Entreprise GPTe Unleashed: Mastering the Art of Generative AI with h2o.ai

Thanks once again for joining in today's webinar. I am the Head of Global Training at h2o. ai and among my responsibilities, I also oversee  content strategy and creation, as well as certification programs.

For me, promoting a  continuous learning culture is very close to my heart, and that's why I'm here today to talk  about the platform called Enterprise H2O GPT. Designed to streamline your experience,  this platform is tailored for those in search of answers from their company's  internal and secure documents and data, all while ensuring non-disclosure to any  external cloud providers if this is what you prefer. So to offer a brief overview,  this platform harnesses generative AI crafted by our top Kaggle Grandmasters  for business and organizational needs, aiming to simplify access to information  within your company's internal data.

You might be aware that privacy, data privacy,  and data confidentiality are at the core of our legislation nowadays, and actually, are  at the core of any sort of legislation. What we did at H2O was to cater for this by  offering you the possibility of not taking the data outside of your company but keeping it  private to yourself while you are also going to be able to leverage the capabilities  of generative AI and especially LLMs. In today's webinar, we will delve into the  following topics.

We are going to have a short introduction, a brief overview to set the stage  for our discussion. We are also going to learn what RAG or Retrieval Augmented Generation is,  and we are going to see why this is important in the context of our platform. Then you are  going to discover a bit about what the H2O generative AI ecosystem is all about and what it  has to offer.

Of course, we are going to set the stage for discussing features and benefits of our  Enterprise H2O GPT, which is our platform that is going to help you act in a very user-friendly  manner to be in control of your data and in control of everything that you want to extract  from your data. We will have a short demo, a live demonstration that showcases the  practical application of the discussed concepts. We are going to jump in right ahead, and  by the end of this entire experience, you are going to be able to have your  questions answered.

So once again, please don't hesitate to ask them in the  Q&A tab, and I'm going to address them at the end of this webinar. Okay, so  let's begin with some opening remarks. So in today's AI landscape, there is significant  excitement around generative AI and large language models.

That's why you are also here, I assume.  This enthusiasm or excitement has prompted companies to recognize the transformative impact  of these technologies because it's a real thing, it's happening, and we are here in  the hype of all these situations, seeing how generative AI and how LLMs are  actually impacting our day-to-day life. So the demand for innovative content,  personalized user experiences, and enhanced communication strategies has  shifted AI priorities, leading nowadays organizations to adapt and explore how these  capabilities can elevate products, services, and overall business strategies.

This evolving AI  landscape not only shapes technological progress but also has significant implications  for legislation and politics globally, as I just said before, and I'm emphasizing  this idea because the demo will contain UA Generative and Artificial Intelligence Act  that I would like to explore with you together. As generative AI and LLMs gain prominence,  policymakers face the challenges of creating frameworks that balance innovation with  ethical considerations and societal impacts, not to mention the data privacy and all  this data security that is so important to be aware of when you want to use  a platform with confidential data. So what exactly sets generative AI and LLMs apart? 

Before we delve into the demo, let's just cover some of the basics. You might already know that  generative AI is a broader category that stands out for creating new data, which is a departure  from traditional AI, which was focusing on analyzing existing information. So in generative  AI, all eyes are on large language models or LLMs.

And these models are armed with a lot of  parameters and trained on a large corpus of data, and this helps them excel at crafting  human-like text. Why does it matter so much? Well, because as human beings ourselves, our fundamental  means of communication relies on language.

We navigate and express ourselves through words  and written text, whether expressing thoughts, sharing ideas, or giving out information. The  power of language highlights the importance of advancements in technologies like  generative AI and large language models. From my perspective, it was inevitable  that this moment occurs sooner or later, and it's better that it is sooner so that we  all learn how to leverage the power of gen AI, even though not all of us are coders.

