Fluid voice-to-voice interaction requires reliable and low-latency detection of when a user has finished speaking. Traditional audio-silence end-pointers add hundreds of milliseconds of delay and fail under hesitations or language-specific phenomena. We present, to our knowledge, the first systematic study of Thai text-only end-of-turn (EOT) detection for real-time agents. We compare zero-shot and few-shot prompting of compact LLMs to supervised fine-tuning of lightweight transformers. Using transcribed subtitles from the YODAS corpus and Thai-specific linguistic cues (e.g., sentence-final particles), we formulate EOT as a binary decision over token boundaries. We report a clear accuracy-latency tradeoff and provide a public-ready implementation plan. This work establishes a Thai baseline and demonstrates that small, fine-tuned models can deliver near-instant EOT decisions suitable for on-device agents.

Although large language models (LLMs) have tremendous utility, trustworthiness is still a chief concern: models often generate incorrect information with high confidence. While contextual information can help guide generation, identifying when a query would benefit from retrieved context and assessing the effectiveness of that context remains challenging. In this work, we operationalize interpretability methods to ascertain whether we can predict the correctness of model outputs from the model's activations alone. We also explore whether model internals contain signals about the efficacy of external context. We consider correct, incorrect, and irrelevant context and introduce metrics to distinguish amongst them. Experiments on six different models reveal that a simple classifier trained on intermediate layer activations of the first output token can predict output correctness with about 75% accuracy, enabling early auditing. Our model-internals-based metric significantly outperforms prompting baselines at distinguishing between correct and incorrect context, guarding against inaccuracies introduced by polluted context. These findings offer a lens to better understand the underlying decision-making processes of LLMs. Our code is publicly available at https://github.com/jiarui-liu/LLM-Microscope

The meteoric rise of foundation models (FMs) has expanded their capabilities far beyond conventional tasks. Creativity, long regarded as a hallmark of human intelligence and a driver of innovation, is now increasingly recognized as a critical dimension of machine intelligence in the era of generative FMs, complementing traditional measures of accuracy. However, existing evaluation frameworks for creativity remain fragmented, relying on ad hoc metrics not firmly grounded in established theories. To address this gap, we introduce C^2-Eval, a holistic benchmark for unified assessment of creativity in FMs. C^2-Eval distinguishes between two complementary forms of creativity: convergent creativity, where tasks admit constrained solutions (e.g., code generation), and divergent creativity, where tasks are open-ended (e.g., storytelling). It evaluates both dimensions using fine-grained criteria derived from social-science theory, focusing on Usefulness, Originality, and Surprise (U-O-S). Through extensive experiments on leading proprietary and open-source models, we analyze trade-offs in their creative capabilities. Our results highlight both the strengths and challenges of current FMs in pursuing a creative machine mind, showing that C^2-Eval is an effective lens for examining the evolving landscape of creative AI.

Deep learning has become a cornerstone of modern machine learning. It relies heavily on vast datasets and significant computational resources for high performance. This data often contains sensitive information, making privacy a major concern in deep learning. Spiking Neural Networks (SNNs) have emerged as an energy-efficient alternative to conventional deep learning approaches. Nevertheless, SNNs still depend on large volumes of data, inheriting all the privacy challenges of deep learning. Homomorphic encryption addresses this challenge by allowing computations to be performed on encrypted data, ensuring data confidentiality throughout the entire processing pipeline. In this paper, we introduce PRIVSPIKE, a privacy-preserving inference framework for SNNs using the CKKS homomorphic encryption scheme. PRIVSPIKE supports arbitrary depth SNNs and introduces two key algorithms for evaluating the Leaky Integrate-and-Fire activation function: (1) a polynomial approximation algorithm designed for high-performance SNN inference, and (2) a novel scheme-switching algorithm that optimizes precision at a higher computational cost. We evaluate PRIVSPIKE on MNIST, CIFAR-10, Neuromorphic MNIST, and CIFAR-10 DVS using models from LeNet-5 and ResNet-19 architectures, achieving encrypted inference accuracies of 98.10%, 79.3%, 98.1%, and 66.0%, respectively. On a consumer-grade CPU, SNN LeNet-5 models achieved inference times of 28 seconds on MNIST and 212 seconds on Neuromorphic MNIST. For SNN ResNet-19 models, inference took 784 seconds on CIFAR-10 and 1846 seconds on CIFAR-10 DVS. These results establish PRIVSPIKE as a viable and efficient solution for secure SNN inference, bridging the gap between energy-efficient deep neural networks and strong cryptographic privacy guarantees while outperforming prior encrypted SNN solutions.

Large language models (LLMs) are increasingly deployed in agentic systems where they map user intents to relevant external tools to fulfill a task. A critical step in this process is tool selection, where a retriever first surfaces candidate tools from a larger pool, after which the LLM selects the most appropriate one. This pipeline presents an underexplored attack surface where errors in selection can lead to severe outcomes like unauthorized data access or denial of service, all without modifying the agent's model or code. While existing evaluations measure task performance in benign settings, they overlook the specific vulnerabilities of the tool selection mechanism under adversarial conditions. To address this gap, we introduce ToolCert, the first statistical framework that formally certifies tool selection robustness. ToolCert models tool selection as a Bernoulli success process and evaluates it against a strong, adaptive attacker who introduces adversarial tools with misleading metadata, and are iteratively refined based on the agent's previous choices. By sampling these adversarial interactions, ToolCert produces a high-confidence lower bound on accuracy, formally quantifying the agent's worst-case performance. Our evaluation with ToolCert uncovers the severe fragility: under attacks injecting deceptive tools or saturating retrieval, the certified accuracy bound drops near zero, an average performance drop of over 60% compared to non-adversarial settings. For attacks targeting the retrieval and selection stages, the certified accuracy bound plummets to less than 20% after just a single round of adversarial adaptation. ToolCert thus reveals previously unexamined security threats inherent to tool selection and provides a principled method to quantify an agent's robustness to such threats, a necessary step for the safe deployment of agentic systems.

