RLAIF
Reinforcement Learning from AI Feedback — replaces human raters in RLHF with an AI rater. Anthropic + Google papers 2023.
Pitch: scales endlessly (no recruiting of raters). Risk: bias amplification, since the AI rater has its own biases. Comparable quality to RLHF on most benchmarks per Google paper "RLAIF vs RLHF" (2023). Used by Claude (via Constitutional AI), Gemini, several open-source models.