These  innovations profoundly shape the landscape in which we interact, exchange, and interpret  information in the digital era. LLMs shine because they understand and produce remarkably  natural, contextually relevant language. The buzz around them stems from their profound  impact on natural language processing (NLP), which helps them understand intricate language  patterns and context.

So that's why LLMs are so important, and that's why generative  AI, which is a broader category of LLMs, is so important. In addition to large language  models, there are generative models for audio, video, and images. However, we will  focus on text today because humans are prone to wanting to communicate  through words and written text.

It's only natural for us to consider incorporating  these LLMs and generative AI into our daily work and lives. The benefits behind these tendencies  are significant and encompass at least several compelling reasons. Content creation is  one of them, as these technologies excel at generating diverse and engaging content. 

Creative assistance is another aspect, providing valuable support in creative endeavors  by assisting with ideas and innovation. Natural language understanding is crucial,  with models that grasp the intricacies of language and context, enhancing communication  and comprehension. Personalization of the entire user experience and data augmentation,  in case we don't have enough data and want to create similar data, are additional  benefits.

These are just a couple of them. However, it's not all smooth sailing for  generative AI because it comes with its own set of challenges. Here are some key  points that I observed and that I am going to address through presenting to you the  platform because our platform was created as an answer to these challenges,  and I am super happy about that.

Biasing data, for example, is a challenge.  Imagine training a language model using data that unintentionally favors certain perspectives, leading to biased or skewed outputs.  That's something we really want to avoid, and platforms that offer gen capabilities need  to be able to identify the bias in the data and mention it.

What's great about different tools  is that when you are creating the problem or the question based on the documents, you can  write and say, "Be mindful of any bias in the data that might occur. " Our platform has been  tested, and our algorithms can observe the bias in the data in most cases. It means that we  are not only at the forefront of technology, creating new content, but we are also able to  identify biases through the algorithm itself.

And that algorithm is able to come back to  us and let us know that that's the case, actually, and that's wonderful. Lack of control: Think about trying to navigate  through unpredictable outcomes where overseeing and guiding the generated content becomes a very  difficult task. So it's not up to us, the user, to control the information but to the platforms  that are actually offering these products.

Security concerns and data privacy: That's  something that needs to be addressed over and over again because as a user, I want to make  sure that when I upload my data to the platform, if I want that data to stay on my local service  on-premise, for example, I can have that option. It's not mandatory for me if I want to use the  service to upload it on the cloud, and that's definitely a possibility when looking at gen  AI capabilities that we, here at H2O AI, offer. Resource intensiveness: That's yet another point  that I am interested in developing because the challenge of dealing with a hefty demand for  computational power is big.

This computational power is required not only in the training  phase but also in the deployment phase of the model. Not all individuals and companies will  be able to leverage its potential if they don't have the hardware. So being able as a software  provider to give access to efficient algorithms that are also not so intensively computational  in terms of resources is extremely important.

Understanding context: That's another  challenge that gen AI faces because if you're not providing additional context  in the form of prompts or written text, it would be very difficult for the GPT to give  you the answers. Understanding context is very difficult, and the difficulty of teaching  a system to grasp the subtleties of the context are big. But we want to make sure that we  ensure that the content that is being generated is going to make sense and be useful to us. 

That's where we started our discussion about developing a platform that wants to give this  possibility of having more context to learn about. So understanding context stands out for me as a  crucial aspect and significant challenge for large language models. When you ask a question to an  LLM, it generates a response without considering additional text.

To receive meaningful answers,  you must excel at crafting precise prompts or ask the right questions in a very detailed manner.  Additionally, you need to make sure that the latest information to the algorithm is being  provided, not only from five months ago. However, as a user, this is, in most cases, out of  our control, and thus, LLMs or generative models have limitations in understanding context,  resulting in superficial solutions to problems.

Now, I'm going to explain the idea of RAG. This  is where RAG, or Retrieval Augmented Generative Models, comes into play. RAG addresses the  challenge of context understanding by using what data scientists call a retrieval system. 

