Data and labels, three ways to learn, overfitting, and how language models write.
Machine learning is how most modern AI gets its abilities. The idea is simple: rather than telling a computer exactly what to do, we give it examples and let it find the pattern.
Everything starts with data: rows of examples. Each example is described by features, the measurable details, such as a house's size, location and age. When we also know the right answer for each example, such as the price the house actually sold for, that answer is called the label.
A model is trained on one part of the data and tested on a separate part it has never seen. The test tells us how well it will do in the real world.
If a model scores brilliantly on its training data but badly on new data, it has overfitted: it memorised the examples instead of learning the general pattern. A good model generalises.
The chat assistants you may use every day are built on large language models (LLMs). An LLM is trained on a vast amount of text to do one thing extremely well: predict the next small piece of text, called a token, given everything before it. Repeat that prediction word after word and you get fluent paragraphs.
Because the model is predicting what sounds right rather than looking facts up, it can produce confident statements that are false. This is called a hallucination.
Comments
Loading comments…