Practical AI for Instructors and Students Part 3: Prompting AI

Summary

In this educational video, Professor Ethan Malek and Angelina Malik from the Wharton School introduce viewers to AI, focusing on crafting effective prompts for AI interaction. They emphasize the importance of context, clear instructions, and examples to guide AI in tasks. The video demonstrates step-by-step prompt construction, using tools like GPT4 and Bing, and highlights the iterative process of refining prompts and incorporating human feedback for optimal AI performance in educational settings.

Takeaways

• 📝 **Context Matters**: Telling the AI who it is provides the right context for its responses, making its output more relevant and effective.
• 📐 **Clear Instructions**: Providing clear instructions to the AI is crucial, as it lacks inherent knowledge of the user's context or goals.
• 📚 **Examples and Steps**: Giving the AI examples and step-by-step directions helps it understand the task better and perform more effectively.
• 🔧 **Iterative Process**: Prompting is not a one-time task; it involves experimentation and refinement to achieve the desired outcome.
• 💬 **Conversational Approach**: Interacting with the AI as if it were a conversational partner helps in guiding it towards the desired output.
• 🔄 **Prompting is Part of Interaction**: Creating the perfect prompt is less important than the ongoing interaction with the AI.
• 🎨 **Customization and Constraints**: Adding context, perspective, and constraints to the prompt can guide the AI to produce more tailored content.
• 🔍 **Utilize AI Tools with Internet Access**: Tools like Bing can leverage internet access to include academic research in their responses, enhancing the quality of information.
• 🤖 **Integrate Human Expertise**: Including specific points or knowledge that the user wants to be covered ensures that the AI's output aligns with human expertise.
• 🤔 **Step-by-Step Thinking**: Asking the AI to think and outline its response step by step can lead to more structured and coherent content.
• 🔄 **Iterative Revision**: Regularly revising and providing feedback on the AI's output helps in refining the final result, making the process more collaborative.
• 👨‍🏫 **Act as an Editor and Instructor**: Users should take on the role of an editor and instructor, guiding the AI and shaping its responses to meet their needs.
• 🧙‍♂️ **Magic is Not Real**: The process of working with AI is more about experience and practice rather than following a magical formula.
• 🔄 **Embrace Randomness**: Recognize that AI systems have built-in randomness, and use this to your advantage by experimenting with different prompts and iterations.

Q & A

What is the main focus of the video series presented by Ethan Malek and Angelina Malik?

The main focus of the video series is to provide an introduction to AI, particularly its application in education, and how to effectively use it from a foundational level.

Why is it important to tell the AI who it is when giving instructions?

Telling the AI who it is provides the right context from which to start, as the AI has been trained on billions of documents and needs this context to understand the perspective and environment it is operating in.

What is the significance of giving clear instructions to AI?

Clear instructions are crucial because AI does not know the user's context or intentions. The more context provided, the more effectively the AI can perform the desired task.

Why should examples and steps be given to AI when assigning a specific project?

Examples and steps help the AI understand what the user wants and think step by step, leading to a more effective performance in completing the task.

What are the two important caveats mentioned about constructing prompts for AI?

The two caveats are: 1) Prompting isn't magic and is about experience, with each generation of AI models making it easier. 2) Prompting is just a part of the interaction with AI, which is more important than getting the perfect prompt.

How can AI models like GPT-4 and chat GPT be used to improve writing?

AI models can be used to generate initial drafts, provide context and perspective, and follow specific instructions or constraints to enhance the quality and style of writing.

What is the advantage of using Bing in creative mode over GPT-4 for certain tasks?

Bing in creative mode has access to the internet, allowing it to look up academic research or additional information that GPT-4, which is not connected to the internet, cannot access.

How can including oneself in the AI model improve the output?

By adding specific points or instructions that one wants the AI to include, the output becomes more integrated with human input, leading to a more personalized and targeted result.

What is the 'Chain of Thought' thinking and how can it be used with AI?