This system fetches relevant information to help the generative model better grasp  the context, ensuring that the generated content fits the situation accurately.  This way, for example, RAG enhances the model's ability to create content that makes  sense and aligns well with a given context. RAG is an AI system that blends retrieval  models to find information quickly and accurately from a large text database, for  example, a PDF document or a huge amount of structured or unstructured databases  like search engines.

It then combines this with generative models that create  new text based on a prompt or an input, useful for tasks like writing articles  or generating chatbot responses. So I assume that the majority of you have  until now used more generative models, but the majority of you have not actually had  the occasion to implement retrieval models through the interface by extracting information  from a given document. This is what we are going to do in the demo.

But before, let's once again  make the distinction between the two of them. We have generative models and retrieval models  that work together to create something called RAG or Retrieval Augmented Generative Models.  They are like machines, LLMs on steroids, let's say like that, because you give them also  documents to work with.

Combining these models makes a RAG system powerful and flexible. For  instance, it can use a retrieval model to find relevant information swiftly, and then it can  use a generative model to create a summary or a response based on that information. This  proves valuable in situations where information is intricate, and a simple retriever  or generation approach may fall short.

Now, you might be thinking, "Andrea, I  got it. I see the distinctions between generative AI and large language models. I  understand the pros and cons of generative AI specifically.

I see how RAG can help with  contextualization and context understanding in general. But how does this all affect or  connect to the H2O AI ecosystem that you want to present? " And I would say that's a great  question, and I'm aiming to address this next.

In advancing our generative AI capabilities, our  focus has been on harnessing open-source models, which culminate with the development of our H2O  GPT, and this is open source. You can test it for yourself on our website, and I would highly  encourage you to start playing with H2O GPT. This system includes a large language model that  you can select according to your own needs.

You have the possibility to select your own  large language model. It has a command-line interface and a user-friendly graphical  interface. I already did a webinar on that, so I would encourage you to go on  YouTube if you're interested also to learn more about it.

But that's  the first stage that we focused on. Then, what we wanted to do is to build on  top of it. We've introduced the LLM Studio, a framework that is equipped with a no-code user  interface, which is tailored for the fine-tuning of large language models.

Because the majority  of the GPTs that are out there don't offer you this possibility of fine-tuning. So H2O LLM  Studio is a versatile tool that streamlines the entire fine-tuning process and allows you easy  adjustments for various hyperparameters without the need for coding expertise. Once again, we want  to take the tools and the knowledge that we have acquired through our machine learning engineers  and Kaggle Grandmasters and put it in the hands of the people who are not necessarily interested  in learning how to code in Python or going in a lot of depth to modify the parameters of the  tools.

That's why we wanted to make it simple. As part of our commitment to innovation,  we also integrated RAG technology into our platform. So we leveraged vector databases  within RAG, which enhances the capabilities of our platform.

These vector databases improve  the system's efficiency in general to retrieve relevant information. Remember, we discussed  retrieval models. With the vector databases, we are able to improve the retrieval of  the data that we apply our models on.

The integration across H2O GPT, LLM  Studio, and RAG with vector databases actually strengthened our approach to  enhancing generative AI and contextual understanding in general. So our team  of machine learning experts has created a lot of generative tools. It would take  a while for me to talk about all of them, but there's a quick overview here on this slide  at what our H2O AI and gen ecosystem includes.

At the top of the slide, we started by introducing  tools to simplify machine learning in general, such as AutoML capabilities. We developed  advanced models using deep learning, adapted, and deployed these models on-premise or on the  cloud. We streamlined the app creation process with user-friendly interfaces accessible  to everyone, and that was the first step for us.

The second step is exactly what you  see underneath when we started to incorporate capabilities of large language models across our  platform. This means we now offer tools for the entire generative AI lifecycle from importing and  embedding data to pre-processing, fine-tuning, evaluating, and deploying everything related  to gen. This extensive package is referred to as H2O LLM Studio Suite.

