Glossary
RLHF (Reinforcement Learning from Human Feedback)
A way of training AI models by having humans rate or compare outputs, then using those ratings to reinforce better behavior over time.
RLHF is one of the main reasons modern AI assistants feel helpful rather than robotic. Instead of learning only from text on the internet, the model gets feedback from people — for example, raters might choose which of two responses is more useful, accurate, or appropriately cautious. That feedback is used to push the model toward producing more of what humans prefer. Most of the AI models you interact with today — ChatGPT, Claude, Gemini — went through some version of this process. As a behavior architect, understanding RLHF helps you understand why models have the tendencies they do, and why changing behavior often requires going back to the training process rather than just adjusting a prompt.