Understanding the evolution of conversational AI is crucial for learning generative AI as it provides foundational knowledge and context. This historical perspective reveals how AI technologies have progressed from simple, rule-based systems to complex machine learning and deep learning models that are core to both conversational and generative AI.

This section explores the evolution of conversational AI, culminating in an in-depth look at LLMs, the technological backbone of contemporary chatbots.

What is conversational AI?

Conversational AI refers to technologies that enable machines to engage in human-like dialogue, comprehend complex commands, and respond intelligently. This is achieved through machine learning and natural language processing capabilities, enabling the system to learn, understand, and improve over time. The following figure demonstrates one such conversation:

Figure 1.1 – Conversations with...

Generative AI refers to a field of AI (as stated in the preceding figure) that focuses on creating or generating new content, such as images, text, music, video, code, 3D objects, or synthetic data that is not directly copied or replicated from existing data. It involves training deep learning models to understand patterns and relationships within a given dataset and then using that knowledge to generate novel and unique content. The following is a visualization of what generative AI is:

Figure 1.3 – What is generative AI?

It is a broad field whose primary function is to generate novel content. Examples of generative AI models include image generation models such as DALL-E and MidJourney, text generation models such as GPT-4, PaLM, and Claude, code generation models such as Codex, audio generation tools such as MusicLM, and video generation models such as SORA.

The rise of generative AI in 2022-23

Generative AI...

Generative AI has a wide range of applications across various industries, presenting several use cases that can bring significant benefits to businesses, and the applications are continuing to grow at a fast pace. In this section, we will discuss popular tasks and models and examine the latest emerging business applications that have gained significant traction recently.

Let’s begin with text generation models.

Text

Text generation models can be used for diverse tasks as outlined here. In the following, we have mentioned the most popular tasks that we have seen architecting solutions with our customers:

• Summarization: They can condense long documents, such as textbook chapters or detailed product descriptions, into concise summaries while retaining the key information.
• Question answering: These models can provide accurate answers to questions, which is particularly useful in automating the creation of FAQ...

Cloud computing has been instrumental in bringing LLMs to a wider audience. LLMs use large-scale GPU processing to learn and generate human-like text, image, audio, and video, engaging in increasingly interactive and intelligent ways.

This section highlights several advantages of leveraging LLMs in a cloud environment:

• Scalability: Cloud computing enables users to access high-performance computing such as GPUs as necessary to run LLMs. This makes it easy to scale applications as required based on consumption needs.

  Since LLM models such as GPT are heavy API-driven workloads, there is a need for API management services, such as Azure APIM, that help achieve scalability, security, and high availability across regions. They can also capture telemetry that can help determine token usage and error logging across organizations. We discuss scaling strategies on Azure in Chapter 7.

• Affordability: There is no need...

Developing an idea and moving it into production is a multi-phase process that typically involves ideation, validation, development, testing, and deployment. The multi-phase process of developing an idea and moving it into production is crucial because it methodically transforms a concept into a viable product.

Take a look at the following image about overlooking a crucial aspect:

Figure 1.8 – Two entrepreneurs engaging in a humorous discussion about overlooking expenses

The above image satirically showcases how some organizations claim to build AI from scratch, when in reality, they're just utilizing API calls to services like OpenAI. It humorously uncovers this exaggeration when asked about the Open AI bills, mocking the notion of starting from scratch.

Each phase serves a distinct purpose: ideation fosters innovation, validation ensures market demand and feasibility...

The aim of this introductory chapter was to highlight the history, core concepts, and other essential information necessary for readers to develop an end-to-end generative AI solution on the cloud. We have explored the evolution of chatbots from simple rule-based systems to multimodal, context-aware, and action-oriented agentic LLMs. We delved into the rise of generative AI, focusing on LLMs and foundation models as well as their relationship and key attributes. The differences between open source and closed source models were examined, alongside trending business applications drawn from our experiences. In the rapidly evolving landscape of AI, we’ve examined a few leading models, including text, image, audio, and video generation. These models represent the forefront of AI technology, showcasing remarkable capabilities in creating high-quality, lifelike content. We then highlighted how cloud computing facilitates the development of secure, scalable, cost-efficient,...

• Nvidia Generative AI: https://www.nvidia.com/en-us/glossary/data-science/generative-ai/#:~:text=Generative%20AI%20models%20use%20neural,semi%2Dsupervised%20learning%20for%20training
• CSET Georgetown University: https://cset.georgetown.edu/article/what-are-generative-ai-large-language-models-and-foundation-models/#:~:text=Using%20the%20term%20%E2%80%9Cgenerative%20AI,system%20that%20works%20with%20language
• Databricks course: https://microsoft-academy.databricks.com/learn/course/1765/play/12440/llms-and-generative-ai