How AI Actually Works

Ideas to remember

1. 01AI, machine learning and deep learning nest like boxes: all deep learning is machine learning, and all machine learning is AI, but not the other way round.
2. 02Machine learning learns from examples instead of hand written rules; a spam filter is the classic case.
3. 03You already use AI in maps, recommendations, autocomplete and photo tagging, and each result is a best guess that can be wrong.
4. 04A large language model predicts the next word, so it can write a fluent, confident answer that is simply false, called a hallucination.
5. 05Training is the slow, costly learning done once by the builder; inference is the quick everyday use, and the model does not keep learning from you.
6. 06A model knows only its training data, so gaps and slants in that data shape and can bias every answer.

Words

Artificial intelligence (AI)

The broad goal of getting computers to do tasks that normally need human intelligence.

Machine learning (ML)

A method where the system is shown many examples and works out the patterns for itself.

Deep learning

Machine learning with neural networks built in many stacked layers.

Large language model (LLM)

AI trained on enormous amounts of text that builds an answer one word at a time.

Hallucination

An answer that sounds right but is false, written in the same assured style as a correct one.

Training vs inference

Training builds the model from data once; inference is each everyday use of it.