Topics Ready...

And welcome to Introduction to Generative AI. (00:01)

My name is Dr. Gwendolyn Stripling. (00:04)

And I am the artificial intelligence technical curriculum developer here at Google Cloud. (00:06)

In this course, you learn to define generative AI, explain how generative AI works, describe generative AI model types, and describe generative AI applications. (00:14)

Generative AI is a type of artificial intelligence technology that can produce various types of content, including text, imagery, audio, and synthetic data. (00:28)

But what is artificial intelligence? (00:41)

Well, since we are going to explore generative artificial intelligence, let's provide a bit of context. (00:44)

So two very common questions asked are what is artificial intelligence and what is the difference between AI and machine learning. (00:51)

One way to think about it is that AI is a discipline, like physics for example. (01:01)

AI is a branch of computer science that deals with the creation of intelligence agents, which are systems that can reason, and learn, and act autonomously. (01:08)

Essentially, AI has to do with the theory and methods to build machines that think and act like humans. (01:20)

In this discipline, we have machine learning, which is a subfield of AI. (01:30)

It is a program or system that trains a model from input data. (01:35)

That trained model can make useful predictions from new or never before seen data drawn from the same one used to train the model. (01:40)

Machine learning gives the computer the ability to learn without explicit programming. (01:49)

Two of the most common classes of machine learning models are unsupervised and supervised ML models. (01:56)

The key difference between the two is that, with supervised models, we have labels. (02:03)

Labeled data is data that comes with a tag like a name, a type, or a number. (02:09)

Unlabeled data is data that comes with no tag. (02:16)

This graph is an example of the problem that a supervised model might try to solve. (02:20)

For example, let's say you are the owner of a restaurant. (02:26)

You have historical data of the bill amount and how much different people tipped based on order type and whether it was picked up or delivered. (02:29)

In supervised learning, the model learns from past examples to predict future values, in this case tips. (02:39)

So here the model uses the total bill amount to predict the future tip amount based on whether an order was picked up or delivered. (02:47)

This is an example of the problem that an unsupervised model might try to solve. (02:57)

So here you want to look at tenure and income and then group or cluster employees to see whether someone is on the fast track. (03:02)

Unsupervised problems are all about discovery, about looking at the raw data and seeing if it naturally falls into groups. (03:11)

Let's get a little deeper and show this graphically as understanding these concepts are the foundation for your understanding of generative AI. (03:21)

In supervised learning, testing data values or x are input into the model. (03:31)

The model outputs a prediction and compares that prediction to the training data used to train the model. (03:37)

If the predicted test data values and actual training data values are far apart, that's called error. (03:45)

And the model tries to reduce this error until the predicted and actual values are closer together. (03:54)

This is a classic optimization problem. (04:01)

Now that we've explored the difference between artificial intelligence and machine learning, and supervised and (04:05)

unsupervised learning, let's briefly explore where deep learning fits as a subset of machine learning methods. (04:11)

While machine learning is a broad field that encompasses many different techniques, deep learning is a type (04:20)

of machine learning that uses artificial neural networks, allowing them to process more complex patterns than machine learning. (04:25)

Artificial neural networks are inspired by the human brain. (04:34)

They are made up of many interconnected nodes or neurons that can learn to perform tasks by processing data and making predictions. (04:37)

Deep learning models typically have many layers of neurons, which allows them to learn more complex patterns than traditional machine learning models. (04:47)

And neural networks can use both labeled and unlabeled data. (04:56)

This is called semi-supervised learning. (05:00)

In semi-supervised learning, a neural network is trained on a small amount of labeled data and a large amount of unlabeled data. (05:03)

The labeled data helps the neural network to learn the basic concepts of the task while the unlabeled data helps the neural network to generalize to new examples. (05:12)

Now we finally get to where generative AI fits into this AI discipline. (05:24)

Gen AI is a subset of deep learning, which means it uses artificial neural networks, can process both labeled and unlabeled data using supervised, unsupervised, and semi-supervised methods. (05:30)

Large language models are also a subset of deep learning. (05:45)

Deep learning models, or machine learning models in general, can be divided into two types, generative and discriminative. (05:51)

A discriminative model is a type of model that is used to classify or predict labels for data points. (05:59)

Discriminative models are typically trained on a data set of labeled data points. (06:06)

And they learn the relationship between the features of the data points and the labels. (06:10)

Once a discriminative model is trained, it can be used to predict the label for new data points. (06:17)

A generative model generates new data instances based on a learned probability distribution of existing data. (06:25)

Thus generative models generate new content. (06:33)

Take this example here. (06:38)

The discriminative model learns the conditional probability distribution or the probability of y, our output, given x, (06:40)

our input, that this is a dog and classifies it as a dog and not a cat. (06:46)

The generative model learns the joint probability distribution or the probability of x and y and predicts (06:54)

the conditional probability that this is a dog and can then generate a picture of a dog. (07:02)

So to summarize, generative models can generate new data instances while discriminative models discriminate between different kinds of data instances. (07:09)

