Deep Reinforcement Learning from Human Preferences

Major category

Foundational work

Quadrant

Not applicable

Architecture

Not applicable

Prediction paradigm

Not applicable

Subcategories

Training optimization & distillation

Source review status

Explicit in survey

Category review. This 2017 preference-learning foundation couples human clip comparisons, a learned scalar reward and model-free policy optimization. It supplies historical RLHF methodology for later post-training, not future-state modeling or a language/vision/action representation module. Reading evidence