The entire thing  underneath the rectangle is the LLM Studio Suite, with Enterprise H2O GPT being just one of its  offerings. So underneath Enterprise H2O GPT is one of the offerings consolidated within a platform  that incorporates the entire LLM Studio Suite, and that's what I am aiming to present to you  today. H2O AI is Enterprise generative AI, or H2O GPT offers advanced document chat  features, including fact-based question answering, a pre-loaded Wikipedia state, offline mode.

It  also includes a persistent database with vector embedding, and this system supports a lot of  document types that you can query and inquire at your own desire by using easy-to-use interfaces.  It employs RAG, so this entire interface uses retrieval models and generative models to give  you the best and most accurate answers. This RAG technology is used for context understanding and  data understanding overall, guaranteeing that the entire Enterprise GPT ensures safety, control, and  transparency for the users.

In addition, it offers the Python API, which lets you use the H2 GPT's  advanced capabilities in Python projects, and that's for more coders here. With this API, and  if you are a code-savvy person, you can leverage the platform further and tap into large language  models and apply gen for content creation. By coding, you can work with the vector database for  managing embeddings, implement RAG to enhance data understanding, and all is seamlessly integrated  into your Python workflows.

So you can use the interface as such, or you can use the Python API,  and it is a powerful tool for data scientists and developers to leverage these features within  Python applications. I think that is also another way that we come to the core of us being machine  learning engineers in the company and Kaggle grandmasters. We also think that even though we  want to facilitate, show the things for everyone, we can also give the possibility for expert  coders to go beyond and above to personalize and customize their models even more.

So by  saying that, allow me to head to the website and access the Enterprise H2O GPT once again. Please  don't hesitate to ask your questions in the Q&A section. I will take them at the end of this demo. 

I would love to see what type of curiosities you have with respect to what I presented until now.  So don't hesitate to do that. Also, for this demo, you can follow along because this is a platform  that is open source and accessible for you in a trial version.

So I would highly encourage  you to do that. What I'm going to do right now is to connect on h2o. ai, and that's our website. 

If you hover over on the top of the screen to the platform tab, the first thing that you are  going to see is a column called generative AI. And so the first row in this column is  Enterprise H2O GPT, and it is also written here, private powerful LLM Search Assistant for internal  teams to answer questions from large volumes of documents, websites, and workplace content.  So what is great here is that we have this emphasis on internal teams.

So we have within  this platform a lot of collaborative tools so that when your company is interested in working  with this platform, you are able to share all your projects and what we call collections with  your team members. You are able to publish them. You are able to actually make them public.

So all  sorts of things that we incorporated within the platform so that we enhance these collaborative  capabilities. So I clicked on the tool itself, and here you have at the beginning a video that  showcases a bit what this platform is about. It is about information retrieval on internal  data, privately hosted LLMs, and this way your data stays with you.

So that's like the strong  core point that I am willing to emphasize. So in addition to having H2 GPT capabilities where you  would ask questions and receive generic answers, you might also be interested in uploading your  own internal documents and extracting information from them by leveraging the generative models  but actually using retrieval models to find the information. So just think about having let's  say a client's internal document that is unique in the world just created by your company for  your company's internal use.

This information you will not find anywhere else. You would like  to keep this document internal to the company, safely stored. Nevertheless, you would like to  actually improve your experience and your work, make it more efficient and faster and more  reactive by querying hundreds of pages of this document in a couple of rows.

And so you would  receive actually here the output of the document with the page number as well as the accuracy  percentage after the output is being created. And that's awesome because not only you will receive  a short summary of what your interest is about by querying the Enterprise generative platform, you  would also receive the pages previews of where those answers are going to be found. So without  further ado, let's just click on the TR now, and we are going to be redirected to a new tab  called Enterprise H2 GPT, and the link is the following one: H2Ogpt.

gen. h2o. ai.