The top image shows a traditional machine learning model which attempts to learn the relationship between the data and the label, or what you want to predict. (07:21)

The bottom image shows a generative AI model which attempts to learn patterns on content so that it can generate new content. (07:30)

A good way to distinguish what is gen AI and what is not is shown in this illustration. (07:40)

It is not gen AI when the output, or y, or label is a number or a class, for example spam or not spam, or a probability. (07:46)

It is gen AI when the output is natural language, like speech or text, an image or audio, for example. (07:57)

Visualizing this mathematically would look like this. (08:08)

If you haven't seen this for a while, the y is equal to f of x equation calculates the dependent output of a process given different inputs. (08:12)

The y stands for the model output. (08:23)

The f embodies the function used in the calculation. (08:25)

And the x represents the input or inputs used for the formula. (08:29)

So the model output is a function of all the inputs. (08:33)

If the y is the number, like predicted sales, it is not gen AI. (08:36)

If y is a sentence, like define sales, it is generative as the question would elicit a text response. (08:43)

The response would be based on all the massive large data the model was already trained on. (08:51)

To summarize at a high level, the traditional, classical supervised and unsupervised learning process takes training code and label data to build a model. (08:59)

Depending on the use case or problem, the model can give you a prediction. (09:09)

It can classify something or cluster something. (09:15)

We use this example to show you how much more robust the gen AI process is. (09:18)

The gen AI process can take training code, label data, and unlabeled data of all data types and build a foundation model. (09:25)

The foundation model can then generate new content. (09:33)

For example, text, code, images, audio, video, et cetera. (09:36)

We've come a long away from traditional programming to neural networks to generative models. (09:42)

In traditional programming, we used to have to hard code the rules for distinguishing a (09:48)

cat-- the type, animal; legs, four; ears, two; fur, yes; likes yarn and catnip. In the (09:52)

wave of neural networks, we could give the network pictures of cats and dogs and ask (10:03)

is this a cat and it would predict a cat. In the generative wave, we as (10:07)

users can generate our own content, whether it be text, images, audio, video, et cetera, for (10:14)

example models like PaLM or Pathways Language Model, or LaMDA, Language Model for Dialogue Applications, ingest (10:21)

very, very large data from the multiple sources across the internet and build foundation language models (10:30)

we can use simply by asking a question, whether typing it into a prompt or verbally (10:36)

talking into the prompt itself. So when you ask it what's a cat, it can give (10:43)

you everything it has learned about a cat. Now we come to our formal definition. What (10:48)

is generative AI? Gen AI is a type of artificial intelligence that creates new content based (10:55)

on what it has learned from existing content. The process of learning from existing content is (11:00)

called training and results in the creation of a statistical model when given a prompt. AI (11:07)

uses the model to predict what an expected response might be and this generates new content. (11:13)

Essentially, it learns the underlying structure of the data and can then generate new samples that (11:21)

are similar to the data it was trained on. As previously mentioned, a generative language model (11:27)

can take what it has learned from the examples it's been shown and create something entirely (11:35)

new based on that information. Large language models are one type of generative AI since they (11:40)

generate novel combinations of text in the form of natural sounding language. A generative image model (11:48)

takes an image as input and can output text, another image, or video. For example, under (11:56)

the output text, you can get visual question answering while under output image, an image completion (12:04)

is generated. And under output video, animation is generated. A generative language model takes text as (12:12)

input and can output more text, an image, audio, or decisions. For example, under the output (12:21)

text, question answering is generated. And under output image, a video is generated. We've stated that (12:28)

generative language models learn about patterns and language through training data, then, given some text, they (12:36)

predict what comes next. Thus generative language models are pattern matching systems. They learn about patterns (12:43)

based on the data you provide. Here is an example. Based on things it's learned from (12:51)

its training data, it offers predictions of how to complete this sentence, I'm making a sandwich (12:58)

with peanut butter and jelly. Here is the same example using Bard, which is trained on (13:05)

a massive amount of text data and is able to communicate and generate humanlike text in (13:12)

response to a wide range of prompts and questions. Here is another example. The meaning of (13:17)

life is-- and Bart gives you a contextual answer and then shows the highest probability response. (13:26)

The power of generative AI comes from the use of transformers. (13:35)

Transformers produced a 2018 revolution in natural language processing. (13:40)

At a high level, a transformer model consists of an encoder and decoder. (13:45)

The encoder encodes the input sequence and passes it to the decoder, which learns how to decode the representation for a relevant task. (13:50)

In transformers, hallucinations are words or phrases that are generated by the model that are often nonsensical or grammatically incorrect. (14:01)

Hallucinations can be caused by a number of factors, including the model is not trained on enough data, or the model is (14:13)

trained on noisy or dirty data, or the model is not given enough context, or the model is not given enough constraints. (14:22)

Hallucinations can be a problem for transformers because they can make the output text difficult to understand. (14:33)

They can also make the model more likely to generate incorrect or misleading information. (14:40)

A prompt is a short piece of text that is given to the large language model as input. (14:46)

