Three kinds of learning signal
Every learner needs something that says “better” or “worse”. Labels, structure in the data, or a reward — where that signal comes from decides almost everything else.
40 min
After this lesson you can
- Classify a learning problem by the source of its training signal
- Run k-means by hand and explain why it always settles, but not always in the best place
- Explain how self-supervised learning manufactures labels out of raw data
- Place pretraining, supervised fine-tuning, RLHF and RLVR on the same map
In this lesson
- The map
- Supervised learning: someone supplied the answer
- Unsupervised learning: structure without answers
- Self-supervised learning: labels manufactured from the data
- Reinforcement learning: a score for what you did
- Where a chat model sits
This lesson is part of Descent Pro
Descent Pro unlocks every lesson on the route, from base camp to the summit. Every plan starts with a free trial.