Chain of Thought thinking is a method where the AI is asked to think step by step, outlining and proving its points before writing the final content. This helps the AI stay on task and allows for better output quality.

Why is it recommended to interact with AI interactively rather than trying to create the perfect prompt?

Interactively working with AI allows for continuous feedback and guidance, which vastly improves the work. It is more effective than a single perfect prompt because it adapts to the evolving needs and preferences of the user.

How does the randomness in AI systems affect the user's experience?

The randomness built into AI systems means that different prompts can yield different outputs, even from the same system at different times. This can be advantageous as it allows users to select the most suitable response or to refresh and try again for better results.

What role does the user play in the process of using AI for writing tasks?

The user acts as a guide, editor, and instructor, taking the lead in shaping the AI's output by providing feedback, selecting preferred content, and giving directions for improvement.

Outlines

00:00

🤖 Effective Prompting Techniques for AI in Education

This paragraph introduces the topic of the video, which is about teaching AI how to assist effectively in educational settings. The speaker, Ethan Malek, a professor at Wharton, and his colleague Angelina Malik, discuss the importance of context when instructing AI and the necessity of clear, example-driven instructions. They emphasize that AI needs to understand who it is and what is expected of it to perform tasks effectively. The paragraph also introduces the concept of 'prompting' as a skill that improves with experience and experimentation rather than following a rigid formula. It sets the stage for a step-by-step guide on constructing prompts for AI.

05:02

🔍 Enhancing AI Prompts with Context and Constraints

The second paragraph delves into the specifics of constructing effective prompts for AI, using the example of writing a paragraph about the importance of group work. It illustrates how adding context and perspective can significantly alter the AI's output. The paragraph also discusses the importance of providing examples and steps to guide the AI in a specific direction. The use of advanced techniques such as adding constraints and instructing the AI to think step by step is highlighted, showing how these methods can lead to more sophisticated and tailored outputs. The paragraph concludes with a demonstration of how to refine prompts interactively, emphasizing the iterative nature of working with AI to achieve the desired results.

10:03

👨‍🏫 Integrating Human Expertise with AI in Content Creation

This paragraph explores the integration of human expertise with AI-generated content, focusing on the collaborative approach to content creation. It discusses the use of AI tools like Bing, which can access the internet to incorporate academic research into its responses. The paragraph demonstrates how to insert specific points known by the user into the AI's output to ensure accuracy and relevance. It also introduces the concept of 'Chain of Thought' prompting, which encourages the AI to outline and revise its work methodically. The importance of human interaction in guiding the AI, refining its output, and avoiding getting stuck in loops is emphasized, highlighting the role of the user as an active participant in the content creation process.

Keywords

💡AI

AI, or Artificial Intelligence, refers to the simulation of human intelligence in machines that are programmed to think like humans and mimic their actions. In the context of the video, AI is the central theme, with a focus on how it can be utilized in education. The script discusses various ways to interact with AI, emphasizing the importance of clear instructions and contextual information to guide its responses effectively.

💡Interactive pedagogy

Interactive pedagogy is an educational approach that emphasizes the use of technology to facilitate interactive learning experiences. In the video, the concept is mentioned in relation to the study of AI conducted by the speakers, suggesting that AI can be a powerful tool in creating engaging and interactive educational content.

💡Prompting

Prompting, in the context of AI, involves giving the AI clear and specific instructions to guide its output. The video script discusses the art of crafting effective prompts to direct the AI in producing desired results. It highlights the iterative process of refining prompts and the importance of examples and steps in this process.

💡Context

Context is crucial when interacting with AI, as it helps the AI understand the situation or scenario it is being asked to address. The script mentions the importance of providing context to the AI, such as identifying oneself or the subject matter, to ensure that the AI's responses are relevant and accurate.

