Despite rapid progress in LLM agents, performance on long-horizon, tool-using tasks remains fragile. To better understand this fragility, we ask a simple question: \emph{do all actions contribute equally to failure?} Analyzing execution traces on $τ$-Bench (Airline/Retail) and SWE-Bench Verified, we decompose trajectories into \emph{mutating} (environment-changing) vs.\ non-mutating steps and formalize \emph{decisive deviations}, earliest action, level divergences that flip success to failure. A logistic regression reveals that each additional deviation in a mutating action reduces the odds of success by upto $92\%$ on Airline and upto $96\%$ on Retail for SoTA models. In contrast, deviations in non-mutating actions have little to no effect. Errors also grow with context length as agents drift from role and act on stale constraints. Motivated by these observations, we introduce \cm{}, a model-agnostic, gradient-free, test-time safeguard that (i) adds mutation-gated verification, (ii) injects \emph{Targeted Reflection} before mutating steps, and (iii) performs block-based context cleaning. \cm{} delivers consistent gains, e.g., Qwen3-Thinking: +28\% \emph{relative} on Airline, +11\% on Retail, and +7\% on SWE-Bench Verified; Claude: +9\%/+7\%. We further identify ceiling effects in $τ$-Bench, where annotation errors and underspecified tasks artificially cap model performance. To address this, we release $τ$-Bench Verified, which restores benchmark headroom through targeted revisions. Our results argue for action-level analysis, targeted safeguards, and reliable evaluations as prerequisites for robust multi-turn agents.

Diffusion language models (DLMs) are emerging as a powerful and promising alternative to the dominant autoregressive paradigm, offering inherent advantages in parallel generation and bidirectional context modeling. However, the performance of DLMs on code generation tasks, which have stronger structural constraints, is significantly hampered by the critical trade-off between inference speed and output quality. We observed that accelerating the code generation process by reducing the number of sampling steps usually leads to a catastrophic collapse in performance. In this paper, we introduce efficient Sampling with Adaptive acceleration and Backtracking Enhanced Remasking (i.e., Saber), a novel training-free sampling algorithm for DLMs to achieve better inference speed and output quality in code generation. Specifically, Saber is motivated by two key insights in the DLM generation process: 1) it can be adaptively accelerated as more of the code context is established; 2) it requires a backtracking mechanism to reverse the generated tokens. Extensive experiments on multiple mainstream code generation benchmarks show that Saber boosts Pass@1 accuracy by an average improvement of 1.9% over mainstream DLM sampling methods, meanwhile achieving an average 251.4% inference speedup. By leveraging the inherent advantages of DLMs, our work significantly narrows the performance gap with autoregressive models in code generation.

Large Language Models (LLMs) with safe-alignment training are powerful instruments with robust language comprehension capabilities. These models typically undergo meticulous alignment procedures involving human feedback to ensure the acceptance of safe inputs while rejecting harmful or unsafe ones. However, despite their massive scale and alignment efforts, LLMs remain vulnerable to jailbreak attacks, where malicious users manipulate the model to produce harmful outputs that it was explicitly trained to avoid. In this study, we find that the safety mechanisms in LLMs are predominantly embedded in the middle-to-late layers. Building on this insight, we introduce a novel white-box jailbreak method, SABER (Safety Alignment Bypass via Extra Residuals), which connects two intermediate layers $s$ and $e$ such that $s < e$, through a residual connection. Our approach achieves a 51% improvement over the best-performing baseline on the HarmBench test set. Furthermore, SABER induces only a marginal shift in perplexity when evaluated on the HarmBench validation set. The source code is publicly available at https://github.com/PalGitts/SABER.

The emergence of large-language models (LLMs) has enabled a new class of semantic data processing systems (SDPSs) to support declarative queries against unstructured documents. Existing SDPSs are, however, lacking a unified algebraic foundation, making their queries difficult to compose, reason, and optimize. We propose a new semantic algebra, SABER (Semantic Algebra Based on Extended Relational algebra), opening the possibility of semantic operations' logical plan construction, optimization, and formal correctness guarantees. We further propose to implement SABER in a SQL-compatible syntax so that it natively supports mixed structured/unstructured data processing. With SABER, we showcase the feasibility of providing a unified interface for existing SDPSs so that it can effectively mix and match any semantically-compatible operator implementation from any SDPS, greatly enhancing SABER's applicability for community contributions.

