Lesson 4: Reading one word at a time, and why it hit a wall
17 min read·13 Oct 2026
Module 5 · Lesson 4 · About 35 minutes · Needs: Module 4 Lesson 1 (tensors, a training loop, the 32/8 ticket split); no model key needed
The situation
Dev has been reading ahead. "Every older tutorial builds text models from RNNs and LSTMs. This course goes from embeddings straight to transformers. Are we skipping something?"
Yes, on purpose. You will build the smallest useful recurrent neural network (RNN) in PyTorch, train it in under a second on the 32 training tickets, and watch it read "The vacuum stopped charging after 3 months." one word at a time. Then you will measure the two problems that made the field move on. Module 6 is the answer to both.