1 of 63

Introductions

•Rename Yourself

•First Name

•Role

•T = Teacher

•A = Administrator

•C = Coach

•O = Other

•AEA = AEA

•Y or N if used ChatGPT

•Example: Aaron AEA Y

Resource Page and Slides https://bit.ly/AICSIN

2 of 63

Empowering Building Leaders with AI: Actionable Insights for Everyday Tasks

3 of 63

Agenda

    • AI Overview
    • Current Landscape of Generative AI
    • Exploring and Using ChatGPT
    • Prompt Engineering Overview Part 1
    • ChatGPT Experimentation – Basics
    • Prompt Engineering Overview Part 2
    • ChatGPT Experimentation – Deeper Dive
    • Full Circle Conversation

4 of 63

How are you feeling about AI, REALLY?

5 of 63

Today we invite you to….

  • Be curious
  • Seek synergies
  • Share expertise

6 of 63

What is all this AI stuff?

What has happened?

7 of 63

8 of 63

Let’s remember AI is not new

  • AI is used by
    • Sprint to detect patterns in customer dissatisfaction
    • Netflix to determine what type of movies people like to see
    • Sports teams to predict in-game player reactions and contest outcomes
    • Apple uses AI to recognize your face before allowing you to access your devices

9 of 63

10 of 63

November 30, 2022

ChatGPT was released for public use.

11 of 63

March 14, 2023

ChatGPT 4 was released for public use.

12 of 63

May 18, 2023

ChatGPT App

13 of 63

May 25, 2023

ChatGPT App available in 11 more countries

14 of 63

15 of 63

We need to question our assumptions about the way things are now.��They will not always be this way!

16 of 63

AI Players

  • Big players
    • Microsoft, with its Bing AI (and Copilot)
    • Google, with Bard
    • OpenAI with ChatGPT-4
  • Keep an eye on

17 of 63

Google AI – Awareness, but not focus today

18 of 63

Google Examples

19 of 63

Google Examples

20 of 63

Google Examples

21 of 63

  • Use images in your prompts: Images are a big part of how we put our imaginations to work. At I/O we announced we’re bringing the capabilities of Google Lens into Bard. Whether you want more information about an image or just need help coming up with a caption, you can now upload images with prompts and Bard will analyze the photo to help. This feature is now live in English, and we’ll expand to new languages soon.

22 of 63

23 of 63

Reframe Thinking

  • How can we transform the Age of AI from being the Age of Available Answers to the Age of Articulate Asking?

24 of 63

HI + AI = New Path Forward���AI is a co-pilot NOT auto-pilot

25 of 63

Examine, Dismantle, Rethink, Redesign

26 of 63

Explore ChatGPT

27 of 63

Prompt Engineering Adventure

  • What prompts did you try?
  • Take note of the strengths and weaknesses of the responses.
  • What worked well?
  • What could be improved?
  • Think about how these insights can be applied to your educational context.
  • Prompts not as important as interaction.

28 of 63

Four C’s of Using ChatGPT

  • Work of Erin Lenihan and Rachel Rozen

29 of 63

Section 1

The Fundamentals of Prompt Engineering

30 of 63

The AI Classroom

31 of 63

32 of 63

Set the scene

Be clear and concise

Examples

Create an academic quiz

Read the following text and be prepared to answer questions on it

Create an outline of a presentation on…..

Grade this answer to this question and give reasons for your judgment

33 of 63

AI needs to know what needs to be accomplished and how to approach the work

Examples

You are a 5th grader

You are an academic professor of ……

You are an expert in the field of……

You are William Shakespeare. Answer all questions using the knowledge Shakespeare had and in his style.

34 of 63

Be clear, concise, and specific in expectations

Examples

Write three examples of a blog post title based on the following information……

Create 10 multiple choice questions using Bloom’s Taxonomy to make sure the questions develop a deeper understanding on the following content…..

This is the exam question……..This is the answer you are using to grade…. Provide a score and reasons for the following response.

35 of 63

Establish boundaries and clear expectations of output

Examples

Write in 100 words

Format in the style of a tweet

Write in British

36 of 63

Think About

  • 3-5 example tasks or questions you would like a computer to generate text for.
    • Travel plans
    • Seating charts
    • Lesson plans
    • Get To Know You Activities
    • Recipes

37 of 63

Sample Work-Related Tasks

  • Create an email template to announce information about an upcoming event.
  • Create a memo template explaining to staff about a new process.
  • Develop an action plan to align with district or building goals.
  • Create a visual that aligns with district or building goals.
  • Take a document and develop a slide deck.
  • Affinity group themes in a given text.
  • Develop a social media campaign to promote student attendance.
  • Create a facilitation protocol for a department meeting.

38 of 63

Let’s Practice

  • Practice creating your own and seeing what happens.
  • Remember it is a CHAT so engage in follow up prompts
  • Use resources on website if stuck and need examples

39 of 63

Section 2

Next level techniques and education focused

40 of 63

Zero-Shot vs Few-Shot Learning

41 of 63

Zero-Shot Learning

  • Zero-shot learning: In zero-shot learning, the AI model has not seen any examples of the specific task it's being asked to perform. It relies on its general understanding of language and context to provide relevant responses. This demonstrates the AI's ability to generalize from the training data it has seen and adapt to novel situations.