So this  is open for everyone to use. I'm going to actually walk with you a bit through the platform  so you understand the main functionalities of the platform, and we are going to upload the document  just to see what the capabilities look like. So here you are prompted on the homepage where you  have the dashboard with your identifiers and the possibility to create a new collection, take  the collection as a project where you are going to have all your chat sessions as well as all  your documents.

As for the moment, I uploaded the document. Let me actually see if I did that.  Okay.

And this is a document called EU AI act. So I can see all my documents here when I clicked on  the documents or in the form of a list. What is great about this tab is that I can actually select  the documents and delete them if I'm interested.

So I'm going to have a row with the type of the  file size of it, number of pages, whether or not the status is being complete or indexed in this  case, as well as when this was uploaded. So I'm going to pick it so that we can delete it. I want  to delete this file so that we will re-upload it once again.

So on the homepage, once again, you  have the number of active chat sessions, number of collections, and number of documents. This is  the page that you will also be able to see when you connect on it with zero collections, chats,  and documents. This is something that we need to create ourselves.

Underneath, you are going to see  the entire history of recent chats, collections, and documents. And here, just underneath the home  button on the left side of the screen, you will be able to see the three tabs and their particular  visualization. So you have collections, documents, and chats.

If I click on collection, what I  can do is actually to create a new collection. And I can actually select, let's say, webinar  Enterprise GPT, and I can give a description. So take this as a project creation, and I can  create this collection.

And here I will be able to upload all my documents that I am interested  in using. So once I created the collection, I will be able to see all my collections if I  click once again on the tab. So here you will see all your collections or your projects.

And  mind you, this is a private collection. So if I am using company Enterprise installed application,  I will be able to make it public. Nevertheless, it's not the case because this is an open  source.

So we just want to have the control over these functionalities. Within these collections,  actually, what I can do is to view the collection itself. And then I am able to enter within the  collection that I just created where I can see the number of documents that are being stored, the  size of all these documents, as well as when this collection has been created.

On the right side of  the screen top, I can click on the settings. And so here I can modify the collection name, the  description. I can also share it with my peers or my colleagues in case this is installed on my  own company servers.

And what is great about this is actually that I can set prompts by default  that will propagate all over my documents and chats that I'm going to use within the collection  themselves. So by default, we have an example, but of course, we can change it. So here we  have the first system prompt: "You are H2O GPT, an intelligent retrieval augmented gen AI system  developed by H2O AI.

" So we are giving context to the collection, what this collection is,  generation approach, rack type to use. So here, if I click on the drop-down menu, I have four  options. I have the option to use only LLMs, so not rag actually.

So I have the option to  generate information without retrieval models. I have the option to use retrieval models, and I  have two more advanced options. Of course, we have documentation about this that goes into further  details where you are able to use embedded LLMs, rags, as well as query once or twice.

So that's  a detail here that maybe code-savvy people are interested in. By default, we are using rag. And  since I had the presentation about what it means, now you are able to understand that rag uses  retrieval models that are going to extract information from the documents that we are going  to upload here.

And we are going to actually have a personalized output that uses generative  models for output creation as well as rag models, retrieval models, for personalized outputs. Next,  we have a rag prompt after, before context. So pay attention and remember the information below which  will help to answer the question or imperative after the context ends.

Of course, you can change  it. And usually, if you're using H2 GPT or any other GPTs, you might rephrase this paragraph  by giving bullet points. And so you are saying, "Every time when I'm putting something, take into  consideration that I want this paragraph that I'm going to input in the chat with bullet points,  rephrased maybe shortened, etc.

" Right? So this is a guideline for the GPT to learn how to treat your  text, what is the style, maybe if you have clients that let's say are over 60 years of age, you would  like to use a more polite version when answering than to people under 20 years of age. You would  like to adapt to the times, to the customs, etc.

So all of this is great, and I find it so  useful in comparison to any GPTs that don't have this option. You also have the possibility to use  a rag prompt after context. So according to only the information in the document sources provided  within the text above.

