Atlas

Post-training & Alignment

RLHF

Learning from human preference instead of ground truth.

You can jump straight in, but this star assumes Fine-tuning. Not recommended as a first stop.

Four layers of depth

Each layer ends with a quiz. Finish layer 4 and you own this concept.

  1. L1IntuitionWhat RLHF is, why plain fine-tuning isn't enough, and the buzzwords decoded.5m
    40 XP
  2. L2MechanicsThe PPO training loop step by step: rollouts, reward scoring, KL penalty, and the policy update.8m
    70 XP
  3. L3CodeImplement the core pieces of a PPO update for language models: reward shaping and the clipped objective.12m
    110 XP
  4. L4FoundationsThe KL-constrained RL objective in full, why it has a closed-form optimal policy, and PPO's variance-reduction math.15m
    180 XP

Where this leads

33 stars in the atlas.