Motivated by the surge of large language models, there has been a push to formally characterize the symbolic abilities intrinsic to the transformer architecture. A programming language, called RASP, has been proposed, which can be directly compiled into transformer weights to implement these algorithms. However, the tasks that can be implemented in RASP are often uncommon to learn from natural unsupervised data, showing a mismatch between theoretical capabilities of the transformer architecture, and the practical learnability of these capabilities from unsupervised data. We propose tracr-injection, a method that allows us to distill algorithms written in RASP directly into a pre-trained language model. We showcase our method by injecting 3 different algorithms into a language model. We show how our method creates an interpretable subspace within the model's residual stream, which can be decoded into the variables present in the code of the RASP algorithm. Additionally, we found that the proposed method can improve out-of-distribution performance compared to our baseline, indicating that indeed a more symbolic mechanism is taking place in the inner workings of the model. We release the code used to run our experiments.

Efficiently modeling and exploiting opponents is a long-standing challenge in adversarial domains. Large Language Models (LLMs) trained on extensive textual data have recently demonstrated outstanding performance in general tasks, introducing new research directions for opponent modeling. Some studies primarily focus on directly using LLMs to generate decisions based on the elaborate prompt context that incorporates opponent descriptions, while these approaches are limited to scenarios where LLMs possess adequate domain expertise. To address that, we introduce a two-stage Strategy-Augmented Planning (SAP) framework that significantly enhances the opponent exploitation capabilities of LLM-based agents by utilizing a critical component, the Strategy Evaluation Network (SEN). Specifically, in the offline stage, we construct an explicit strategy space and subsequently collect strategy-outcome pair data for training the SEN network. During the online phase, SAP dynamically recognizes the opponent's strategies and greedily exploits them by searching best response strategy on the well-trained SEN, finally translating strategy to a course of actions by carefully designed prompts. Experimental results show that SAP exhibits robust generalization capabilities, allowing it to perform effectively not only against previously encountered opponent strategies but also against novel, unseen strategies. In the MicroRTS environment, SAP achieves a $85.35\%$ performance improvement over baseline methods and matches the competitiveness of reinforcement learning approaches against state-of-the-art (SOTA) rule-based AI. Our code is available at https://github.com/hsushuai/SAP.

Recent improvement in large language model performance have, in all likelihood, been accompanied by improvement in how well they can approximate the distribution of their training data. In this work, we explore the following question: which properties of text domains do LLMs faithfully approximate, and how well do they do so? Applying observational approaches familiar from corpus linguistics, we prompt a commonly used, opensource LLM to regenerate text from two domains of permissively licensed English text which are often contained in LLM training data -- Wikipedia and news text. This regeneration paradigm allows us to investigate whether LLMs can faithfully match the original human text domains in a fairly semantically-controlled setting. We investigate varying levels of syntactic abstraction, from more simple properties like sentence length, and article readability, to more complex and higher order properties such as dependency tag distribution, parse depth, and parse complexity. We find that the majority of the regenerated distributions show a shifted mean, a lower standard deviation, and a reduction of the long tail, as compared to the human originals.

Systematic reviews are comprehensive literature reviews that address highly focused research questions and represent the highest form of evidence in medicine. A critical step in this process is the development of complex Boolean queries to retrieve relevant literature. Given the difficulty of manually constructing these queries, recent efforts have explored Large Language Models (LLMs) to assist in their formulation. One of the first studies,Wang et al., investigated ChatGPT for this task, followed by Staudinger et al., which evaluated multiple LLMs in a reproducibility study. However, the latter overlooked several key aspects of the original work, including (i) validation of generated queries, (ii) output formatting constraints, and (iii) selection of examples for chain-of-thought (Guided) prompting. As a result, its findings diverged significantly from the original study. In this work, we systematically reproduce both studies while addressing these overlooked factors. Our results show that query effectiveness varies significantly across models and prompt designs, with guided query formulation benefiting from well-chosen seed studies. Overall, prompt design and model selection are key drivers of successful query formulation. Our findings provide a clearer understanding of LLMs' potential in Boolean query generation and highlight the importance of model- and prompt-specific optimisations. The complex nature of systematic reviews adds to challenges in both developing and reproducing methods but also highlights the importance of reproducibility studies in this domain.

The proliferation of open-sourced Large Language Models (LLMs) and diverse downstream tasks necessitates efficient model selection, given the impracticality of fine-tuning all candidates due to computational constraints. Despite the recent advances in LLM selection, a fundamental research question largely remains nascent: how can we model the dynamic behaviors of LLMs during fine-tuning, thereby enhancing our understanding of their generalization performance across diverse downstream tasks? In this work, we propose a novel theoretical framework that provides a proper lens to assess the generalization capabilities of LLMs, thereby enabling accurate and efficient LLM selection for downstream applications. In particular, we first derive a PAC-Bayesian Generalization Bound that unveils fine-tuning dynamics of LLMs and then introduce LENSLLM, a Neural Tangent Kernel (NTK)-based Rectified Scaling Model that enables accurate performance predictions across diverse tasks while maintaining computational efficiency. Extensive empirical results on 3 large-scale benchmarks demonstrate that our model achieves up to 91.1% accuracy and reduces up to 88.5% computational cost in LLM selection, outperforming 5 state-of-the-art methods. We open-source our proposed LENSLLM model and corresponding results at LensLLM.io.

