[Music] the AI landscape is by any definition complicated and confusing there are many subfields or domains that are defined somewhat inconsistently consider all the terms you might have heard machine learning neural networks natural language processing large language models the list goes on many of these common terms point to different AI domains but in an overlapping and inconsistent way let's parse it a bit sometimes AI domains are described by the nature of the inputs they receive for example natural language processing or NLP is a domain that focuses on processing human language inputs voice recognition translation and modern spell check all rely on some version of NLP on the other hand sometimes AI domains are described by the nature of their output the tasks they achieve a great example is generative AI or gen AI a domain that focuses on using AI to generate new text image or audio content that g in chat GPT it stands for generative and we'll get to the p and t later and you can already see the domains overlapping since chat GPT relies on natural language inputs it also falls into the NLP domain a third and the most common way to segment the landscape of AI domains is by the type of model used this is where you'll hear terms like machine learning and deep learning using the model type lens the first domain is that of the rule-based AI models which were the first AI models used and are still used in many applications today the build step of these types of models involves hardcoding rules with if then logic if the input is a then output C humans set the rulle and the AI model applies them this type of model could be used for example to detect fraud in credit card transactions if certain characteristics are detected then Mark fraud however there's a big limitation to this human developers must know the rules what does determine fraud really fraudsters are constantly changing tactics detecting their patterns means following a moving Target so a different type of model might be used instead one that allows the machine to help determine which rules will best indicate frauds this is the machine learning or ml vast domain covering hundreds or thousands of different model types just like in the regression example we reviewed building an ml model in this case would involve three steps humans would complete the first two steps choosing the input variables and mathematical form the machine would then complete the third step finding the optimal parameter value via a training process in our store sales prediction example we used a straight line form and a regression to determine the optimal parameters many different mathematical forms can be used in the ml domain in a similar way say a curve a so-called tree or even a neural network each form does the math in a different way and the training approach will vary accordingly the neural network has been a major research focus in recent years it gives rise to a complex form with many parameters by mimicking the layout of the human brain calculations are made across a network of nodes called neurons arranged in layers we'll explain the details in a later chapter larger neural networks ones with more than two neuron layers between the inputs and outputs are considered deep and deep neural networks are so popular that they've defined their own AI domain a subset of machine learning called deep learning this is the domain of many of the most advanced AI models in use today like large language models llms and image generating models any machine learning model including a deep learning model needs to be trained this is the learning step in the next chapter we'll discuss the different ways to train a model [Music]