This paper systematically studies how soft errors propagate during Large Language Model (LLM) inference using a novel fault-injection framework, providing critical insights and mitigation strategies f…

This paper demonstrates that Concept Bottleneck Models (CBMs), despite their interpretability, are highly vulnerable to targeted adversarial attacks that manipulate semantic concepts, and proposes SPE…

The paper analyzes the failure modes of aggressive 2-bit quantization in large reasoning models, proposing lightweight controls like FP16 planning and loop rescue to restore accuracy and achieve pract…

This systematic mapping survey reviews label-efficient approaches for code vulnerability detection, synthesizing five paradigm families and providing a decision guide to navigate trade-offs.

The paper introduces TELBench and the DRIFT framework to enable fine-grained, span-level error localization in deep-research agents, significantly improving the ability to pinpoint exactly where an ag…

This paper evaluates the causal reasoning abilities of large language models and finds that they rely heavily on lexical pattern matching rather than structural reasoning.

The paper analyzes LLM vulnerability detection using mechanistic interpretability, finding that models primarily rely on safety detectors rather than direct vulnerability signature recognition.

This paper proposes a method to recover recoverability structure from failed traces of post-trained language models, enabling test-time routing and post-training analysis.

This paper proposes DeMix, a novel framework for simultaneously diagnosing erroneous samples and their error types in machine learning models.

The paper introduces MemTrace, a framework that treats LLM memory pipelines as traceable graphs to systematically diagnose and automatically correct memory-related errors, boosting performance by up t…

This paper analyzes failure modes in collaborative visual reasoning systems, demonstrating that naive shared workspaces can amplify hallucinations and proposing diagnostics for improving communication…

This paper investigates how different types of compressed reasoning data (Explicit, Composed, Implicit CoT) affect LLM performance during post-training, finding that the choice of compression and subs…

The paper introduces an Item Response Theory (IRT)-based indicator that effectively identifies likely mislabeled items in existing LLM benchmarks, revealing systematic errors in labeling and model spe…

The paper proposes projectional decoding, a novel framework that integrates a partial graph model alongside text generation to ensure the semantic validity of LLM-generated software artifacts.

The paper introduces Thinking as Compression (TaC), a novel paradigm showing that the inherent reasoning process of a large language model can naturally compress long context inputs, outperforming ded…

The paper reframes LLM reliability from an impossible universal problem to a manageable, local patch-based problem, showing that sufficient interventions can be found by focusing on recurring failure…

MOSAIC is a multi-objective framework that efficiently allocates a fixed supervised fine-tuning budget by turning failure profiles into actionable data mixtures, significantly improving model alignmen…

The paper theoretically explains that optimizing LLMs solely on outcomes leads to brittle reasoning (Reward-Induced Manifold Collapse) by favoring low-complexity shortcuts, and proposes process-based…

The paper introduces 5WBENCH, a new benchmark for causal unlearning, and proposes MAAT, a novel three-phase framework that achieves high forgetting and high retention specifically on complex 'Why'-typ…

The paper introduces Safety Bottleneck Regularization (SBR), a novel defense mechanism that anchors LLM safety by constraining the unembedding layer, effectively preventing harmful fine-tuning (HFT) e…