Medical audio signals, such as heart and lung sounds, play a crucial role in clinical diagnosis. However, analyzing these signals remains challenging: traditional methods rely on handcrafted features or supervised deep learning models that demand extensive labeled datasets, limiting their scalability and applicability. To address these issues, we propose CaReAQA, an audio-language model that integrates a foundation audio model with the reasoning capabilities of large language models, enabling clinically relevant, open-ended diagnostic responses. Alongside CaReAQA, we introduce CaReSound, a benchmark dataset of annotated medical audio recordings enriched with metadata and paired question-answer examples, intended to drive progress in diagnostic reasoning research. Evaluation results show that CaReAQA achieves 86.2% accuracy on open-ended diagnostic reasoning tasks, outperforming baseline models. It also generalizes well to closed-ended classification tasks, achieving an average accuracy of 56.9% on unseen datasets. Our findings show how audio-language integration and reasoning advances medical diagnostics, enabling efficient AI systems for clinical decision support.

Large Vision Language Models (LVLMs) excel in various vision-language tasks. Yet, their robustness to visual variations in position, scale, orientation, and context that objects in natural scenes inevitably exhibit due to changes in viewpoint and environment remains largely underexplored. To bridge this gap, we introduce V$^2$R-Bench, a comprehensive benchmark framework for evaluating Visual Variation Robustness of LVLMs, which encompasses automated evaluation dataset generation and principled metrics for thorough robustness assessment. Through extensive evaluation on 21 LVLMs, we reveal a surprising vulnerability to visual variations, in which even advanced models that excel at complex vision-language tasks significantly underperform on simple tasks such as object recognition. Interestingly, these models exhibit a distinct visual position bias that contradicts theories of effective receptive fields, and demonstrate a human-like visual acuity threshold. To identify the source of these vulnerabilities, we present a systematic framework for component-level analysis, featuring a novel visualization approach for aligned visual features. Results show that these vulnerabilities stem from error accumulation in the pipeline architecture and inadequate multimodal alignment. Complementary experiments with synthetic data further demonstrate that these limitations are fundamentally architectural deficiencies, scoring the need for architectural innovations in future LVLM designs.

Malicious URLs persistently threaten the cybersecurity ecosystem, by either deceiving users into divulging private data or distributing harmful payloads to infiltrate host systems. Gaining timely insights into the current state of this ongoing battle holds significant importance. However, existing reviews exhibit 4 critical gaps: 1) Their reliance on algorithm-centric taxonomies obscures understanding of how detection approaches exploit specific modal information channels; 2) They fail to incorporate pivotal LLM/Transformer-based defenses; 3) No open-source implementations are collected to facilitate benchmarking; 4) Insufficient dataset coverage.This paper presents a comprehensive review of malicious URL detection technologies, systematically analyzing methods from traditional blacklisting to advanced deep learning approaches (e.g. Transformer, GNNs, and LLMs). Unlike prior surveys, we propose a novel modality-based taxonomy that categorizes existing works according to their primary data modalities (URL, HTML, Visual, etc.). This hierarchical classification enables both rigorous technical analysis and clear understanding of multimodal information utilization. Furthermore, to establish a profile of accessible datasets and address the lack of standardized benchmarking (where current studies often lack proper baseline comparisons), we curate and analyze: 1) publicly available datasets (2016-2024), and 2) open-source implementations from published works(2013-2025). Then, we outline essential design principles and architectural frameworks for product-level implementations. The review concludes by examining emerging challenges and proposing actionable directions for future research. We maintain a GitHub repository for ongoing curating datasets and open-source implementations: https://github.com/sevenolu7/Malicious-URL-Detection-Open-Source/tree/master.

Static analysis plays a crucial role in software vulnerability detection, yet faces a persistent precision-scalability tradeoff. In large codebases like the Linux kernel, traditional static analysis tools often generate excessive false positives due to simplified vulnerability modeling and overapproximation of path and data constraints. While large language models (LLMs) demonstrate promising code understanding capabilities, their direct application to program analysis remains unreliable due to inherent reasoning limitations. We introduce BugLens, a post-refinement framework that significantly enhances static analysis precision for bug detection. BugLens guides LLMs through structured reasoning steps to assess security impact and validate constraints from the source code. When evaluated on Linux kernel taint-style bugs detected by static analysis tools, BugLens improves precision approximately 7-fold (from 0.10 to 0.72), substantially reducing false positives while uncovering four previously unreported vulnerabilities. Our results demonstrate that a well-structured, fully automated LLM-based workflow can effectively complement and enhance traditional static analysis techniques.

The limited reasoning capabilities of small language models (SLMs) cast doubt on their suitability for tasks demanding deep, multi-step logical deduction. This paper introduces a framework called Small Reasons, Large Hints (SMART), which selectively augments SLM reasoning with targeted guidance from large language models (LLMs). Inspired by the concept of cognitive scaffolding, SMART employs a score-based evaluation to identify uncertain reasoning steps and injects corrective LLM-generated reasoning only when necessary. By framing structured reasoning as an optimal policy search, our approach steers the reasoning trajectory toward correct solutions without exhaustive sampling. Our experiments on mathematical reasoning datasets demonstrate that targeted external scaffolding significantly improves performance, paving the way for collaborative use of both SLM and LLM to tackle complex reasoning tasks that are currently unsolvable by SLMs alone.