Large language models (LLMs) empowered by chain-of-thought reasoning have achieved impressive accuracy on complex tasks but suffer from excessive inference costs and latency when applied uniformly to all problems. We propose SABER (Switchable and Balanced Training for Efficient LLM Reasoning), a reinforcement learning framework that endows LLMs with user-controllable, token-budgeted reasoning. SABER first profiles each training example's base-model thinking token usage and assigns it to one of the predefined budget tiers. During fine-tuning, the model is guided by system prompts and length-aware rewards to respect its assigned budget. In parallel, we incorporate no-think examples to ensure the model remains reliable even when explicit reasoning is turned off. SABER further supports four discrete inference modes--NoThink, FastThink, CoreThink, and DeepThink, enabling flexible trade-offs between latency and reasoning depth. Extensive evaluations on math reasoning (MA TH, GSM8K), code generation (MBPP), and logical reasoning (LiveBench-Reasoning) demonstrate that SABER achieves high accuracy under tight budgets, graceful degradation, and effective cross-scale and cross-domain generalization. In particular, SABER-FastThink cuts reasoning length by 65.4% and yields a 3.6% accuracy gain compared with the base model on the MA TH benchmark.

Multimodal in-context learning (ICL) has emerged as a key capability of Large Vision-Language Models (LVLMs), driven by their increasing scale and applicability. Despite its promise, effective ICL in the multimodal setting remains challenging due to the inherent complexity of image-text inputs and the high sensitivity of ICL performance to input configurations. In this work, we shed light on the core mechanism underlying multimodal ICL, identifying task mapping as a crucial factor in configuring robust in-context demonstration (ICD) sequences. Building on these insights, we propose \textit{SabER}, a lightweight yet powerful decoder-only transformer equipped with task-aware attention, which intelligently selects and arranges ICDs from a demonstration library in an autoregressive fashion. This design enables fine-grained feature extraction and cross-modal reasoning, iteratively refining task mapping to generate high-quality ICD sequences. Through extensive experiments covering five LVLMs and nine benchmark datasets, SabER not only demonstrates strong empirical performance, but also provides deeper understanding of how task semantics interact with multimodal ICDs. Our findings highlight the importance of principled ICD sequence configuration and open new avenues to enhance multimodal ICL in a wide range of real-world scenarios.

Large language models (LLMs) have demonstrated impressive impact in the field of natural language processing, but they still struggle with several issues regarding, such as completeness, timeliness, faithfulness and adaptability. While recent efforts have focuses on connecting LLMs with external knowledge sources, the integration of knowledge bases (KBs) remains understudied and faces several challenges. In this paper, we introduce KnowledGPT, a comprehensive framework to bridge LLMs with various knowledge bases, facilitating both the retrieval and storage of knowledge. The retrieval process employs the program of thought prompting, which generates search language for KBs in code format with pre-defined functions for KB operations. Besides retrieval, KnowledGPT offers the capability to store knowledge in a personalized KB, catering to individual user demands. With extensive experiments, we show that by integrating LLMs with KBs, KnowledGPT properly answers a broader range of questions requiring world knowledge compared with vanilla LLMs, utilizing both knowledge existing in widely-known KBs and extracted into personalized KBs.

Developing monolingual large Pre-trained Language Models (PLMs) is shown to be very successful in handling different tasks in Natural Language Processing (NLP). In this work, we present AraMUS, the largest Arabic PLM with 11B parameters trained on 529GB of high-quality Arabic textual data. AraMUS achieves state-of-the-art performances on a diverse set of Arabic classification and generative tasks. Moreover, AraMUS shows impressive few-shot learning abilities compared with the best existing Arabic PLMs.

Technological advances in holography, robotics, and 3D printing are starting to realize the vision of a holodeck. These immersive 3D displays must address user safety from the start to be viable. A holodeck's safety challenges are novel because its applications will involve explicit physical interactions between humans and synthesized 3D objects and experiences in real-time. This pioneering paper first proposes research directions for modeling safety in future holodeck applications from traditional physical human-robot interaction modeling. Subsequently, we propose a test-bed to enable safety validation of physical human-robot interaction based on existing augmented reality and virtual simulation technology.

The Arcade Learning Environment (ALE) is proposed as an evaluation platform for empirically assessing the generality of agents across dozens of Atari 2600 games. ALE offers various challenging problems and has drawn significant attention from the deep reinforcement learning (RL) community. From Deep Q-Networks (DQN) to Agent57, RL agents seem to achieve superhuman performance in ALE. However, is this the case? In this paper, to explore this problem, we first review the current evaluation metrics in the Atari benchmarks and then reveal that the current evaluation criteria of achieving superhuman performance are inappropriate, which underestimated the human performance relative to what is possible. To handle those problems and promote the development of RL research, we propose a novel Atari benchmark based on human world records (HWR), which puts forward higher requirements for RL agents on both final performance and learning efficiency. Furthermore, we summarize the state-of-the-art (SOTA) methods in Atari benchmarks and provide benchmark results over new evaluation metrics based on human world records. We concluded that at least four open challenges hinder RL agents from achieving superhuman performance from those new benchmark results. Finally, we also discuss some promising ways to handle those problems.