Yubo Li

Large language models systematically fail when a salient surface cue conflicts with an unstated feasibility constraint. We study this through a diagnose-measure-bridge-treat framework. Causal-behavioral analysis of the ‘car wash problem’ across six models reveals approximately context-independent sigmoid heuristics: the distance cue exerts 8.7 to 38 times more influence than the goal, and token-level attribution shows patterns more consistent with keyword associations than compositional inference. The Heuristic Override Benchmark (HOB) – 500 instances spanning 4 heuristic by 5 constraint families with minimal pairs and explicitness gradients – demonstrates generality across 14 models: under strict evaluation (10/10 correct), no model exceeds 75%, and presence constraints are hardest (44%). A minimal hint (e.g., emphasizing the key object) recovers +15 pp on average, suggesting the failure lies in constraint inference rather than missing knowledge; 12/14 models perform worse when the constraint is removed (up to -39 pp), revealing conservative bias. Parametric probes confirm that the sigmoid pattern generalizes to cost, efficiency, and semantic-similarity heuristics; goal-decomposition prompting recovers +6 to 9 pp by forcing models to enumerate preconditions before answering. Together, these results characterize heuristic override as a systematic reasoning vulnerability and provide a benchmark for measuring progress toward resolving it.

Agent Skills are structured packages of procedural knowledge that augment LLM agents at inference time. Despite rapid adoption, there is no standard way to measure whether they actually help. We present SkillsBench, a benchmark of 86 tasks across 11 domains paired with curated Skills and deterministic verifiers. Each task is evaluated under three conditions: no Skills, curated Skills, and self-generated Skills. We test 7 agent-model configurations over 7,308 trajectories. Curated Skills raise average pass rate by 16.2 percentage points (pp), but effects vary widely by domain (+4.5pp for Software Engineering to +51.9pp for Healthcare) and 16 of 84 tasks show negative deltas. Self-generated Skills provide no benefit on average, showing that models cannot reliably author the procedural knowledge they benefit from consuming. Focused Skills with 2-3 modules outperform comprehensive documentation, and smaller models with Skills can match larger models without them. We publish the dataset and evaluation harness to assist developers and researchers in future work at skillsbench.ai.

Most work on confidence in large language models has focused on estimation, uncertainty quantification, and calibration. In deployed systems, however, the key question is how confidence should be used to govern behavior. This survey studies confidence utilization: the use of confidence-related signals to control system decisions. We formalize this perspective through a unified framework in which confidence is defined over decision units under a local state and then consumed by a policy to determine actions. Using this lens, we organize the literature across full LLM lifecycle: training, inference, model selection and cascading, retrieval-augmented generation, risk management, and agentic control. We compare methods by signal source, decision unit, and functional role, and conclude by highlighting open challenges in confidence semantics, composition, source attribution, decision-aware evaluation, and robustness. Overall, the survey positions confidence not only as an estimation target, but as a control primitive for building more reliable and trustworthy LLM systems.

Large Language Models (LLMs) have revolutionized conversational AI, yet their robustness in extended multi-turn dialogues remains poorly understood. Existing evaluation frameworks focus on static benchmarks and single-turn assessments, failing to capture the temporal dynamics of conversational degradation that characterize real-world interactions. In this work, we present the first comprehensive survival analysis of conversational AI robustness, analyzing 36,951 conversation turns across 9 state-of-the-art LLMs to model failure as a time-to-event process.

Recent advancements in large language models (LLMs) have revolutionized their ability to handle single-turn tasks, yet real-world applications demand sophisticated multi-turn interactions. This survey provides a comprehensive review of recent advancements in evaluating and enhancing multi-turn interactions in LLMs. Focusing on task-specific scenarios—from instruction following in diverse domains such as math and coding to complex conversational engagements in roleplay, healthcare, education, and even adversarial jailbreak settings—we systematically examine the challenges of maintaining context, coherence, fairness, and responsiveness over prolonged dialogues. The paper organizes current benchmarks and datasets into coherent categories that reflect the evolving landscape of multi-turn dialogue evaluation. In addition, we review a range of enhancement methodologies under multi-turn settings, including model-centric strategies (contextual learning, supervised fine-tuning, reinforcement learning, and new architectures), external integration approaches (memory-augmented, retrieval-based methods, and knowledge graph), and agent-based techniques for collaborative interactions. Finally, we discuss open challenges and propose future directions for research to further advance the robustness and effectiveness of multi-turn interactions in LLMs.

Large Language Models (LLMs) have shown remarkable capabilities across various tasks, but their deployment in high-stake domains requires consistent and coherent behavior across multiple rounds of user interaction. This paper introduces a comprehensive framework for evaluating and improving LLM response consistency, making three key contributions . First, we introduce Position-Weighted Consistency (PWC), a metric designed to capture both the importance of early-stage stability and recovery patterns in multi-turn interactions. Second, we present MT-Consistency, a carefully curated benchmark dataset spanning diverse domains and difficulty levels, specifically designed to evaluate LLM consistency under various challenging follow-up scenarios. Third, we introduce Confidence-Aware Response Generation (CARG), a framework that significantly improves response stability by explicitly integrating internal model confidence scores during the generation process. Experimental results demonstrate that CARG significantly improves response stability without sacrificing accuracy, offering a practical path toward more dependable LLM behavior in critical, real-world deployments.