Next-token Prediction

Lesson goals

• Probabilities over alternatives
• Token-by-token generation
• Sampling affects outputs

Core idea

The basic mechanism behind writing, summarization, and reasoning behavior.

1. Probabilities over alternatives: understand it in context, identify inputs and outputs, and define how the result will be verified.

2. Token-by-token generation: understand it in context, identify inputs and outputs, and define how the result will be verified.

3. Sampling affects outputs: understand it in context, identify inputs and outputs, and define how the result will be verified.

Practical example

Consider a real application of “Next-token Prediction.” Start small, identify available data, and compare the system with a clear baseline before expanding it.

Check your understanding

1. Explain “Next-token Prediction” in your own words.
2. Give one appropriate and one misleading application.
3. What risk or limitation should be monitored?