Retrieval-Augmented Generation (RAG) mitigates hallucinations by grounding generation in retrieved evidence, yet failures persist when models underutilize the retrieved context. While prior work attributes this to conflicts between external context and internal parametric knowledge encoded in Feed-Forward Networks (FFNs), our empirical analyses reveal that FFN-derived signals are unstable and insufficient for reliable diagnosis across diverse contexts. We therefore propose the Attention Escape Score (AES), a training-free diagnostic framework that provides a mechanistic lens on RAG hallucinations via dynamic attention routing failures. We models the failures as weakened Semantic Anchoring, increased Contextual Distraction, and heightened Generation Inertia, moving beyond static component-level analyses. Extensive experiments show that AES significantly outperforms state-of-the-art baselines with robust cross-model generalization. Furthermore, we introduce Pruning Escape Routes (PER), a lightweight inference-time intervention that suppresses hallucinations by attenuating identified escape pathways, achieving improved faithfulness with negligible degradation to complex reasoning and general capabilities.

Automatic harmfulness metrics and LLM-based judges are central to LLM safety assessment, yet differences in evaluation criteria, methodologies, and output formats may yield inconsistent judgments for the same model response. The lack of systematic research on these evaluators undermines the credibility and consistency of LLM safety assessments. To bridge this gap, we introduce HarmMetric Eval, a benchmark for both overall and fine-grained assessment of harmfulness evaluators. We construct a dataset of representative harmful prompts paired with diverse harmful responses and non-harmful counterparts across multiple categories. We propose a flexible scoring mechanism that rewards metrics for correctly ranking harmful responses above non-harmful ones, making it applicable to nearly all metrics and judges with varying output formats and scoring scales. Using HarmMetric Eval, we find that conventional reference-based metrics such as ROUGE and METEOR can outperform existing LLM-based judges in fine-grained harmfulness evaluation, challenging prevailing assumptions about LLM superiority. Fine-grained analysis reveals that LLM judges tend to overestimate irrelevant and useless responses. Furthermore, we build a new harmfulness judge by incorporating fine-grained criteria into its prompt template and leveraging reference-based metrics to fine-tune its base LLM. The resulting judge demonstrates leading performance on HarmMetric Eval and two external human-labeled datasets.

Ensuring Large Language Model (LLM) safety is crucial, yet the lack of a clear understanding about safety mechanisms hinders the development of precise and reliable methodologies for safety intervention across diverse tasks. To better understand and control LLM safety, we propose the Expected Safety Impact (ESI) framework for quantifying how different parameters affect LLM safety. Based on ESI, we reveal distinct safety-critical patterns across different LLM architectures: In dense LLMs, many safety-critical parameters are located in value matrices (V) and MLPs in middle layers, whereas in Mixture-of-Experts (MoE) models, they shift to late-layer MLPs. Leveraging ESI, we further introduce two targeted intervention paradigms for safety enhancement and preservation, i.e., Safety Enhancement Tuning (SET) and Safety Preserving Adaptation (SPA). SET can align unsafe LLMs by updating only a few safety-critical parameters, effectively enhancing safety while preserving original performance. SPA safeguards well-aligned LLMs during capability-oriented intervention (e.g., instruction tuning) by preventing disruption of safety-critical weights, allowing the LLM to acquire new abilities while maintaining safety capabilities. Extensive evaluations on different LLMs demonstrate that SET can reduce the attack success rates of unaligned LLMs by over 50% with only a 100-iteration update on 1% of model weights. SPA can limit the safety degradation of aligned LLMs within 1% after a 1,000-iteration instruction fine-tuning on different tasks. Our code is available at: https://github.com/ZJU-LLM-Safety/SafeWeights-ACL

Code optimization remains a core objective in software development, yet modern compilers struggle to navigate the enormous optimization spaces. While recent research has looked into employing large language models (LLMs) to optimize source code directly, these techniques can introduce semantic errors and miss fine-grained compiler-level optimization opportunities. We present HintPilot, which bridges LLM-based reasoning with traditional compiler infrastructures via synthesizing compiler hints—annotations that steer compiler behavior. HintPilot employs retrieval-augmented synthesis over compiler documentation and applies profiling-guided iterative refinement to synthesize semantics-preserving and effective hints. Upon PolyBench and HumanEval-CPP benchmarks, HintPilot achieves up to 6.88x geometric mean speedup over -Ofast while preserving program correctness.