So this is a way to say, "Don't hallucinate, don't create new information  because you're also using generative AI. Just give me the strict information that you found from  the text and don't create, don't make things up. " Right here, for example, as compared  to the challenges that I presented to you, you can say here, "Don't show me any possible  bias within the answer or the data.

" So you can create and put here a address within these  rag prompts any sort of challenges that GAN has, which are not completely taken care of without the  emphasis on specific prompts. And then I can click on update. And what happened is that at the level  of this project, all these prompts are going to be taken into consideration by default whenever  I am uploading a document and asking through a chat information about the document.

So let's do  just that. I'm going to click on "Add Documents. " And I'm going to be able to do multiple  things.

I can upload documents from my own PC, from web pages, and from the file system that  is connected to the server where the platform is installed. Let's say I want to upload the  document from my PC, and I'm going to go to Downloads. Here I have downloaded 350 pages of  a new AI act that I'm able to upload within the platform.

This has been successfully taken into  consideration, and now I'm going to click on ADD. What happens now is that I am brought to this Jobs  tab where I see all my active jobs. As you see, I have already created a couple of jobs where  I deleted a document 7 minutes ago, as well as I'm ingesting a new document.

I have also created  and tested some of the things just to see that as a user, you have the tracking and the history of  what happened on the platform. I see that I have my name here, and this is being uploaded, so the  document is being ingested and now has actually completed to be ingested. So once again, I have  the collection, and within this collection now, I have a private item that is 860 kilobytes. 

Okay, and now what I'm able to do is to search this collection and start my first chat. So  I have this EU AI legislation documentation. Maybe I'm interested to ask specific questions  with respect to this.

What are the top three legislation proposals with respect to generative  AI? Let's see. Like I haven't tested that one, but what happens now is actually Enterprise  GPT searches all the 350 pages of documents and is coming up with an answer.

What I like with  respect to this is that it's still pending, but I can actually visualize particular parts that are  taken into consideration. So when I click on this, I can actually visualize, so I receive a small  summary with respect to my question, and also I receive all the pages that have been concerned  by the output. So here I have a couple of pages, each of these pages with their output and  accuracy score, let's say, with respect to what the algorithm has found.

So the answer is based on  the information provided in the document sources that was a prompt, if you remember. The top three  legislation proposals with respect to gen AI are: Proposal of a regulation on AI systems  which aims to establish a framework for the development, deployment, and use of  AI systems in the European Union. I just took this one because I live in France, so  actually I was curious to see what happens.

Amendment 114, which proposes to authorize the  placing on the market or putting into service of AI systems that have not undergone a  conformity assessment, blah, blah, blah. And Amendment 115, which proposes to require  providers of high-risk AI systems to register their systems in a UA database. Nice, nice,  right?

So actually I really have concrete answers with respect to this document. Of course,  if I would be a lawyer and I would have all the keywords possible to ask more specific questions,  I would. But I'm not, so what I'm going to do now, let's say I want to save this and make it  shorter.

Let's see what happens. So now let's see if actually it was able to detect my query,  hoping that it would take into consideration the previous prompt, which I think is actually the  case. Shorter version of the information.

So the document discusses the proposed EU Artificial  Intelligence Regulation. So actually what I asked is, take this document and try to summarize it  as short as possible. The outcome, like I just consolidated with the help of Enterprise H2 GPT,  350 pages in 10 rows or so, right?

Which is great. I think it's great. And what is also great is that  you are able to go into the chats and normally you will be able to visualize the chats after a while  because you just queried the collection.

So here, I was curious just how the document looks and  what it functions, what it looks like now. I will be able, after this page refreshes, actually  to be able to see the history of all the chats. And what is great is that I can personalize the  chats even more if I want to.

So I personalize the parameters and the settings of this document, but  I can go and let's click on start your first chat. And now, I really have, in a sort of a history  archive, all the information that I was interested in. So, give me the short summarization of Law 15. 