💡GPT4

GPT4, or Generative Pre-trained Transformer 4, is a type of AI model discussed in the video. It is part of a series of foundational models that the speakers use to demonstrate how to craft prompts and interact with AI. The script shows how GPT4 can be directed to write a paragraph with specific instructions and style.

💡Bing

Bing, in the script, refers to Microsoft's search engine, which has a creative mode that utilizes the same AI model as GPT4 but with an added feature of internet connectivity. The video demonstrates how Bing can be used to look up academic research on a given topic, enhancing the AI's responses with current and relevant information.

💡Academic research

Academic research is the systematic investigation and study of a particular subject to establish facts and draw conclusions. In the video, the AI is prompted to look up academic research on group work, showing how AI can be used to incorporate scholarly insights into its responses.

💡Centaur

The term 'centaur' is used metaphorically in the video to describe the collaborative relationship between humans and AI, where both contribute to the creation of content. The script illustrates this concept by showing how the AI can be guided to include specific points known by the human user, creating a hybrid of human and AI-generated content.

💡Chain of Thought

Chain of Thought is a technique mentioned in the video where the AI is asked to think step by step, outlining and then proving its points before writing the final content. This method helps the AI stay focused on the task and allows for better quality control by the user.

💡Randomness

Randomness in AI refers to the variability in the outputs generated by the system, even when given the same prompts. The script notes that AI systems are designed to produce different responses over time, which can be used to the user's advantage by experimenting with prompts and selecting the most suitable output.

Highlights

Introduction to the importance of AI in education and the focus on interactive pedagogy.

The necessity of informing AI about its identity to provide the right context for responses.

Clear instructions are crucial for AI effectiveness due to its lack of inherent understanding of user context.

The value of providing examples and steps to guide AI in specific projects or directions.

The process of constructing a prompt step by step to optimize AI responses.

The importance of recognizing that prompting is not a magical formula but a skill that improves with experience.

The concept of experimenting with AI and asking for its help in task accomplishment.

The role of context and perspective in shaping AI responses, illustrated with the example of writing about group work.

Adding constraints to AI prompts to achieve more tailored and engaging content.

The use of AI's style analysis to inform constraints and improve writing quality.

Incorporating academic research into AI responses using tools with internet access, like Bing.

The integration of human expertise into AI-generated content to ensure accuracy and relevance.

Techniques for prompting AI to think step by step, enhancing the quality of its output.

The interactive nature of working with AI, emphasizing the importance of human guidance and feedback.

The role of the AI user as an editor, selecting and refining content to meet specific needs.

The inherent randomness in AI responses and the strategy of using multiple prompts for diverse outputs.

The art of prompting as a skill developed through practice and experimentation.

Upcoming videos will explore the application of AI in classroom settings for both teachers and students.

Transcripts

[Music]

