Skip to content
AEO Canon · the reference for answer-engine optimization

Lesson 3 of 6

How AI Models Are Trained

Models learn in stages — pretraining, fine-tuning, RLHF — from huge amounts of text. This lesson shows where their abilities and blind spots come from.

Learning objectives

  • Outline the three stages of training an LLM.
  • Explain what training data is and where it comes from.
  • Explain why a trained model is frozen at a knowledge cutoff.

The lesson

Read the full lesson →How Are AI Models Trained?AI models are trained in stages — large-scale pretraining on text to learn language, then fine-tuning and reinforcement learning from human feedback (RLHF) to make them helpful, honest, and safe to use.3 min read

Key takeaways

  • Pretraining (learn language) → fine-tuning (follow instructions) → RLHF (be helpful & safe).
  • Training data is trillions of tokens from the web, books, code, and licensed sources.
  • Training is a frozen snapshot — built-in knowledge stops at the cutoff date.

Knowledge check

Knowledge check

0 / 3

  1. 1. Which is the correct training order?

  2. 2. What is RLHF for?

  3. 3. Why can a model be confidently out of date?