Large Language Models (LLMs) can perform Automatic Modulation Classification (AMC) in an open-set manner without LLM fine-tuning when equipped with carefully designed in-context prompts~\cite{rostami2025plug}. Building on this prior work, we target the practical bottlenecks of long prompt contexts and large model sizes that impede in-the-loop deployment. We present Discretized Statistics in-Context Automatic Modulation Classification (DiSC-AMC), a token- and parameter-efficient variant that: (i) discretizes higher-order statistics and cumulants into compact symbolic tokens, (ii) prunes the exemplar list via a lightweight k-top neural prefilter and filters misleading/low-impact features using rationales extracted from prior LLM responses, and (iii) enforces label-only predictions through a calibrated prompt template. Together, these changes reduce both input/output tokens and the model parameter footprint by more than half while maintaining competitive accuracy. On synthetic AMC with ten modulation types under noise, a 7B \textit{DeepSeek-R1-Distill-Qwen} baseline achieves 5.2% accuracy, whereas our system, using an approximately 5B-parameter \textit{Gemini-2.5-Flash}~\cite{comanici2025gemini} model, attains 45.5% accuracy. These results demonstrate that careful discretization and context selection can cut inference cost by over 2x while preserving the advantages of prompt-based AMC and enabling practical in-the-loop use.

Security Operations Centers (SOCs) are overwhelmed by tens of thousands of daily alerts, with only a small fraction corresponding to genuine attacks. This overload creates alert fatigue, leading to overlooked threats and analyst burnout. Classical detection pipelines are brittle and context-poor, while recent LLM-based approaches typically rely on a single model to interpret logs, retrieve context, and adjudicate alerts end-to-end -- an approach that struggles with noisy enterprise data and offers limited transparency. We propose CORTEX, a multi-agent LLM architecture for high-stakes alert triage in which specialized agents collaborate over real evidence: a behavior-analysis agent inspects activity sequences, evidence-gathering agents query external systems, and a reasoning agent synthesizes findings into an auditable decision. To support training and evaluation, we release a dataset of fine-grained SOC investigations from production environments, capturing step-by-step analyst actions and linked tool outputs. Across diverse enterprise scenarios, CORTEX substantially reduces false positives and improves investigation quality over state-of-the-art single-agent LLMs.

Agents backed by large language models (LLMs) often rely on external tools drawn from marketplaces where multiple providers offer functionally equivalent options. This raises a critical point concerning fairness: if selection is systematically biased, it can degrade user experience and distort competition by privileging some providers over others. We introduce a benchmark of diverse tool categories, each containing multiple functionally equivalent tools, to evaluate tool-selection bias. Using this benchmark, we test seven models and show that unfairness exists with models either fixating on a single provider or disproportionately preferring earlier-listed tools in context. To investigate the origins of this bias, we conduct controlled experiments examining tool features, metadata (name, description, parameters), and pre-training exposure. We find that: (1) semantic alignment between queries and metadata is the strongest predictor of choice; (2) perturbing descriptions significantly shifts selections; and (3) repeated pre-training exposure to a single endpoint amplifies bias. Finally, we propose a lightweight mitigation that first filters the candidate tools to a relevant subset and then samples uniformly, reducing bias while preserving good task coverage. Our findings highlight tool-selection bias as a key obstacle for the fair deployment of tool-augmented LLMs. Large language models (LLMs) have transformed natural language processing, achieving near-human performance on tasks ranging from code generation to creative writing (Naveed et al., 2024; Luo et al., 2024). Y et LLMs cannot directly act in the world: they cannot query databases, fetch live information, or invoke external services. Additionally, their knowledge remains frozen at training time, leaving them prone to "hallucinations" when asked about events beyond their cutoff (Ji et al., 2023). Augmenting LLMs with external "tools" / APIs addresses these shortcomings by allowing models to delegate specialized functions to dedicated services (Qu et al., 2025). It endows LLMs with the ability to act, a core capability often associated with LLM agents (Chowa et al., 2025). A crucial step within the typical tool-usage pipeline is the multi-stage tool-selection process: given an instruction to the LLM, (i) retrieve a short list of the most relevant candidate tools based on the user query (with, e.g., highest semantic similarity) from a potentially large database of tools, (ii) insert their metadata into the prompt, (iii) have the LLM reason and pick one to solve (one of) the necessary user task(s). However, this process introduces a new challenge: bias (see Figure 1).

Detecting hallucinations in Large Language Model-generated text is crucial for their safe deployment. While probing classifiers show promise, they operate on isolated layer-token pairs and are LLM-specific, limiting their effectiveness and hindering cross-LLM applications. In this paper, we introduce a novel approach to address these shortcomings. We build on the natural sequential structure of activation data in both axes (layers $\times$ tokens) and advocate treating full activation tensors akin to images. We design ACT-ViT, a Vision Transformer-inspired model that can be effectively and efficiently applied to activation tensors and supports training on data from multiple LLMs simultaneously. Through comprehensive experiments encompassing diverse LLMs and datasets, we demonstrate that ACT-ViT consistently outperforms traditional probing techniques while remaining extremely efficient for deployment. In particular, we show that our architecture benefits substantially from multi-LLM training, achieves strong zero-shot performance on unseen datasets, and can be transferred effectively to new LLMs through fine-tuning. Full code is available at https://github.com/BarSGuy/ACT-ViT.