The Transformer architecture has revolutionized the field of sequence modeling and underpins the recent breakthroughs in large language models (LLMs). However, a comprehensive mathematical theory that explains its structure and operations remains elusive. In this work, we propose a novel continuous framework that rigorously interprets the Transformer as a discretization of a structured integro-differential equation. Within this formulation, the self-attention mechanism emerges naturally as a non-local integral operator, and layer normalization is characterized as a projection to a time-dependent constraint. This operator-theoretic and variational perspective offers a unified and interpretable foundation for understanding the architecture's core components, including attention, feedforward layers, and normalization. Our approach extends beyond previous theoretical analyses by embedding the entire Transformer operation in continuous domains for both token indices and feature dimensions. This leads to a principled and flexible framework that not only deepens theoretical insight but also offers new directions for architecture design, analysis, and control-based interpretations. This new interpretation provides a step toward bridging the gap between deep learning architectures and continuous mathematical modeling, and contributes a foundational perspective to the ongoing development of interpretable and theoretically grounded neural network models.

Distilling long reasoning traces (10K+ tokens) from stronger teacher models into smaller student LLMs via SFT has emerged as a standard paradigm. This approach is practical and efficient: it leverages the ease of generating abundant reasoning data from stronger models and provides a direct, data-driven way to teach less capable models better reasoning. While previous work has largely focused on prompt selection with responses from a single teacher, the equally important problem of choosing the best response when multiple teacher outputs are available for a single prompt remains underexplored. This challenge becomes important in a multi-teacher setting, where different students may benefit from the outputs of different teachers. This paper fills that gap with a systematic study of response selection for reasoning distillation. We first show that the current method, which picks responses the student assigns the highest global log-probability (global naturalness), fails when responses come from multiple teachers, i.e., global naturalness no longer correlates with downstream performance, especially as the reasoning traces from strong teachers become longer. To overcome this problem, we introduce Local Naturalness, which measures the student's log-probabilities over short, sequential reasoning steps conditioned only on a small local window. Local Naturalness enables two applications: 1) Teacher Selection: Aggregating local scores across prompts reliably identifies the most helpful teacher. 2) Response Selection from a Multiple Teachers: When mixing answers from many teachers, Local Naturalness boosts a 32B student's accuracy on math benchmarks by 9.4pp over global selection, also surpassing the performance achieved by training on data from the single best teacher. These results highlight the power of localized data quality evaluation and data mixing for more effective reasoning distillation.

Personalized AI agents are becoming central to modern information retrieval, yet most evaluation methodologies remain static, relying on fixed benchmarks and one-off metrics that fail to reflect how users' needs evolve over time. These limitations hinder our ability to assess whether agents can meaningfully adapt to individuals across dynamic, longitudinal interactions. In this perspective paper, we propose a conceptual lens for rethinking evaluation in adaptive personalization, shifting the focus from static performance snapshots to interaction-aware, evolving assessments. We organize this lens around three core components: (1) persona-based user simulation with temporally evolving preference models; (2) structured elicitation protocols inspired by reference interviews to extract preferences in context; and (3) adaptation-aware evaluation mechanisms that measure how agent behavior improves across sessions and tasks. While recent works have embraced LLM-driven user simulation, we situate this practice within a broader paradigm for evaluating agents over time. To illustrate our ideas, we conduct a case study in e-commerce search using the PersonalWAB dataset. Beyond presenting a framework, our work lays a conceptual foundation for understanding and evaluating personalization as a continuous, user-centric endeavor.

Large Language Models (LLMs) can produce catastrophic responses in conversational settings that pose serious risks to public safety and security. Existing evaluations often fail to fully reveal these vulnerabilities because they rely on fixed attack prompt sequences, lack statistical guarantees, and do not scale to the vast space of multi-turn conversations. In this work, we propose QRLLM, a novel, principled Certification framework for Catastrophic risks in multi-turn Conversation for LLMs that bounds the probability of an LLM generating catastrophic responses under multi-turn conversation distributions with statistical guarantees. We model multi-turn conversations as probability distributions over query sequences, represented by a Markov process on a query graph whose edges encode semantic similarity to capture realistic conversational flow, and quantify catastrophic risks using confidence intervals. We define several inexpensive and practical distributions: random node, graph path, adaptive with rejection. Our results demonstrate that these distributions can reveal substantial catastrophic risks in frontier models, with certified lower bounds as high as 70\% for the worst model, highlighting the urgent need for improved safety training strategies in frontier LLMs.

Time series anomaly detection plays a crucial role in a wide range of fields, such as healthcare and internet traffic monitoring. The emergence of large language models (LLMs) offers new opportunities for detecting anomalies in the ubiquitous time series data. Traditional approaches struggle with variable-length time series sequences and context-based anomalies. We propose Soft Prompt Enhanced Anomaly Recognition (SPEAR), a novel approach to leverage LLMs for anomaly detection with soft prompts and quantization. Our methodology involves quantizing and transforming the time series data into input embeddings and combining them with learnable soft prompt embeddings. These combined embeddings are then fed into a frozen LLM. The soft prompts are updated iteratively based on a cross-entropy loss, allowing the model to adapt to time series anomaly detection. The use of soft prompts helps adapt LLMs effectively to time series tasks, while quantization ensures optimal handling of sequences, as LLMs are designed to handle discrete sequences. Our experimental results demonstrate that soft prompts effectively increase LLMs' performance in downstream tasks regarding time series anomaly detection.