AI-generated text (AIGT) detection evasion aims to reduce the detection probability of AIGT, helping to identify weaknesses in detectors and enhance their effectiveness and reliability in practical applications. Although existing evasion methods perform well, they suffer from high computational costs and text quality degradation. To address these challenges, we propose Self-Disguise Attack (SDA), a novel approach that enables Large Language Models (LLM) to actively disguise its output, reducing the likelihood of detection by classifiers. The SDA comprises two main components: the adversarial feature extractor and the retrieval-based context examples optimizer. The former generates disguise features that enable LLMs to understand how to produce more human-like text. The latter retrieves the most relevant examples from an external knowledge base as in-context examples, further enhancing the self-disguise ability of LLMs and mitigating the impact of the disguise process on the diversity of the generated text. The SDA directly employs prompts containing disguise features and optimized context examples to guide the LLM in generating detection-resistant text, thereby reducing resource consumption. Experimental results demonstrate that the SDA effectively reduces the average detection accuracy of various AIGT detectors across texts generated by three different LLMs, while maintaining the quality of AIGT.

Recent advances in artificial intelligence (AI), particularly in the domain of large language models (LLMs), have significantly amplified concerns around AI safety and security. One critical aspect of these concerns is the vulnerability of LLMs to backdoor attacks--a malicious strategy whereby an attacker injects specific triggers into training data, resulting in "sleeper agents" that behave normally until activated by particular inputs [6]. These backdoored models (also known as sleeper agents or trojaned models) pose a serious threat as they cannot be detected by standard evaluation methods and manifest undesirable or harmful behaviors only upon exposure to particular triggers in the input [2]. Triggers can take on many forms, ranging from simple single-token lexical triggers to complex semantic triggers [9]. The significance of studying backdoor vulnerabilities arises from two primary threat models: Data-poisoned sleeper agents These involve deliberate poisoning of the training data to trigger specific harmful behaviors under attacker-defined conditions [3]. Real-world implications are substantial; for instance, autonomous vehicles might misinterpret modified road signs, potentially leading to fatal accidents, or software coding assistants might generate insecure code when prompted by certain organisations, making the organisations software systems vulnerable to attack if the generated code is not carefully inspected [3]. Deceptive instrumental alignment Plausibly, models could develop deceptive behaviors organically during training [8]. These models exhibit compliant behaviors in training and evaluation phases but deviate from their developer-defined goals once deployed. While naturally occurring deceptive models have not yet been reported, the training process does select for such behaviour [6].

This review synthesizes the nascent but critical field of developmental interpretability for Large Language Models. We chart the field's evolution from static, post-hoc analysis of trained models to a dynamic investigation of the training process itself. We begin by surveying the foundational methodologies, including representational probing, causal tracing, and circuit analysis, that enable researchers to deconstruct the learning process. The core of this review examines the developmental arc of LLM capabilities, detailing key findings on the formation and composition of computational circuits, the biphasic nature of knowledge acquisition, the transient dynamics of learning strategies like in-context learning, and the phenomenon of emergent abilities as phase transitions in training. We explore illuminating parallels with human cognitive and linguistic development, which provide valuable conceptual frameworks for understanding LLM learning. Finally, we argue that this developmental perspective is not merely an academic exercise but a cornerstone of proactive AI safety, offering a pathway to predict, monitor, and align the processes by which models acquire their capabilities. We conclude by outlining the grand challenges facing the field, such as scalability and automation, and propose a research agenda for building more transparent, reliable, and beneficial AI systems.

Developing dataset-specific models involves iterative fine-tuning and optimization, incurring significant costs over time. This study investigates the transferability of state-of-the-art (SOTA) models trained on established datasets to an unexplored text dataset. The key question is whether the knowledge embedded within SOTA models from existing datasets can be harnessed to achieve high-performance results on a new domain. In pursuit of this inquiry, two well-established benchmarks, the STSB and Mohler datasets, are selected, while the recently introduced SPRAG dataset serves as the unexplored domain. By employing robust similarity metrics and statistical techniques, a meticulous comparative analysis of these datasets is conducted. The primary goal of this work is to yield comprehensive insights into the potential applicability and adaptability of SOTA models. The outcomes of this research have the potential to reshape the landscape of natural language processing (NLP) by unlocking the ability to leverage existing models for diverse datasets. This may lead to a reduction in the demand for resource-intensive, dataset-specific training, thereby accelerating advancements in NLP and paving the way for more efficient model deployment.

Language models are increasingly used in Brazil, but most evaluation remains English-centric. This paper presents Alvorada-Bench, a 4,515-question, text-only benchmark drawn from five Brazilian university entrance examinations. Evaluating twenty models under zero-shot, role-playing, and chain-of-thought prompting, producing 270,900 responses with structured self-reports of confidence, perceived difficulty, and Bloom level. The top models exceed 94% accuracy overall, but accuracy declines on Mathematics and on the engineering oriented IME and ITA exams, indicating persistent weaknesses in multi-step reasoning. Confidence is well calibrated and correlates with perceived difficulty, revealing that models can accurately assess their own certainty capabilities. A cost accuracy analysis shows that high accuracy is achievable at under $2 per 1K tokens. On ENEM 2024 the top model (O3) achieved perfect scores in Languages subject questions while even the weakest system (GPT-4.1 Nano) only underperforms humans in Mathematics. Through exams that distill decades of Brazilian educational priorities and assess millions of students yearly, Alvorada-Bench establishes whether language models can navigate the intersection of language, culture, and reasoning that defines academic readiness in Brazil.

