Gen AI Crash Course: Understanding the Buzz Words

Let's go all the way back to the basics and set the groundwork on what exactly AI is. Learn the history (including its ancient origins) and types of AI, including Narrow, AGI, and Super-intelligent AI.

Start to understand the different subfields of AI, and how the build on top of each other. Be introduced to machine learning, which is the next broadest category in the AI realm. Machine learning is a program or system that trains a model from input data that can make useful predictions from new or never before seen data. It utilizes both labeled and unlabeled data, and gives a computer the ability to learn without programming.

The basic definition of what is Generative AI, what the three workloads are, and what kind of tools are classified as GenAI.

Understanding the GenAI hype cycle, and what the predictions mean. How people are rushing to implement solutions like ChatGPT just to say they have AI, not realizing they are selecting the incorrect AI tools for their needs. Start to be introduced to the concept of "AI Ready"

Learning the two types of machine learning models (supervised and unsupervised learning), and their usage of data labels.

What are deep learning, neural networks, and semi-supervised learning. Learn the parts of the neural networks (such as neurons/nodes), and how neural networks send/receive information.

An Introduction to the two types of deep learning models: discriminative and generative.

Generative AI utilizes both ML and deep learning techniques leveraging supervised, un-supervised and semi-supervised methods in order to generate new content. In addition to putting this puzzle piece together, learn the three main phases of GenAI training.

What are the output types of Gen AI, and understand what Gen AI can/cannot do.

A quick overview of the market landscape of Generative AI tools.

Get a breakdown of the 4 main components of GenAI and become introduced to the technical infrastructure diagram Google uses to explain the GenAI puzzle pieces.

Understanding what foundation models are, the different types, and how the term is used.

Introduction to parameters and why Large Language Models (LLMs) are mentioned everywhere (since this is the type of Gen AI model you use the most)

Learn how parameters, nodes/neurons, weights/biases, and how certain information "wins" over others. This is probably one of the most technical sections of the course, so take it slow.

Understanding the basic unit GenAI LLMs utilize as the building blocks

Attention is all you need. The groundbreaking invention of transformers and how they made Generative AI possible.

Understanding positional encodings, attention and self-attention

An introduction to the newest age of neural network infrastructure, used by some Google models and Grok-1

An introduction to this unit, and explaining that these two technical methods are the true way people are "fine-tuning" and grounding models nowadays.

Learn what vectors and vector databases are, and how they work.

Vectors usually generated by neural networks designed to be consumed by a generative model for fine-tuning enhancements

The preferred fine-tuning/grounding protocol for LLMs currently by development teams. Combining LLMs and private business data to deliver better output.

See the flow diagrams and exactly how RAG works

Introduction to the section, understanding what prompts are, and what my everyday prompt method is.

See what I'm preaching live in action, and see the quality of the output drastically improve.

Create custom instructions in a chat interface like ChatGPT, watch your length, choose your words with care, set output limits, and use punctuation correctly.

The 7 main components you can use to enhance your image generation prompts and make them more precise.

See what I preached live in action.

5 mistakes you can avoid making when prompting

The dangers of hallucinations, what causes them, and one of the worst hallucinations I've seen in my career

The risks of generating misleading fake videos and deceptive narratives, leading to mistrust, propaganda and information warfare.

Understand some of the threats to privacy and security, including biometric profiling, un-authorized use of your likeness, data leaks, and risks of blackmail.

Explore legal challenges with Generative AI, stemming from the fact that copyright laws were made with humans as the creators.

Explaining what this section is, and a personal story of how I encounter these challenges frequently in airports.

Establishing the ground rules that your data should be protected more than ever, at all costs (and that doesn't just mean physical data), and sharing the sneaky example of LinkedIn Contributor Articles.

Evaluating how your data is being used to train models when using general access (ie directly from the chat interface, directly using their APIs with no protections, etc).

Discussing the ways to add data protect levels, including leveraging RAG, the Azure OpenAI Service, etc. This will also teach you the right questions to ask the software companies saying they are leveraging AI to make sure they're protecting your data, too.

An introduction to the claims of copyright infringement with AI model training data usage, the tech companies' defense of US Fair Use laws, and the argument for transformative works.

A quick introduction to what is fair use laws and what these lawsuits will evaluate.

Focused on the purpose and nature of the usage, and how much it brings something new or "transforms" the original work.

Evaluating the second factor of how much of the copyrighted work was used, and OpenAI's contention of typical fair use law to make a distinction between how much of it was distributed vs. privately used (something normal fair use does not consider)

How artistic works (such as music, video, or artistic works) vs more fact based copyrighted works are evaluated differently.

Evaluating how impact to income or market of a copyrighted work in a fair use case.

Drawing from my experience previously in music copyright law, I offer a different point of view, as I believe that the way its transformed could also be evaluated in a different way similar to entertainment copyright law precedents. Also the conclusion for this unit.