
TLDR
Machine learning is a method of building AI where systems learn from data rather than being explicitly programmed with rules. It is the foundation of most modern AI, including language models and image generators.
Traditional software follows explicit rules written by programmers: if the customer clicks this button, do that. Machine learning flips this: show the system thousands of examples and let it figure out the rules itself.
A machine learning model is trained by adjusting millions of internal values (parameters) until the model's outputs match the expected outputs in the training data. This process, called gradient descent, is repeated billions of times during training.
Machine learning has three main categories: supervised learning (learning from labeled examples), unsupervised learning (finding patterns in unlabeled data), and reinforcement learning (learning by trial and error with rewards).
Deep learning is a subset of machine learning that uses neural networks with many layers. All modern large language models, image generators, and voice recognition systems are deep learning models.
Spam filtering
A spam filter is trained on millions of emails labeled spam or not spam. It learns to recognize patterns (certain words, sender domains, link patterns) that indicate spam.
Recommendation systems
Netflix, Spotify, and YouTube recommendation engines use machine learning to find patterns in what you watch and listen to, then predict what you will like next.
Language models
ChatGPT is a deep learning model trained with a combination of supervised learning (on curated text) and reinforcement learning from human feedback (RLHF).
AI is the broad goal of making computers intelligent. Machine learning is one method for achieving that goal. Most modern AI is built using machine learning, but not all AI is machine learning.
To use AI tools as an end user, no. To build and train ML models, yes: linear algebra, calculus, and statistics are important. Frameworks like PyTorch and TensorFlow handle much of the complexity.
Training a small model takes minutes. Training a frontier model like GPT-4 takes months on tens of thousands of specialized chips and costs tens or hundreds of millions of dollars.
Bottom line
Machine learning is a method of building AI where systems learn from data rather than being explicitly programmed with rules. It is the foundation of most modern AI, including language models and image generators.