AL

We propose REAL (Response Embedding-based Alignment for LLMs), a method to improve alignment efficiency by selecting less ambiguous, dissimilar response pairs for annotation. By leveraging embedding similarity in an off-policy manner, REAL reduces label noise and improves alignment quality. Experiments show it boosts performance while cutting annotation effort by up to 65%.

DWDS is a density-weighted diversity strategy for active learning in deep learning. It selects informative and representative samples using geometric insights and a beam search for efficient query selection. DWDS consistently outperforms existing methods under limited labeling budgets.