AWS Flash - Unleashing Innovation: The Generative AI Revolution

We find ourselves in a society where the line between what is created by humans, and what is created by machines, is increasingly blurred. Generative AI has been a turning point in how we create, design, and interact with technology. But how does it work, what are the benefits beyond novelty and what are the risks?

Join us for an approachable introduction into how Generative AI works in a no-nonsense, understandable way; and how we can use this technology not just as a stand-alone tool, but in collaborative partnership to responsibly drive innovation and transformation.

• Course level: Fundamental
• Duration: 1 Hour 30 Minutes

Activities

This course includes presentations based on practical examples, with use-cases and demonstrations.

Course objectives

By the end of this session, attendees will be able to:

• Define artificial intelligence, machine learning, and the three types of machine learning
• Describe how machine learning algorithms learn and subsequently output a machine learning model
• Understand the difference between traditional machine learning algorithms and deep learning algorithms
• Describe how artificial neural networks work
• Understand the difference between discriminative (predictive) AI and generative AI
• Describe how Large Language Models (LLMs) are trained and used for text generation
• Understands the purpose and importance of Foundation Models (FMs) and how prompt engineering and fine-tuning can be used to customize FMs.
• Describe how diffusion models are trained and used for image generation
• Describe the practical applications of generative AI
• Identify issues surrounding responsible and inclusive use of generative AI

Intended audience

This course is intended for:

• Non-technical enthusiasts
• Technical enthusiasts
• Decision makers

Prerequisites

None

Course outline

Introduction

• This section provides overall context of the current state of the artificial intelligence and machine learning landscape. It starts from the basics with a definition of artificial intelligence before drilling-down into machine learning and the three main types of machine learning: supervised, unsupervised, and reinforcement learning. There is a high-level discussion about how these enable artificial intelligence and machine learning to actually learn, which leads into comprehensive coverage about deep learning and artificial neural networks - the technologies which underpin generative AI.
• What is artificial intelligence?
  • What is machine learning?
    Supervised learning
    Unsupervised learning
    Reinforcement learning
• How does machine learning learn?
• What is deep learning?
  Artificial neural networks

• Discriminative AI

Generative AI

In this section we dive-deep into two of the most popular, and well-known, forms of generative AI: text generation and Natural Language Processing, and image generation and diffusion models.

• Text generation and Natural Language Processing

The cornerstone of text generation and Natural Language Processing are Large Language Models (LLMs). We go in-depth about what is an LLM and how these are more advanced and effective than traditional machine learning approaches towards solving text-focused problems - with particular focus on transformer architectures. We then discuss how these work and can be developed to address practical use-cases utilising your own data. This leads into the final part of the module covering Foundation Models (FMs) which significantly lower the cost of entry for organisations towards using generative AI through access to pre-trained, high-performance LLMs.

• What is a Large Language Model?
• How does this differ from past approaches?
• Transformer architectures
• Conditional text generation
• Developing Large Language Models (LLMs)
• Demo: Foundation Models with Amazon SageMaker JumpStart

• Image generation and diffusion models

Dynamic image generation and synthesis has gained significant advancements over the past few years, particularly with the introduction of diffusion based models which is the focus of this section. We discuss how diffusion models work and can be trained to not only generate images, but generate images based on a prompt provided by the user to guide the types of images generated.

• Diffusion models
• Training diffusion models
• Effective diffusion models
• Improving noise prediction and removal
• Latent space
• Text conditioning
• Text-to-image synthesis
• Demo: Stable Difffusion with Amazon SageMaker JumpStart

Practical uses of generative AI

While generative AI has gained widespread attention through consumer-facing applications, real benefits for organisations can come through business-facing applications. While traditional AI approaches have been available for some time, generative AI brings a new wave of possibilities which were not possible with these traditional approaches. This section discusses some of the practical-use cases of generative AI for organisations.

• Content generation
• Prototyping
• Data analytics
• Content analysis
• Chatbots and virtual assistants
• Creativity benefits

Responsible and inclusive AI

Like with any technology, there is scope for misuse and misrepresentation. This final module of the course discusses ways that responsible and inclusive products and services can be built using generative AI.

• Scope of generative AI
• Impact on underrepresented groups
• Toxicity
• Hallucinations
• Intellectual property
• Plagiarism and cheating
• Disruption of the nature of work

Conclusion: Building responsible AI products and services

 The conclusion to the course provides next steps for attendees who are keen to continue their journey in generative AI, with suggestions about courses and learning pathways available through AWS Training & Certification.