Generative AI

What IS Artifical Intelligence?

Intelligence that is not real

Any technology that uses electricity

When computers can do things that usually require human intelligence

Can you think of any examples of AI that you might have interacted with today?

Mario Kart uses AI to control the behavior of computer-controlled opponents and to manage in-game events such as power-ups and obstacles.

Smart speakers and other smart home devices can be controlled using voice commands, which are powered by AI-powered virtual assistants.

Streaming services like Netflix and Disney+ use AI algorithms to recommend shows and movies based on a user's viewing history.

Smartphone cameras use AI to improve the quality of pictures. AI helps the camera analyze the scene by recognizing objects, people, and the lighting conditions. Based on this analysis, it makes automatic adjustments to settings like focus, exposure, and color balance to capture the best possible image.

Captchas are puzzles on the internet that separate humans from robots. Artificial Intelligence (AI) is a smart computer program that can solve these puzzles quickly. By using AI, websites ensure that real people are using them, keeping the internet safe.

AI uses signals from satellites and lots of other data, like maps and traffic conditions, to calculate the best route for us. It can even predict how long it will take to reach our destination based on real-time data.

Artificial Intelligence (AI) is everywhere around us. It is used in various applications, from simple tasks like recommending products on e-commerce websites to complex tasks like self-driving cars.

Generative AI focuses on generating new data from the data it has seen.

Add Image of Molecular Model- Covid image?

Today we are going to focus on generative AI that can generate text and images.

Generative Text

Where are you?

Have you ever used predictive text on a phone? How do you think it works?

Hi! What are you doing?

park

store

library

I'm at the

Hey!Not much. I'm just walking the

path

road

dog

library

park

store

path

road

dog

AI uses language models to suggest what word or phrase might come next as you type.

Let's imagine we are training the AI so it can develop its language model.

Predict the next word.

What happens when you train a language model on all of the text that exists on the internet?

What happens when you train a language model on all of the text that exists on the internet?

You get a large language model!

LARGE

Large

language models use lots and lots of data from the internetto predict the most likely next word.

quick

big

apple

goose

brown

rabbit

little

and

Most likely next word

Most likely next word

fox

dog

rabbit

monkey

Most likely next word

jumped

stepped

leaped

ran

...

Most likely next word

The

GPT (generative pre-trained transformer) is a class of Large Language Model that's really good at writing.

There are two kinds of prompting.

Talk to the chat bot.

Option 2 Conversation

Option 1: Autocomplete

Complete my text.

How can we direct GPT to generate text that aligns with our desired style and preferences? We use something called a prompt.

Let's write a story!

Talk to the chat bot.

Option 2 Conversation

Option 1: Autocomplete

Complete my text.

Let's write some prompts!

Let's write a story!

Option 1: Autocomplete

Talk to the chat bot.

Option 2 Conversation

Complete my text.

Please write me a story about an ice cream truck that visited a school.

The ice cream truck pulled up to the school and...

Create an example of each type for the kids to reference.

Not only can AI generate text.

AI can also generate images!

Let's play a game!

Did AI create this image?

Select the image to find out!

This image was generated by AI.

Did AI create this image?

This picture was taken by a person.

Did AI create this image?

This picture was taken by a person.

Did AI create this image?

This image was generated by AI.

Did AI create this image?

This picture was taken by a person.

Did AI create this image?

This image was generated by AI.

How does AI learn to generate images?

GAN: Generative Adversarial Network

Discriminator

Tries to distinguish whether the example is from training data or was generated by GAN

Generator

AI that generates examples similar to the training data

GAN sends examples from training data and examples it generated.

Example

How does AI learn to generate images?

Generate an image of a puppy that will fool the discriminator.

Discriminator

Tries to distinguish whether the example is from training data or was generated by GAN

GAN

AI that generates examples similar to the training data

How does AI learn to generate images?

Discriminator

Tries to distinguish whether the example is from training data or was generated by GAN

GAN

AI that generates examples similar to the training data

How does AI learn to generate images?

Hmm...I think that came from the training data.

Discriminator

Tries to distinguish whether the example is from training data or was generated by GAN

GAN

AI that generates examples similar to the training data

How does AI learn to generate images?

Hmm...I think that came from the training data.

Nope! I generated it.

Generate image of puppy.

Discriminator

Tries to distinguish whether the example is from training data or was generated by GAN

GAN

AI that generates examples similar to the training data

How does AI learn to generate images?

Ok! Good to know. I'll remember that and learn from it.

Nope! I generated it.

Generate image of puppy.

Discriminator

Tries to distinguish whether the example is from training data or was generated by GAN

GAN

AI that generates examples similar to the training data

Diffusion Model

The diffusion model works by adding noise to a training data image, then learns how to remove that noise.

Diffusion Model

The diffusion model works by adding noise to a training data image, then learns how to remove that noise.

Diffusion Model

The diffusion model works by adding noise to a training data image, then learns how to remove that noise.

Look for examples that show static added to image.

Let's illustrate our stories!

Large language models use lots and lots of data from the internetto make predictions about what words or phrases might come next.

What might be some pros and cons of using the internet to develop large language models?Type your ideas on the post-it notes.

Cons

Pros

Justin- Do we want this after the text section?do we want a pros cons concerning using the internet for generating just text or for all kinds of generative AI?Also, do we want to include a page that disucsses the pros and cons of using generative AI in general (for example, it might help people cheat or cause people to become less creative).