hi welcome to video three of our series

of videos where we're trying to give you

an introduction to Ai and how to use it

with a special focus on education from

the ground up I'm Ethan Malek a

professor at Wharton who has been

studying AI

and Angelina Malik and I study Ai and

interactive pedagogy at the Wharton

School of the University of Pennsylvania

in this video we're going to talk to you

about how to prompt and how to tell the

AI exactly what you'd like it to do and

how to do so effectively

thank you

the first thing to remember is that you

want to tell the AI who it is the reason

you want to do that is because the AI

has been trained on billions of

documents telling it who it is gives it

the right context from which to start

once you tell the AI who it is the

second thing to do is to give the AI

very clear instructions about what you

wanted to do you have to remember that

the AI doesn't know you and doesn't know

any of your contexts so the more context

you give it the more effective it can be

the third thing to remember is to always

give AI examples and steps if you have a

specific project you'd like it to take

on if you have specific directions that

you'd like to give it giving it examples

helps it learn what you want and giving

it steps helps it think step by step and

we'll do a more effective job for you so

let's see how this works in practice

so I'm going to show you how to

construct a prompt sort of Step by Step

there are two important caveats that you

have to know before we do this first off

don't follow this exactly prompting

isn't Magic eighty percent of prompting

is about experience and prompting is

just getting easier with each generation

of these models so this is not a magical

formula you can actually experiment and

the more you use AI the more you'll find

out what's good or bad and how to make

it do what you'd like and even just ask

the AI to help you say I want to

accomplish this task what should I ask

you it will help you with that so that's

the first thing is this isn't magical

there is not a technical programming

language here this is a conversation

with the AI and the second is as you

will talk about it a little bit

prompting is just a piece of this so

getting the right perfect prompt is less

important than interacting with the AI

which we'll talk to you after we show

you how to create a problem so with that

said let's talk about how to create a

good first prompt so let's work out a

prompt together and see how we can make

it better we're going to start with

something basic and build it out for

this particular use case we're going to

use gpt4 and chat GPT which as we

discussed last time is one of a series

of foundational models you're going to

use so you could also do this in

Microsoft Bing which in its creative

mode uses chat GPT for the same model or

you can use it Google's Bard which works

a little bit differently but we're going

to just use gpt4 for this again

prompting Works fairly similarly across

different AI models so let's start with

an example let's say we wanted to write

me a paragraph

about

why group work

is important

and so gpd4 is a very powerful model

it's a little bit slow it will produce

the results here that we need here about

this so that is a good prompt and it's

coming up with a perfectly fine essay

but you'll notice it doesn't know what

context to use what perspective to write

this from

so the first thing we can start to do is

we can add context and perspective so we

can say

act like

an experienced teacher

write a paragraph about why group work

is important and you'll notice that the

entire context of the conversation

changes because now it's like a pedagogy

and it's talking about from a teacher's

perspective we haven't made it into a

teacher and it's not magic but we've

given it context to work in and a

perspective to take and you'll see it

uses that perspective throughout now we

can make this even more advanced by

adding in constraints by telling it

exactly what we want to see so we can

say that act like experienced teacher

write a paragraph about why group Works

important make their writing engaging

but smart to pick a style and by the way

if you don't know what style to use you

can paste in your favorite text into any

of these AIS and ask it to describe that

text to you and then use that

description when you're giving a

constraints we might also want to say

make sure there is

an example and again you can see I'm

just using normal human text to do this

there's nothing magical here and say now

it's being a little bit too engaging and

fun right same group work my deal

Scholars is a vital Cornerstone of

education so maybe you think that's too

flowery we're learning as we go with

this prompt we'll talk more about how

you revise it as you go along but you

can see we get a better more constrained

more interesting piece of writing as a

result of this

so now we can do something a little bit

different

for this example we're going to switch

over to a tool that has access to the

internet which is Bing Bing is using in

creative mode purple mode the exact same

model as chat gpt's gpt4 but it has an

internet connection so here we're going

to paste in the exact same thing that we

had before

and we're going to add a additional

qualifier we're going to say

look up the academic research on group

work

now this would not work in GPT it's not

connect to the internet but it works

here

well in bing so let's try this with look

up the academic research work and what

you'll see is Bing is now going to

actually do that for us here so it's now

looking up doing actual web search to

find the answer and we're going to get a

better result you'll also notice that

Bing will provide sources as a result

these sources are not always 100 correct

you'll want to again check the work to

make sure there's no hallucinations but

I think you can see how we get better

results now here that are both including

the example including the context but

now including actual academic research

we can go one step further with this by

adding ourselves into this model we

wanted to make particular points we're

smart we're experts we don't want the AI

to be the expert so we can tell it make

the following points

[Music]