Let's see if it recognizes that this is the law and not an amendment here. So that's what I asked.  Okay, okay, sure.

There's a short answer based on the provided document sources. Law 15, also known  as the Regulation on Artificial Intelligence, is proposed to ensure that systems are developed  and used in a way that respects human rights. Now, make this shorter, please.

Also, observe  the fact that you also have all the references, and this is great if you want to go into further  details. Like, I think we all use that PDF option, like Ctrl+F, to find a particular keyword  in a PDF. Well, take this one as the most updated and advanced version of that thing that  also incorporates AI into it, right?

And so, it keeps account of the context, gives you  information. Now, I see that my recent chat was created a couple of seconds ago, and I have a  preview with respect to what that chat was about. This is great because when I click on it, what  can happen is that I'm able to share this chat.

I can copy this link and share it with my friends. I  can also click the button "Show Related Chat. " So, here, I will have all the related chats.

Let's say  we have hundreds of chats over the month. You're going to see all the related chats that have the  same type of information taken care of. What I can do here is actually go beyond and modify the  settings of the entire collection or project by adjusting and making it more personalized.

So,  for example, I can change the style. I can put here "Use the language of a 10-year-old child. "  Let's say I would like for this to be very easy to understand, and I'm updating the information,  so I would like maybe to make it shorter.

But now, I am focusing on also having a simpler language.  I could have just said that, right? So, maybe I don't understand all the things in this  document because they are too specific.

So, what I'm trying to use here are prompts that  are predefined so that we can modify the settings and use them in a format that takes into  consideration that type of language level or uses the format that is more frequent for non-native  speakers of the English language. So, this is it, guys. The idea here is once again to explore it by  yourselves.

You're going to have a lot of fun with these collections and just test it by yourselves,  uploading documents and extracting information out of it in a very intelligent manner. That's  what this entire platform is about. So, don't hesitate to play around.

The last thing  I want to show you is the fact that you can actually generate API keys and incorporate  them within your Python notebooks. So, here's an API that you can generate, and this is  what this code would look like. So, you're able to create a new collection.

We have specific  functions and attributes for this. You're able to use this to upload documents and then create  chat sessions, query the collections, and reply within these queries. You can also create a test  API that you can attach to your own collection.

And you can generate this API that I'm going  to copy and keep to incorporate within my own notebooks. There is not enough time for us to  do that, but just be aware of the fact that you can use the APIs within Python notebooks to  leverage all the potential of the Enterprise GPT platform if you so desire. So, we did a short  presentation of the Enterprise H2O GPT platform, and I'm really hoping that you were also  able to test yourself some of the things.

To sum up our presentation for today, we  began by introducing the difference between generative AI and large language models.  Large language models are a subcategory that concerns only text data from the  broader category that generative AI is. We also delved into the various benefits  and challenges associated with generative AI, setting the stage for an insightful discussion. 

I hope through this session, we also learned what RAG is all about or retrieval augmented  generation. We saw that it uses something called retrieval models and generative models  so that it leverages the capabilities of LLMs beyond the generalization point and adapts it to  specific text that you are going to give to it. We also looked at the H2O generative AI  ecosystem and saw what the advantages of having a seamlessly integrated platform are  all about.

The fact that we offer an end-to-end generative AI platform for any sort of generative  AI lifecycle is something that is super useful, at least from what I see from our clients. Our  journey here ended with the demo, and I hope I provided to you a practical demonstration and  showcased the functionalities of Enterprise H2O GPT, how it can be seamlessly  integrated into real-world scenarios. Before taking questions, if you're interested  in learning about our LLM learning path that I am going to present, it's a course that  I created.

Don't hesitate to connect on YouTube and take the courses out there.  We have a lot of courses and information that I would encourage you to check out. It  explains theoretical concepts and then shows how they apply to our ecosystem.