How Machines Think
Data Is the Ingredient
6 min+50 XPStep 1/9
Data is what learning eats
A learning system needs examples — lots of them. To teach a model 'cat vs dog', you show thousands of labeled photos. To teach a spam filter, labeled emails. To teach a translator, millions of sentence pairs.
The examples are the dataset, and the dataset decides what the model can learn. No data, no learning. Bad data, bad learning.
insight
Think of the dataset as the textbook the model studies. If the textbook is wrong, incomplete, or skewed, the model graduates with the same problems.