AlgoPlus//machine learning / rlhf
Read the theory

Fine-tuning & RLHF

Align a base model with human feedback: fine-tune, learn a reward, optimize with RL.

Stage
Legend
Active stage
Preferred / high reward
AI Tutor Workspace
In a nutshell
A pretrained base model can predict text but isn't tuned to what people actually want. RLHF aligns it in three stages. First, supervised fine-tuning: keep training on human-written ideal answers to teach the tone of a helpful assistant. Second, build a reward model: people repeatedly pick the better of two answers, and those choices train a separate model to score any response. Third, reinforcement learning: the model (the policy) writes answers, the reward model scores them, and the policy is nudged to earn higher scores — while a leash to the base model keeps it from drifting into nonsense. The payoff is an assistant tuned to be helpful, harmless, and honest.
Ready
Press play to begin the cinematic walkthrough.
Teach taste, not just facts. Show good examples, then let people pick the better of two answers thousands of times, distill that into a 'critic', and let the model practice pleasing the critic.
Key terms
Go deeper in the lesson
Read the full theory, intuition & complexity for Fine-tuning & RLHF.