Detailed explanation
The agent observes a state, picks an action and receives a reward. It updates its policy to maximize cumulative reward. Used in robotics, games, optimization, and LLM alignment via RLHF.
Examples
AlphaGo beating Go champions
Robots learning to walk
RLHF for aligning ChatGPT
Ad bidding optimization
Frequently asked questions
What is RLHF?
Reinforcement Learning from Human Feedback: the model is trained to prefer responses humans rated as better.
Related terms
Deep Learning
A subfield of machine learning using deep neural networks to learn complex representations.
Machine Learning
A branch of AI where models learn patterns from data instead of following explicit rules.
Supervised Learning
Learning from labeled examples (input → expected output).
Unsupervised Learning
Learning from unlabeled data, where the model discovers structures by itself.
Last updated: 7/15/2026