Fine-tuning large language models (LLMs) on domain-specific datasets has become a standard paradigm for adapting LLMs to specialized applications. However, recent work has shown that even fine-tuning on benign task-specific data can substantially weaken the safety capabilities of LLMs. While existing efforts have made progress in identifying data responsible for safety degradation, they usually rely on a single mean vector computed over a specific model with its tokenizer to represent the safety direction, which limits both the effectiveness and transferability of their risk assessment measures. To address these limitations, we propose DataShield, a data assessment framework that identifies risky fine-tuning samples and response segments through consensus subspace alignment over joint safety-critical semantic spaces derived from multiple safety-aligned LLMs.Within these spaces, DataShield extracts consensus safe and unsafe subspaces using semantic spectral decomposition over safe and unsafe data representations. The risk of a data sample or segment is then estimated by measuring its relative alignment with the unsafe and safe subspaces, enabling both sample-level filtering and fine-grained segment-level masking. Compared with state-of-the-art filtering and masking baselines, DataShield reduces ASR by 14.6% with sample filtering and 32.3% with segment masking, while preserving downstream utility and avoiding target-model-specific risk computation.

Clinical agents promise to democratize access to electronic health records (EHRs), yet existing benchmarks fail to reflect the complexity of practical EHR analysis, e.g., often operating on idealized, clean EHRs via static SQL generation rather than interactive execution. In this work, we introduce **EHR-Complex**, a large-scale benchmark designed for interactive clinical database reasoning. Built on the large MIMIC-IV substrate (365K patients, 31 tables, 500M+ records), EHR-Complex comprises about 52K tasks spanning six clinical intents, supporting both patient-level and population-level queries, where each task requires an agent to interact with a sandboxed environment by executing SQL queries or Python code. Notably, EHR-Complex considers the real-world SQL task complexity for longitudinal multi-table aggregation and compositional reasoning, resulting in 31.93 SQL structural components per query on average. Evaluation results on EHR-Complex reveal the clinical difficulty of these EHR reasoning scenarios, with the top-performing model achieving only 62.3% exact-match accuracy. Pass^k consistency drops below 50% for nearly all evaluated models at k=4, exposing broad stochastic fragility. A fine-grained analysis of more than 3,800 failed trajectories for representative LLMs reveals three dominant failure modes: SQL logic errors, medical-code lookup failures, and semantic misunderstandings. EHR-Complex provides a rigorous testbed for clinical agents and highlights remaining gaps in robust reasoning for large-scale EHR analysis.

Continual Pre-training (CPT) serves as a fundamental approach for adapting foundation models to domain-specific applications. Scaling laws for pre-training define a power-law relationship between dataset size and the test loss of an LLM. However, the marginal gains from simply increasing data for CPT diminish rapidly, yielding suboptimal data utilization and inefficient training. To address this challenge, we propose a novel perplexity-aware data scaling law to establish a predictive relationship between the perplexity landscape of domain-specific data and the test loss. Our approach leverages the pre-trained model’s own perplexity on domain data as a proxy for estimating the knowledge gap, effectively quantifying the informational perplexity landscape of candidate training samples. By fitting this scaling law across diverse perplexity regimes, we enable adaptive selection of high-utility data subsets, prioritizing content that maximizes knowledge absorption while minimizing redundancy and noise. Extensive experiments on both medical and general-domain benchmarks demonstrate that our method consistently identifies near-optimal training subsets, achieving superior performance with significantly reduced data consumption.

Privacy policies are essential for users to understand how service providers handle their personal data. However, these documents are often long and complex, as well as filled with technobabble and legalese, causing users to unknowingly accept terms that may even contradict the law. While summarizing and interpreting these privacy policies is crucial, there is a lack of high-quality English parallel corpus optimized for legal clarity and readability. To address this issue, we introduce APPSI-139, a high-quality English privacy policy corpus meticulously annotated by domain experts, specifically designed for summarization and interpretation tasks. The corpus includes 139 English privacy policies, 15,692 rewritten parallel corpora, and 36,351 fine-grained annotation labels across 11 data practice categories. Concurrently, we propose TCSI-pp-V2, a hybrid privacy policy summarization and interpretation framework that employs an alternating training strategy and coordinates multiple expert modules to effectively balance computational efficiency and accuracy. Experimental results show that the hybrid summarization system built on APPSI-139 corpus and the TCSI-pp-V2 framework outperform large language models, such as GPT-4o and LLaMA-3-70B, in terms of readability and reliability. The source code and dataset are available at https://github.com/EnlightenedAI/APPSI-139.

Recently, the self-consistency decoding strategy has shown the ability to improve performance for complex reasoning tasks with large language models (LLMs). However, the costs may be high because the sampling process of the strategy generates some low-probability text, resulting in low-quality reasoning paths. As a consequence, it requires a relatively large sampling number to obtain good aggregation performance. In this paper, we propose an alternative strategy, self-para-consistency. It first generates multiple paraphrases for each test question, then generates reasoning paths for the original and all the paraphrased questions based on greedy decoding, and finally selects the most consistent answer. Since all the candidate paths have relatively high probabilities, the sampling number could be much smaller than the self-consistency strategy. Extensive experiments on complex reasoning datasets demonstrate the effectiveness of our method in reducing the sampling number.