How do alignment techniques such as RLHF and DPO affect the performance of LLMs on LLM-as-a-Judge benchmarks l

Abstract The rapid evolution of large language models (LLMs) has driven a transformative shift in artificial intelligence (AI), reshaping both research paradigms and practical applications. Distinguished from their predecessors by unprecedented scale and advanced capabilities, LLMs necessitate new frameworks for understanding their development, behavior, and societal impact. This survey systematically reviews recent advancements in LLM techniques across four key dimensions: (1) pre-training methodologies, which establish core model capabilities through large-scale self-supervised training, arc

Research goal: How do alignment techniques such as RLHF and DPO affect the performance of LLMs on LLM-as-a-Judge benchmarks like MT-Bench, and what are the trade-offs in terms of helpfulness vs. harmfulness scores?

Autonomous synthesis report generated by SOVEREIGN Research Kernel. Tribunal consensus score: 9.2/10.