so now to make the AI Not Just Produce

random work we want to actually insert

what we know into the prompt and we give

it instructions that are more explicit

so in this case we wanted to make the

following points that group work allows

students to support each other but group

work also needs to be carefully designed

so up till now there's been none of us

in the AI which is of course kind of

boring because you'd like to make sure

that as humans we're in the loop that we

are making playing a role in the

creation of this content so it's really

important to add that in so we're going

to add in a couple of points that we

know that we want to make sure the AI

covers so it doesn't just make up its

own information we're going to say that

we wanted to include that group work

allows students to support each other

and the group work needs to be carefully

designed and then when we do this prompt

the AI will do what it did before which

is look up the information like we

requested it'll still do the work it did

before from the context of a teacher

with the same tone but you'll notice now

including the points that we wanted to

include as opposed to just being AI work

now it's starting to be integrated work

between human and AI sometimes it's

called being a centaur half person half

horse and that's what we're seeing here

there's one even more advanced technique

that we can include which research shows

makes the am more effective and this is

a little bit trickier but it is asking

the AI to think step by step so remember

the AI is just trying to guess what the

next word is in a series of words and I

ask you to think step by step allows it

to create content that it can then look

at and build from so we could say follow

these steps outline the paragraph

prove the outline

write the paragraph so now we're giving

it steps to follow this step-by-step

thinking sometimes called Chain of

Thought thinking is a useful way to get

the AI to stay on task and also for us

to check the output as we go along to

make sure that it's good so you can see

it's outlining the parallel graph it's

going to do first it's giving us the

sentences it's going to make the points

it's going to make the example it's

going to include which we all asked for

and now it's going to improve the

outline so it's going to go back and you

can see it's going to actually already

revising what it's doing so asking it to

look back at its own work often gets

much better output as a result and you

can see it's actually doing a better

transition it's doing better points and

it's going to then produce the final

content here which is the actual end

goal that we wanted and so you can see

you have the sophistication level

increases the quality level increases

and we can check the results to see what

goes wrong if anything does

so I've just spent a bunch of time

showing you how to do a prompt but this

is actually not the right way to

optimally use the AI you don't want to

try and create the world's perfect

prompt instead you want to work with it

interactively and to work with it

interactively you have to push back and

this is really where the magic happens

for instance you could ask it to expand

I like a second Point can you expand on

that

and you can ask it to edit I don't like

the example you used in this paragraph

why don't you change the example

in addition you can also ask it to

include more of your thinking you can

add a point about how group work

requires you as a teacher to choose

whether teams have the same ability

level

and all of this including your own

thinking your own interaction improves

the paragraph if you look back at the

beginning of the paragraph to the final

output you can see that you in the loop

guiding the AI you being in charge and

giving a direction vastly improves the

work

[Music]

sometimes the AI can get into a loop you

may need to clear the memory by hitting

for instance in bing on new topic where

the broom is and just start again each

time you prompt the AI it'll give you a

different output and so you have to take

the lead thank you

so because you have to take the lead

that means you have to do many different

kinds of tasks that aren't normally part

of writing not only are you thinking

about whether the writing is good or bad

but you're also kind of acting as an

editor you're selecting what work you

like you're figuring out whether it's

hey in the right direction you're even

instructing the AI like a teacher which

is part of our teachers are often very

good at working with AI and prompting it

to do things so

an important thing to note about all

this is the AI will give you different

answers every time there's Randomness

built into all of these systems and so

when you give it one prompt you may get

different outputs from different systems

at different times or even the same

system in different ways it's something

you actually take advantage of so you

could say something like give me five

examples of a particular paragraph and

it will give you five different examples

and you can pick the one that you like

you could tell them doing very different

styles and get very different outcomes

but sometimes you'll have to hit refresh

clear the memory on the system and try

again so it's about multiple efforts

multiple attempts to try the same prompt

multiple experiments with you giving

feedback to get a prompt response that

you like so prompting becomes more art

than science which brings us back to the

point of experience this is not magic it

is about practicing until you get good

at this it's about trying the AI for the

tasks that you want to use it for it's

about treating it a bit like a human

again even though it's not a human and

giving it feedback and instructions

about what you want to see have happen

in the next set of videos we're going to

dive into how to use this in a classroom

both as a teacher and as a student

[Music]