Web agents have shown great promise in performing many tasks on ecommerce website. To assess their capabilities, several benchmarks have been introduced. However, current benchmarks in the e-commerce domain face two major problems. First, they primarily focus on product search tasks (e.g., Find an Apple Watch), failing to capture the broader range of functionalities offered by real-world e-commerce platforms such as Amazon, including account management and gift card operations. Second, existing benchmarks typically evaluate whether the agent completes the user query, but ignore the potential risks involved. In practice, web agents can make unintended changes that negatively impact the user account or status. For instance, an agent might purchase the wrong item, delete a saved address, or incorrectly configure an auto-reload setting. To address these gaps, we propose a new benchmark called Amazon-Bench. To generate user queries that cover a broad range of tasks, we propose a data generation pipeline that leverages webpage content and interactive elements (e.g., buttons, check boxes) to create diverse, functionality-grounded user queries covering tasks such as address management, wish list management, and brand store following. To improve the agent evaluation, we propose an automated evaluation framework that assesses both the performance and the safety of web agents. We systematically evaluate different agents, finding that current agents struggle with complex queries and pose safety risks. These results highlight the need for developing more robust and reliable web agents.

Depression is a common mental health condition that can lead to hopelessness, loss of interest, self-harm, and even suicide. Early detection is challenging due to individuals not self-reporting or seeking timely clinical help. With the rise of social media, users increasingly express emotions online, offering new opportunities for detection through text analysis. While prior research has focused on languages such as English, no studies exist for Sorani Kurdish. This work presents a machine learning and Natural Language Processing (NLP) approach to detect depression in Sorani tweets. A set of depression-related keywords was developed with expert input to collect 960 public tweets from X (Twitter platform). The dataset was annotated into three classes: Shows depression, Not-show depression, and Suspicious by academics and final year medical students at the University of Kurdistan Hewlêr. Four supervised models, including Support Vector Machines, Multinomial Naive Bayes, Logistic Regression, and Random Forest, were trained and evaluated, with Random Forest achieving the highest performance accuracy and F1-score of 80%. This study establishes a baseline for automated depression detection in Kurdish language contexts.

Large language models (LLMs) have rapidly advanced in recent years, achieving remarkable performance across a wide range of natural language processing tasks. However, this progress has come at the cost of increasingly large model sizes, which pose significant challenges for deployment, scalability, and energy efficiency. To address these limitations, post-training pruning has emerged as a promising approach for reducing model size and inference latency without the need for retraining. Despite these advantages, many existing pruning methods result in substantial performance degradation or require computationally expensive fine-tuning. In this work, we introduce Z-Pruner, a novel post-training pruning method designed to induce sparsity in pretrained LLMs without any retraining. Unlike conventional approaches, Z-Pruner leverages both weight update magnitudes and activation patterns to identify and eliminate redundant parameters more effectively. Our method is model-agnostic, efficient, and easy to implement. We evaluate Z-Pruner using multiple widely-used LLM architectures, including LLaMA-2, LLaMA-3, and OPT, across a diverse set of standard language benchmarks. Experimental results demonstrate that Z-Pruner surpasses state-of-the-art pruning methods that require intensive weight updates. Specifically, Z-Pruner achieves the lowest perplexity scores and the highest overall average score for zero-shot accuracy. We have made the corresponding codes publicly available at https://github.com/sazzadadib/Z-Pruner.

The high dimensional and semantically complex nature of textual Big data presents significant challenges for text clustering, which frequently lead to suboptimal groupings when using conventional techniques like k-means or hierarchical clustering. This work presents Semantic Deep Embedded Clustering (SDEC), an unsupervised text clustering framework that combines an improved autoencoder with transformer-based embeddings to overcome these challenges. This novel method preserves semantic relationships during data reconstruction by combining Mean Squared Error (MSE) and Cosine Similarity Loss (CSL) within an autoencoder. Furthermore, a semantic refinement stage that takes advantage of the contextual richness of transformer embeddings is used by SDEC to further improve a clustering layer with soft cluster assignments and distributional loss. The capabilities of SDEC are demonstrated by extensive testing on five benchmark datasets: AG News, Yahoo! Answers, DBPedia, Reuters 2, and Reuters 5. The framework not only outperformed existing methods with a clustering accuracy of 85.7% on AG News and set a new benchmark of 53.63% on Yahoo! Answers, but also showed robust performance across other diverse text corpora. These findings highlight the significant improvements in accuracy and semantic comprehension of text data provided by SDEC's advances in unsupervised text clustering.

As organizations adopt retrieval-augmented generation (RAG) for their knowledge management systems (KMS), traditional market research deliverables face new functional demands. While PDF reports and slides have long served human readers, they are now also "read" by AI systems to answer user questions. To future-proof reports being delivered today, this study evaluates information loss during their ingestion into RAG systems. It compares how well PDF and PowerPoint (PPTX) documents converted to Markdown can be used by an LLM to answer factual questions in an end-to-end benchmark. Findings show that while text is reliably extracted, significant information is lost from complex objects like charts and diagrams. This suggests a need for specialized, AI-native deliverables to ensure research insights are not lost in translation.