no ground truth like in Retrieval Augmented Generation Chatbots or classical Machine Learning • Solution? Just ask the client! • Write a prompt • Identify pitfalls in outputs • Correct • Generate outputs for the client • Gather feedback • Repeat
  • 24. 24 Heroes to Zero #4: More Clean For More Green { "skill": "Self Awareness", "placement": "Pre", "activity": "The main character in this story is good at understanding what people need. Before reading, think about what you're good at. What are your strengths? Write about or discuss with someone a time when you used one of your strengths to accomplish something.", "title": "Knowing Your Strengths", "category": "Grow" }
  • 25. 25 Rini and Jojo Work It Out: The Super-Fast Apple Sorting Machine { "skill": "Physical", "placement": "Post", "activity": "Rini and Jojo made an apple sorting machine. Gather some small balls or rolled up socks. See if you can gently toss them into a basket or box. Take a step back each time to make it more challenging. How many can you get in a row?", "title": "Toss and Sort", "category": "Play“ }
  • 26. Summary 26 • Activity generation time reduced from days to seconds • More creative, less repetitive activities • Solution delivered within a month • High value with out of the box tools • Scaling to zero architecture
  • 27. 27 Key takeaways • There are founds for AI4Good • LLMs are not only user facing chatbots • Close collaboration with a client was crucial as there was no ground truth • Start small and build around feedback and failure
  • 28. Thank you Questions? How to reach us? dziezyc.ai LinkedIn: Ewelina Kucal
  • 29. 29 Disclaimer • This publication contains proprietary information that is protected by copyright. SoftwareOne reserves all rights thereto. • SoftwareOne shall not be liable for possible errors in this document. Liability for damages directly and indirectly associated with the supply or use of this document is excluded as far as legally permissible. • The information presented herein is intended exclusively as a guide offered by SoftwareOne. The publisher's product use rights, agreement terms and conditions and other definitions prevail over the information provided herein. The content must not be copied, reproduced, passed to third parties or used for any other purposes without written permission of SoftwareOne • Copyright© by SoftwareOne. All Rights Reserved. SoftwareOne is a registered trademark of SoftwareOne. All other trademarks, service marks or trade names appearing herein are the property of their respective owners.