42 of 63

Few-Shot Learning

  • In few-shot learning, the AI model has only seen a limited number of examples of the task it's being asked to perform. The model can leverage this small amount of information to adapt and provide relevant responses.

43 of 63

Zero-shot vs. few-shot learning

  • Sentiment Analysis:
  • Task: Determine the sentiment of the following text - "I absolutely loved the movie! The visual effects were stunning, and the plot was captivating.”
  • In this case, ChatGPT would need to analyze the text and classify the sentiment as positive, negative, or neutral, without having seen explicit examples of this specific text during training.

Sentiment Analysis:

  • Example 1: Text: "I couldn't stand the terrible acting and weak storyline." Sentiment: Negative
  • Example 2: Text: "The performance was absolutely breathtaking and inspiring." Sentiment: Positive
  • Task: Determine the sentiment of the following text - "The food was mediocre, but the service was fantastic."
  • In this case, ChatGPT is provided with a couple of examples to help guide its understanding of sentiment analysis before being asked to determine the sentiment of a new text.

44 of 63

Zero-shot vs. few-shot learning

Writing a Haiku:

Task: Write a haiku about a sunset.

A haiku is a form of Japanese poetry consisting of three lines with a 5-7-5 syllable pattern. ChatGPT would be expected to compose a haiku on the given topic, even if it hasn't seen specific examples of sunset haikus in its training data.

Writing a Haiku:

  • Example 1: Rain on the window, Soft patter of falling drops, Nature's lullaby.
  • Example 2: Moonlight on the waves, A shimmering sea of stars, Silent night's beauty.
  • Task: Write a haiku about a blooming garden.

By providing two examples of haikus, ChatGPT is given context about the format and structure of the poetry before being asked to create a new haiku on the given topic.

45 of 63

One approach, called Chain of Thought prompting, gives the AI an example of how you want it to reason before you make your request, as you can see in the illustration from the paper.

46 of 63

Prompting Technique

Definition

Quality Prompt Example

Zero-shot Prompting

Zero-shot prompting enables a model to make predictions about previously unseen data without the need for any additional training. It is used to generate natural language text without the need for explicit programming or pre-defined templates​1​.

"Write a short story about a brave astronaut."

One-shot Prompting

One-shot prompting is used to generate natural language text with a limited amount of input data such as a single example or template. This can allow for the creation of predictable outputs from the large language model​2​.

"Given this example 'Once upon a time, there was a brave astronaut who saved the spaceship from disaster', write a short story about a brave astronaut."

Few-shot Prompting

Few-shot prompting is a technique where the model is given a small number of examples, typically between two and five, in order to quickly adapt to new examples of previously seen objects. This technique can allow for the creation of more versatile and adaptive text generation models​3​.

"Given these examples '1. Once upon a time, there was a brave astronaut who saved the spaceship from disaster. 2. The brave astronaut courageously fixed the spaceship's engine, just in time', write a short story about a brave astronaut."

Chain of Thought Prompting

Chain of Thought prompting is a recently developed prompting method that encourages the large language model to explain its reasoning. By showing the model some few-shot exemplars where the reasoning process is explained, the model will also show the reasoning process when answering the prompt. This often leads to more accurate results​4​.

"Given these examples with explanations '1. Once upon a time, there was a brave astronaut who saved the spaceship from disaster. This sentence introduces the protagonist and the conflict. 2. The brave astronaut courageously fixed the spaceship's engine, just in time. This sentence shows the resolution of the conflict', write a short story about a brave astronaut."

Fine-tuning Prompting

Fine-tuning is a technique where a pre-trained model is further trained on a smaller, specific dataset to adapt the model to a particular task. This can be used to improve performance on specific tasks or to instill specific behaviors in the model.

"After fine-tuning the model with a collection of short stories about brave astronauts, prompt the model with 'Write a short story about a brave astronaut'."

47 of 63

PROMPTING TECHNIQUE

DEFINITION

QUALITY PROMPT EXAMPLE

IMPACT ON OUTPUT ACCURACY

Zero-shot Prompting

Zero-shot prompting enables a model to make predictions about previously unseen data without the need for any additional training. It is used to generate natural language text without the need for explicit programming or pre-defined templates

"Solve the equation: 2x + 3 = 7"

Since this is a straightforward problem and language models like GPT-3 have been trained on a variety of mathematical problems, the accuracy of the output would likely be high. However, for more complex problems, zero-shot prompting may not be as effective without additional context.

One-shot Prompting

One-shot prompting is used to generate natural language text with a limited amount of input data such as a single example or template. This can allow for the creation of predictable outputs from the large language model

"Given this example 'Solving the equation 3x + 2 = 8 gives x = 2', solve the equation: 2x + 3 = 7"

Providing a solved example may guide the model to better understand the desired output format, potentially increasing the accuracy of the output. However, it's worth noting that the quality of the example provided is crucial in this case.

