See It Work

Before we touch any math or Python code, run two 5-minute experiments. No machine learning background needed.

• How AI Chat Works: When a chatbot replies, it looks like it planned the whole paragraph in advance. In reality, it only ever guesses the very next token, glues that choice onto the input, and repeats.
• Why Transformers Won: To guess the next token, a model often needs a clue from many words earlier. You’ll test why passing a single rolling memory (Chapter 2: RNNs) fades on long sentences, and how looking back directly (Chapter 3: Attention) fixes it.

Start with the first experiment: How AI Chat Works →