How do AI, machine learning, and LLMs fit inside each other?

The clearest way to picture these terms is as a set of Russian dolls, each one inside the next.

The outer doll is artificial intelligence: any system that does intelligent-seeming work, learned or hand-written. Inside it sits machine learning, the part of AI where the system learns patterns from data instead of following rules a person wrote out.^[1] That nesting gives a useful rule of thumb: all machine learning is AI, but not all AI is machine learning. A pocket calculator or a rule-based program can be AI without ever learning anything.

Inside machine learning sits deep learning, which uses neural networks with many stacked layers to do tasks like classification and prediction.^[2] “Deep” is a literal description: the layer count can range “from three to several hundred or thousands,” and most researchers draw the line at more than two.^[2] Deep learning is what made the recent jump in AI possible, because these layered networks can pick up far more subtle patterns than earlier methods. Older machine learning “often involved hand-crafted feature engineering,” a person deciding by hand what to measure in the data, while “a deep learning process can learn which features to optimally place at which level on its own.”^[2] That’s the real division of labor: shallower ML still needs a person to tell it what to look for, deep learning works that out for itself.

The innermost doll, for our purposes, is the large language model. An LLM is a neural network, so it is a deep-learning system, trained on huge amounts of text.^[3] So when you use an LLM, you are using a large language model, which is a deep-learning model, which is machine learning, which is AI. Four labels, one object.