Few-shot Prompting

Few-shot prompting is a technique where the model is given a small number of examples, typically between two and five, in order to quickly adapt to new examples of previously seen objects. This technique can allow for the creation of more versatile and adaptive text generation models【

"Given these examples '1. Solving the equation 3x + 2 = 8 gives x = 2. 2. Solving the equation 4x - 2 = 10 gives x = 3', solve the equation: 2x + 3 = 7"

Providing multiple solved examples can improve the model's understanding of the task, possibly leading to more accurate and reliable outputs.

Chain of Thought Prompting

Chain of Thought prompting is a recently developed prompting method that encourages the large language model to explain its reasoning. By showing the model some few-shot exemplars where the reasoning process is explained, the model will also show the reasoning process when answering the prompt. This often leads to more accurate results

"Given these examples with explanations '1. To solve the equation 3x + 2 = 8, subtract 2 from both sides to get 3x = 6, then divide by 3 to get x = 2. 2. To solve the equation 4x - 2 = 10, add 2 to both sides to get 4x = 12, then divide by 4 to get x = 3', solve the equation: 2x + 3 = 7"

As Chain of Thought prompting encourages the model to explain its reasoning, the output will likely contain both the solution and the steps to arrive at the solution. This can result in a better understanding of the problem-solving process and potentially higher accuracy, especially for more complex problems.

Fine-tuning Prompting

Fine-tuning is a technique where a pre-trained model is further trained on a smaller, specific dataset to adapt the model to a particular task. This can be used to improve performance on specific tasks or to instill specific behaviors in the model.

"After fine-tuning the model with a collection of algebraic equations and their solutions, prompt the model with 'Solve the equation: 2x + 3 = 7'."

Fine-tuning the model on a specific task can significantly improve the accuracy of the output for that task. However, this approach requires a relevant dataset for fine-tuning and may limit the model's performance on tasks outside of the specific domain.

48 of 63

Prompt Chaining

  • Break complex tasks into manageable sub-tasks, and work on prompts that are given one "job" to be done. Then, the output of one prompt becomes the input for the next.

49 of 63

AI productivity benefits

Images for presentations

Fill out a form

Rewrite a document

Email Templates

Create new examples, problems, samples

50 of 63

Sample Prompts�From Zain Kahn

Learn any complex topic in just a few minutes:

Prompt: "Explain [insert topic] in simple and easy terms that any beginner can understand.”

Get ChatGPT to write in your style.�Prompt: "Analyze the writing style from the text below and write a 200 word piece on [insert topic]”

Train ChatGPT to generate prompts for you.�Prompt: "You are an AI designed to help [insert profession]. Generate a list of the 10 best prompts for your profession.

Ask ChatGPT to help you become better at using ChatGPT.�Prompt: "Create a beginner's guide to using ChatGPT. Topics should include prompts, priming, and personas. Include examples where necessary. The guide should be no longer than 500 words."lf. The prompts should be about [insert topic].”

Eliminate writer's block.�Prompt: "I'm writing a blog post about [insert topic]. I can't come up with a catchy title. Give me a list of 5 suggestions for the blog title for this piece.”

Generate new ideas.�Prompt: “I want to [insert task or goal]. Generate [insert desired outcome] for [insert task or goal]."

51 of 63

Try the following

Few Shot - give examples to follow

Provide it steps to follow

Incorporate your own ideas with the AI

Ask to explain it’s thinking

Push back

Expand on the second point

Have it ask you questions to better understand what you want

Try different tones, descriptors, roles, etc.

Create additional prompts or ask feedback on assignments

52 of 63

Let’s Practice Again

Try some new interactions

53 of 63

Section 3

AI layered tools

54 of 63

Word of Caution

  • Always check data and privacy use
  • What exists today will most likely be infused in big companies soon
  • Don’t pay for anything!

55 of 63

Let’s Practice Again

Try some tools

56 of 63

Full Circle

  • Take a minute to reflect on your learning journey.
    • How are you feeling?
    • What new insights and/or inquiries do you have?�

57 of 63

What if instead of teaching our students to answers questions, we teach them how to question answers?

58 of 63

Transformative Potential of AI for Educators

  • AI isn't here to replace teachers, but to empower them.
  • AI’s potential in personalized learning.
  • Provide universal access to education, especially for students in remote or disadvantaged areas.
  • AI in data-driven decision making in education policy.

59 of 63

Melanie Borden

…you can’t automate relationships

60 of 63

Examine, Dismantle, Rethink, Redesign

61 of 63

Perhaps a reframe of thinking. Instead of seeing tools as cheating we explore how learning how to prompt is to teach how to��� �

  • Ask Questions
  • Answer Questions
  • Question Answers
  • Question Questions

62 of 63

The future of education is human led, purpose driven, and technology augmented.��Dwayne Matthews

63 of 63

Thank you