Artificial Intelligence
280 views
Reward Modeling
Quick Definition
Training models to predict human preferences for RLHF
Full Definition
Training a model to predict human preferences and provide reward signals for reinforcement learning from human feedback.
Examples
RLHF training pipeline, preference learning, alignment
Related Terms
rlhf
ai-alignment
reinforcement-learning
More Artificial Intelligence Terms
Word Embedding
Dense vector representation capturing word semantic meaning
Feature Importance
Quantifying each input features contribution to predictions
Symbolic AI
AI using symbolic logic and explicit knowledge rules
Prompt Injection
Attack manipulating AI behavior by overriding instructions
Knowledge Graph
Structured graph of entities and their relationships
Chain-of-Thought
Technique encouraging AI to show step-by-step reasoning