What did we learn? These models are trained on large scale data from the internet. These models can produce content at the level that a human could produce. The models don't always produce what you want --> Prompt writing is working with the AI to create What could go wrong?

Cons: Offensive content: It's possible for large language models to access offsensive content on the internet. Since it is learning, it may not always realize that the content should not be used. Bias: The data on the internet may not be representative of the whole population, and can reflect biases of the people who create and share it. This can result in large language models that also contain these biases. Data privacy: To train on internet data, large language models must collect and store large amounts of data from the internet. This raises concerns about data privacy and who has access to the data.

Pros and cons of using the internet to develop large language models.

Pros

Cons

Pros: Large amounts of data: Using the internet as a source for training data allows large language models to access an enormous amount of information. This is important because large amounts of data are necessary to build accurate language models. Diverse data: The internet is home to a wide variety of languages, dialects, and writing styles. By training on data from the internet, the large language model can develop language models that can handle a broad range of inputs. Continuous learning: The large language model can be constantly updated with new data from the internet, allowing it to adapt and improve over time.

Cons: Offensive content: It's possible for large language models to access offsensive content on the internet. Since it is learning, it may not always realize that the content should not be used. Bias: The data on the internet may not be representative of the whole population, and can reflect biases of the people who create and share it. This can result in large language models that also contain these biases. Data privacy: To train on internet data, large language models must collect and store large amounts of data from the internet. This raises concerns about data privacy and who has access to the data.

Next

Pros and cons of using the internet to develop large language models.

Pros

Cons

Pros: Large amounts of data: Using the internet as a source for training data allows large language models to access an enormous amount of information. This is important because large amounts of data are necessary to build accurate language models. Diverse data: The internet is home to a wide variety of languages, dialects, and writing styles. By training on data from the internet, the large language model can develop language models that can handle a broad range of inputs. Continuous learning: The large language model can be constantly updated with new data from the internet, allowing it to adapt and improve over time.

Cons: Offensive content: It's possible for large language models to access offsensive content on the internet. Since it is learning, it may not always realize that the content should not be used. Bias: The data on the internet may not be representative of the whole population, and can reflect biases of the people who create and share it. This can result in large language models that also contain these biases. Data privacy: To train on internet data, large language models must collect and store large amounts of data from the internet. This raises concerns about data privacy and who has access to the data.

Next

Do we want to include this anywhere? I have tried to get more specific with prompts and AI really just doesn't get banana splits.

"Eating a banana split"

Next

Talk about Prompts: Prompts are the way humans specify their intent to get the model to produce something. For example, you can provide a prompt like "a red apple on a table" to a text-to-image model, and it will generate an image of a red apple on a table. Play with a Text-to-Image Model: Let's try using a text-to-image model like Stable Diffusion or DALL-E to generate some images. You can provide a prompt, and the model will generate an image based on it. Maybe you can generate some images to go with the story you wrote earlier.

Generative AI and Large Language Models

1. WHAT: How AI is used now and ask how it can be used in the future
2. Introduce that machines now have the ability to use similar process that humans use to create
1. Show a picture that was created by generative AI
2. Explain neural networks (how detailed?) and generative AI
3. All students to simulate generative AI (how should this look on the interactive?)
4. Play a game where students guess if a work was created by a human or AI
3. Introduce Large Language Models
1. Predictive text example
2. label data
3. Students simulate
1. Categorization of word activity?
2. Writing prompts that include a style

4. HOW: Neural networks4. The possibilities

1. Ways AI may be used in the future (provided examples and kids answer)
2. Limitations of AI: There is no reasoning
1. Examples of how AI did not have the context that a human would have, banna split
2. It sounds confident but is wrong
3. biases- some languages don't have he/she, teachers are always female
4. I just makes up stuff (hallicinations) and comes off as very confident
3. We do not want AI to replace what we do but to empower us as we make decisions, we remain in control
1. Examples of art created on midjourney- show the process of creating the final product through AI

Wrap-up: What do you think about the future of generative AI? Some people are afraid that these technologies will take over their jobs, while others believe that new jobs will be created. What are your thoughts on this?

Why is AI significant...why is it important for kids to learn about AI? How does it compare to other technological advances throughout history. Artificial Intelligence (AI) is significant because it has the potential to transform the way we live, work, and interact with each other. By using algorithms and computational power, AI can automate tasks, analyze large amounts of data, and even learn from experience. This can lead to new discoveries, better decision-making, and increased efficiency in various fields, from healthcare to finance to transportation. AI is important for kids to learn about because it will likely play an increasingly important role in their future. As AI becomes more prevalent in society, it will create new job opportunities and require new skills. By understanding the basics of AI, kids can better prepare themselves for the future workforce and even create their own innovative solutions using AI. AI is different from other technological advances throughout history because it involves the development of intelligent machines that can perceive their environment, reason, and learn. While previous technological advances such as the invention of the steam engine or the telephone have revolutionized industries, AI has the potential to change the very nature of work and human interaction. It also raises ethical and societal questions, such as the impact on employment, privacy, and bias. It is important for kids to learn about these issues so they can be prepared to make informed decisions as citizens and creators in a world increasingly shaped by AI.