AI Fundamentals: The Hierarchy Explained

AI is a field; machine learning is a subfield

Artificial intelligence is the broadest discipline, similar to physics. Machine learning is a subfield of AI, just as thermodynamics is a subfield of physics. Deep learning is a subset of machine learning, and large language models sit at the intersection of deep learning and generative models.

The nested structure of AI disciplines from broadest to most specific

What machine learning does

Machine learning is a program that uses input data to train a model. That trained model can then make predictions based on data it has never seen before. For example, a model trained on Nike sales data can predict how well a new Adidas shoe would sell based on Adidas sales data.

Supervised vs. unsupervised learning

Supervised learning models use labeled data (e.g., blue dot = pickup, yellow dot = delivery) to make predictions. Unsupervised learning models use unlabeled raw data and look for natural groupings or patterns. After supervised models make predictions, they compare results to training data and try to close gaps; unsupervised models do not.

Key difference: supervised models require labeled data; unsupervised models work with raw, unlabeled data

Deep learning uses artificial neural networks

Deep learning is a type of machine learning that uses artificial neural networks inspired by the human brain. These networks consist of layers of nodes and neurons; more layers mean a more powerful model. Deep learning enables semi-supervised learning, where a model trains on a small amount of labeled data and a large amount of unlabeled data.

Semi-supervised learning in practice

A bank detecting fraud might label only 5% of transactions as fraudulent or legitimate, leaving 95% unlabeled due to resource constraints. The deep learning model learns concepts from the 5% labeled data, then applies those learnings to the 95% unlabeled data, creating a more robust dataset for future predictions.

Semi-supervised learning: bank labels 5% of transactions, model learns and applies to remaining 95%

Discriminative models classify; generative models create

Discriminative models learn the relationship between data labels and classify new data points (e.g., fraud/not fraud, cat/dog). Generative models learn patterns in training data and generate something entirely new based on those patterns. A simple test: if output is a number, classification, or probability, it is not generative AI; if output is text, speech, image, or audio, it is generative AI.

Discriminative models classify existing data; generative models create new samples

Five main generative AI model types

Text-to-text models (ChatGPT, Google Bard) generate language. Text-to-image models (Midjourney, DALL-E, Stable Diffusion) generate and edit images. Text-to-video models (Google Imagen Video, CogVideo, Make a Video) generate and edit video. Text-to-3D models (OpenAI's Shap-E) create game assets. Text-to-task models (Gmail summarize feature in Bard) perform specific tasks like summarizing unread emails.

Five generative AI model types and their applications

LLMs are pre-trained then fine-tuned

Large language models are first pre-trained on massive datasets to solve common language problems like text classification, question answering, document summarization, and text generation. They are then fine-tuned using smaller, industry-specific datasets for specialized purposes in retail, finance, healthcare, entertainment, and other fields.

LLM development: pre-training on general data, then fine-tuning for specific domains

Real-world example: hospital medical diagnostics

A hospital uses a pre-trained large language model from a major tech company and fine-tunes it with its own first-party medical data to improve diagnostic accuracy from X-rays and other medical tests. This is a win-win: large companies develop general-purpose LLMs at scale, while smaller institutions like hospitals leverage those models with their domain-specific data without building from scratch.

LLMs and generative AI are related but distinct

Although there is overlap, large language models and generative AI are not the same thing. LLMs are a subset of deep learning and are pre-trained on large datasets then fine-tuned for specific purposes. Generative AI is a broader category that includes any model generating new content (text, image, video, audio, 3D).