Data-driven approaches have revolutionized the design and optimization of photonic metadevices by harnessing advanced artificial intelligence methodologies. This review takes a model-centric perspective that synthesizes emerging design strategies and delineates how traditional trial-and-error and computationally intensive electromagnetic simulations are being supplanted by deep learning frameworks that efficiently navigate expansive design spaces. We discuss artificial intelligence implementation in several metamaterial design aspects from high-degree-of-freedom design to large language model-assisted design. By addressing challenges such as transformer model implementation, fabrication limitations, and intricate mutual coupling effects, these AI-enabled strategies not only streamline the forward modeling process but also offer robust pathways for the realization of multifunctional and fabrication-friendly nanophotonic devices. This review further highlights emerging opportunities and persistent challenges, setting the stage for next-generation strategies in nanophotonic engineering.

Automatically generated radiology reports often receive high scores from existing evaluation metrics but fail to earn clinicians' trust. This gap reveals fundamental flaws in how current metrics assess the quality of generated reports. We rethink the design and evaluation of these metrics and propose a clinically grounded Meta-Evaluation framework. We define clinically grounded criteria spanning clinical alignment and key metric capabilities, including discrimination, robustness, and monotonicity. Using a fine-grained dataset of ground truth and rewritten report pairs annotated with error types, clinical significance labels, and explanations, we systematically evaluate existing metrics and reveal their limitations in interpreting clinical semantics, such as failing to distinguish clinically significant errors, over-penalizing harmless variations, and lacking consistency across error severity levels. Our framework offers guidance for building more clinically reliable evaluation methods.

Black box large language models (LLMs) make information extraction (IE) easy to configure, but hard to trust. Unlike traditional information extraction pipelines, the information "extracted" is not guaranteed to be grounded in the document. To prevent this, this paper introduces the notion of a "safe passage": context generated by the LLM that is both grounded in the document and consistent with the extracted information. This is operationalized via a three-step pipeline, SafePassage, which consists of: (1) an LLM extractor that generates structured entities and their contexts from a document, (2) a string-based global aligner, and (3) a scoring model. Results show that using these three parts in conjunction reduces hallucinations by up to 85% on information extraction tasks with minimal risk of flagging non-hallucinations. High agreement between the SafePassage pipeline and human judgments of extraction quality mean that the pipeline can be dually used to evaluate LLMs. Surprisingly, results also show that using a transformer encoder fine-tuned on a small number of task-specific examples can outperform an LLM scoring model at flagging unsafe passages. These annotations can be collected in as little as 1-2 hours.

Large Language Models (LLMs) have shown extraordinary success across various text generation tasks; however, their potential for simple yet essential text classification remains underexplored, as LLM pre-training tends to emphasize generation over classification. While LLMs with instruction tuning can transform classification into a generation task, they often struggle to categorize nuanced texts. One such example is text revision, which involves nuanced edits between pairs of texts. Although simply fine-tuning LLMs for revision classification seems plausible, it requires a large amount of revision annotations, which are exceptionally expensive and scarce in the community. To address this issue, we introduce a plug-and-play layer-wise parameter-efficient fine-tuning (PEFT) framework, i.e., IR-Tuning, which fine-tunes a subset of important LLM layers that are dynamically selected based on their gradient norm distribution, while freezing those of redundant layers. Extensive experiments suggest that IR-Tuning surpasses several layer-wise PEFT baselines over diverse text revisions, while achieving fast convergence, low GPU memory consumption, and effectiveness on small revision corpora.

The alignment of large language models (LLMs) with human values increasingly relies on using other LLMs as automated judges, or ``autoraters''. However, their reliability is limited by a foundational issue: they are trained on discrete preference labels, forcing a single ground truth onto tasks that are often subjective, ambiguous, or nuanced. We argue that a reliable autorater must learn to model the full distribution of preferences defined by a target population. In this paper, we propose a general framework for calibrating probabilistic autoraters to any given preference distribution. We formalize the problem and present two learning methods tailored to different data conditions: 1) a direct supervised fine-tuning for dense, probabilistic labels, and 2) a reinforcement learning approach for sparse, binary labels. Our empirical results show that finetuning autoraters with a distribution-matching objective leads to verbalized probability predictions that are better aligned with the target preference distribution, with improved calibration and significantly lower positional bias, all while preserving performance on objective tasks.

Interest in generative Electrocardiogram-Language Models (ELMs) is growing, as they can produce textual responses conditioned on ECG signals and textual queries. Unlike traditional classifiers that output label probabilities, ELMs are more versatile, supporting domain-specific tasks (e.g., waveform analysis, diagnosis, prognosis) as well as general tasks (e.g., open-ended questions, dialogue). Retrieval-Augmented Generation (RAG), widely used in Large Language Models (LLMs) to ground LLM outputs in retrieved knowledge, helps reduce hallucinations and improve natural language generation (NLG). However, despite its promise, no open-source implementation or systematic study of RAG pipeline design for ELMs currently exists. To address this gap, we present the first open-source RAG pipeline for ELMs, along with baselines and ablation studies for NLG. Experiments on three public datasets show that ELMs with RAG consistently improves performance over non-RAG baselines and highlights key ELM design considerations. Our code is available at: https://github.com/willxxy/ECG-Bench.