Rethinking Agentic RAG: Toward LLM-Driven Logical Retrieval Beyond Embeddings

Recent advances in RAG have shifted toward an agentic paradigm, where LLMs interact with retrieval systems over multiple turns and iteratively refine queries based on intermediate results. At the same time, LLMs have demonstrated a strong ability to construct structured queries that precisely express their information needs. However, contemporary RAG systems remain heavily focused on engineering complex retrieval backends, including dense, hybrid, and graph-based retrieval architectures. In this study, we argue that agentic RAG should delegate greater control to the LLM to steer the retrieval process, while relying on a lightweight retrieval interface that provides fine-grained control and faithfully executes the LLM's structured intent. Guided by this principle, we propose an agentic RAG framework that enables LLMs to formulate retrieval intents using logical expressions while simplifying the retrieval backend to an inverted-index-based system. Extensive experiments show that our framework matches a strong agentic hybrid baseline, while substantially reducing construction and serving cost. Moreover, we show that anchoring the retrieval process in logical queries substantially reduces hallucinations in generated responses.

Reward Hacking in the Era of Large Models: Mechanisms, Emergent Misalignment, Challenges

Reinforcement Learning from Human Feedback (RLHF) and related alignment paradigms have become central to steering large language models (LLMs) and multimodal large language models (MLLMs) toward human-preferred behaviors. However, these approaches introduce a systemic vulnerability: reward hacking, where models exploit imperfections in learned reward signals to maximize proxy objectives without fulfilling true task intent. As models scale and optimization intensifies, such exploitation manifests as verbosity bias, sycophancy, hallucinated justification, benchmark overfitting, and, in multimodal settings, perception--reasoning decoupling and evaluator manipulation. Recent evidence further suggests that seemingly benign shortcut behaviors can generalize into broader forms of misalignment, including deception and strategic gaming of oversight mechanisms. In this survey, we propose the Proxy Compression Hypothesis (PCH) as a unifying framework for understanding reward hacking. We formalize reward hacking as an emergent consequence of optimizing expressive policies against compressed reward representations of high-dimensional human objectives. Under this view, reward hacking arises from the interaction of objective compression, optimization amplification, and evaluator--policy co-adaptation. This perspective unifies empirical phenomena across RLHF, RLAIF, and RLVR regimes, and explains how local shortcut learning can generalize into broader forms of misalignment, including deception and strategic manipulation of oversight mechanisms. We further organize detection and mitigation strategies according to how they intervene on compression, amplification, or co-adaptation dynamics. By framing reward hacking as a structural instability of proxy-based alignment under scale, we highlight open challenges in scalable oversight, multimodal grounding, and agentic autonomy.