Foundations · 1 min read · Updated 2026-10-05
Deep learning and neural networks
Layers build up understanding
In a photo, the first layers notice simple things like edges. Later layers notice shapes, like ears or eyes. The last layers decide it is a cat.
How it learns
At the start, the connections are random, so the answers are wrong. Each time it gets one wrong, it nudges the connections a tiny bit to do better. After millions of nudges, it works.
Why now
Deep learning needs lots of examples and very fast computer chips called GPUs. Around 2012 both became available, and progress took off.
Key takeaways
- A neural network learns by adjusting connections.
- Many layers let it understand complex things step by step.
- Big data and fast chips made it work.
Quick questions
What is a parameter?
One of the adjustable connections inside the network. Big models have billions.
Why GPUs?
Training needs a huge amount of simple sums done at once, and GPUs are built for that.
Is it like a real brain?
Only loosely. It is much simpler than a real brain.
Keep learning
AI vs machine learning vs deep learningAI means computers doing things that seem smart. Machine learning is how most AI is built today: the computer learns from…1 min readRead →Transformers and attentionThe transformer is the design that powers most chatbots. Its key trick, called attention, lets the computer look at all the…1 min readRead →