Large language models (LLMs) demonstrate remarkable capabilities in various complex language tasks, yet they face significant reliability challenges, including factual inaccuracies and generated biases. Uncertainty quantification (UQ) plays a pivotal role in assessing model trustworthiness, particularly for high-stakes applications. However, current UQ methods for LLMs encounter computational efficiency bottlenecks due to their reliance on extensive sampling or external model invocations. In this work, we introduce a novel, sampling-free uncertainty quantification framework centered on hidden layer representation analysis. Our method facilitates real-time uncertainty quantification by modeling hierarchical internal semantic dynamics during the generation process. Through comprehensive experiments on multiple QA datasets and diverse model scales, we show that our approach consistently outperforms existing uncertainty quantification techniques in distinguishing correct from incorrect generations. Our results reveal that analyzing the dynamic evolution of hidden states provides a potent and computationally efficient signal for uncertainty quantification, directly from the model's internal workings, surpassing methods that depend solely on output probabilities or approximations via multiple samples.

Accurate Uncertainty Quantification (UQ) is critical for reliable deployment of Large Language Models (LLMs), yet traditional probability-based metrics often fail to capture the model's true epistemic state.While recent mechanistic approaches leverage hidden state dynamics, they typically aggregate residual stream updates, conflating the distinct roles of parametric memory (Feed-Forward Networks) and contextual processing (Attention).We argue that this aggregation obscures fine-grained mechanistic conflicts, such as memory-context misalignment, that are fundamental indicators of uncertainty.To address this, we introduce Decoupled Update Dynamics (DUD), a framework that explicitly decouples FFN and Attention contributions via noise-induced causal interventions.By quantifying the independent restoration capabilities of each module, we construct a dual-stream dynamic profile that captures the model's internal fragility.Extensive experiments demonstrate that DUD significantly outperforms state-of-the-art baselines in both uncertainty estimation and calibration, while exhibiting superior cross-dataset generalization, validating decoupled dynamics as a robust proxy for model faithfulness.

Query-service relevance prediction in e-commerce search systems faces strict latency requirements that prevent the direct application of Large Language Models (LLMs). To bridge this gap, we propose a two-stage reasoning distillation framework to transfer reasoning capabilities from a powerful teacher LLM to a lightweight, deployment-friendly student model. In the first stage, we address the limitations of general-purpose LLMs by constructing a domain-adapted teacher model. This is achieved through a three-step process: domain-adaptive pre-training to inject platform knowledge, supervised fine-tuning to elicit reasoning skills, and preference optimization with a multi-dimensional reward model to ensure the generation of reliable and preference-aligned reasoning paths. This teacher can then automatically annotate massive query-service pairs from search logs with both relevance labels and reasoning chains. In the second stage, to address the challenges of architectural heterogeneity in standard distillation, we introduce Contrastive Reasoning Self-Distillation (CRSD). By modeling the behavior of the same student model under ``standard'' and ``reasoning-augmented'' inputs as a teacher-student relationship, CRSD enables the lightweight model to internalize the teacher's complex decision-making mechanisms without needing the explicit reasoning path at inference. Offline evaluations and online A/B testing in the Meituan search advertising system demonstrate that our framework achieves significant improvements across multiple metrics, validating its effectiveness and practical value.

The neural basis of semantics has long been a central topic across disciplines such as linguistics, cognitive neuroscience, and artificial intelligence. Traditional approaches in neuroscience often struggle to isolate semantic processing from syntactic influences without compromising the naturalness of sentence comprehension, due to constraints in experimental paradigms. Meanwhile, large language models (LLMs) have shown remarkable alignment with human brain activity, offering new tools to investigate language representation in the brain through brain encoding techniques. However, extracting purely semantic representations from LLMs remains a significant challenge, as their internal embeddings inherently entangle syntax and semantics. To address this challenge, we introduce a semantic representation isolation method based on large language models. Specifically, we propose Augmented Semantic Extraction (ASE), which conditionally perturbs word order to disrupt syntactic structures, thereby isolating semantic representations from the mixed encodings produced by LLMs. We evaluate ASE using multiple probing techniques and demonstrate that it effectively purifies semantic content from language model representations under a linear decoding framework. Leveraging ASE, we perform brain encoding experiments on a functional magnetic resonance imaging (fMRI) dataset collected during natural spoken narrative comprehension. Our findings reveal the cortical organization of semantic processing. The results provide neural evidence that semantic processing constitutes the dominant component of language comprehension in the brain.

Recent advances in Vision Transformers (ViTs) have shown remarkable performance across vision tasks, yet their deep, uniform layer structure introduces significant computational overhead. In this work, we explore the emergent dynamics of ViT layers through the lens of energy-based memory systems, drawing a connection between self-attention and modern Hopfield networks. We introduce a novel metric—Layer Instability Index (LII)—derived from the operational softmax mode and its variability, to quantify the metastability of each Transformer layer over time. Our analysis reveals that certain layers exhibit consistent convergence to attractor-like states, suggesting functional specialisation and early stabilisation. Leveraging this insight, we propose an adaptive training framework that dynamically freezes or skips stable layers based on their energy landscape behavior. Our method reduces training costs while maintaining or improving accuracy. Extensive experiments on ViT-S/B/L on CUB-200-2011, CIFAR-10/100, Food-101, Stanford Dogs, and Beans demonstrate the generality and efficiency of our approach. This work provides new theoretical and practical perspectives for energy-aware optimisation of deep Transformer models.