And it can be used to control the output of the model in a variety of ways. (14:53)

Prompt design is the process of creating a prompt that will generate the desired output from a large language model. (14:59)

As previously mentioned, gen AI depends a lot on the training data that you have fed into it. (15:07)

And it analyzes the patterns and structures of the input data and thus learns. (15:14)

But with access to a browser based prompt, you, the user, can generate your own content. (15:20)

We've shown illustrations of the types of input based upon data. (15:27)

Here are the associated model types. (15:31)

Text-to-text. (15:33)

Text-to-text models take a natural language input and produces a text output. (15:35)

These models are trained to learn the mapping between a pair of text, e.g. for example, translation from one language to another. (15:40)

Text-to-image. (15:49)

Text-to-image models are trained on a large set of images, each captioned with a short text description. (15:50)

Diffusion is one method used to achieve this. (15:58)

Text-to-video and text-to-3D. (16:01)

Text-to-video models aim to generate a video representation from text input. (16:04)

The input text can be anything from a single sentence to a full script. (16:09)

And the output is a video that corresponds to the input text. (16:15)

Similarly, text-to-3D models generate three dimensional objects that correspond to a user's text description. (16:20)

For example, this can be used in games or other 3D worlds. (16:29)

Text-to-task. (16:34)

Text-to-task models are trained to perform a defined task or action based on text input. (16:36)

This task can be a wide range of actions such as answering a question, performing a search, making a prediction, or taking some sort of action. (16:44)

For example, a text-to-task model could be trained to navigate a web UI or make changes to a doc through the GUI. (16:55)

A foundation model is a large AI model pre-trained on a vast quantity of data designed to be adapted (17:05)

or fine tuned to a wide range of downstream tasks, such as sentiment analysis, image captioning, and object recognition. (17:11)

Foundation models have the potential to revolutionize many industries, including health care, finance, and customer service. (17:23)

They can be used to detect fraud and provide personalized customer support. (17:32)

Vertex AI offers a model garden that includes foundation models. (17:38)

The language foundation models include PaLM API for chat and text. (17:43)

The vision foundation models includes stable diffusion, which has been shown to be effective at generating high quality images from text descriptions. (17:48)

Let's say you have a use case where you need to gather sentiments about how your customers are feeling about your product or service. (18:00)

You can use the classification task sentiment analysis task model for just that purpose. (18:07)

And what if you needed to perform occupancy analytics? (18:15)

There is a task model for your use case. (18:19)

Shown here are gen AI applications. (18:23)

Let's look at an example of code generation shown in the second block under code at the top. (18:27)

In this example, I've input a code file conversion problem, converting from Python to JSON. (18:34)

I use Bard. (18:41)

And I insert into the prompt box the following. (18:42)

I have a Pandas DataFrame with two columns, one with the file name and one with the hour in which it is generated. (18:46)

I'm trying to convert this into a JSON file in the format shown onscreen. (18:54)

Bard returns the steps I need to do this and the code snippet. (19:00)

And here my output is in a JSON format. (19:06)

It gets better. (19:10)

I happen to be using Google's free, browser-based Jupyter Notebook, known as Colab. (19:11)

And I simply export the Python code to Google's Colab. (19:18)

To summarize, Bart code generation can help you debug your lines of source code, explain your code to you line by (19:23)

line, craft SQL queries for your database, translate code from one language to another, and generate documentation and tutorials for source code. (19:30)

Generative AI Studio lets you quickly explore and customize gen AI models that you can leverage in your applications on Google Cloud. (19:42)

Generative AI Studio helps developers create and deploy Gen AI models by providing a variety of tools and resources that make it easy to get started. (19:53)

For example, there's a library of pre-trained models. (20:05)

There is a tool for fine tuning models. (20:09)

There is a tool for deploying models to production. (20:12)

And there is a community forum for developers to share ideas and collaborate. (20:15)

Generative AI App Builder lets you create gen AI apps without having to write any code. (20:21)

Gen AI App Builder has a drag and drop interface that makes it easy to design and build apps. (20:28)

It has a visual editor that makes it easy to create and edit app content. (20:35)

It has a built-in search engine that allows users to search for information within the app. (20:39)

And it has a conversational AI Engine that helps users to interact with the app using natural language. (20:43)

You can create your own digital assistants, custom search engines, knowledge bases, training applications, and much more. (20:51)

PaLM API lets you test and experiment with Google's large language models and gen AI tools. (21:01)

To make prototyping quick and more accessible, developers can integrate PaLM API with Maker suite and use it to access the API using a graphical user interface. (21:09)

The suite includes a number of different tools such as a model training tool, a model deployment tool, and a model monitoring tool. (21:21)

The model training tool helps developers train ML models on their data using different algorithms. (21:31)

The model deployment tool helps developers deploy ML models to production with a number of different deployment options. (21:38)

The model monitoring tool helps developers monitor the performance of their ML models in production using a dashboard and a number of different metrics. (21:48)

Thank you for watching our course, Introduction to Generative AI. (22:01)

Introduction to Generative AI(192)