The rise of social media and online communication platforms has led to the spread of Arabic textual posts and memes as a key form of digital expression. While these contents can be humorous and informative, they are also increasingly being used to spread offensive language and hate speech. Consequently, there is a growing demand for precise analysis of content in Arabic text and memes. This paper explores the potential of large language models to effectively identify hope, hate speech, offensive language, and emotional expressions within such content. We evaluate the performance of base LLMs, fine-tuned LLMs, and pre-trained embedding models. The evaluation is conducted using a dataset of Arabic textual speech and memes proposed in the ArabicNLP MAHED 2025 challenge. The results underscore the capacity of LLMs such as GPT-4o-mini, fine-tuned with Arabic textual speech, and Gemini Flash 2.5, fine-tuned with Arabic memes, to deliver the superior performance. They achieve up to 72.1%, 57.8%, and 79.6% macro F1 scores for tasks 1, 2, and 3, respectively, and secure first place overall in the Mahed 2025 challenge. The proposed solutions offer a more nuanced understanding of both text and memes for accurate and efficient Arabic content moderation systems.

The increasing input sequence length in Large Language Models (LLMs) puts significant pressure on key-value (KV) cache storage, making efficient inference challenging. Explicitly distinguishing attention behavior into our self-defined surface memorization and logic construction reveals essential roles in long-context reasoning. We observe that an individual attention head can display various behaviors, with nearly 98.5% effectively ignoring completely irrelevant information. The remaining 1.5% behaves as logic construction, and 0.5% behaves as surface memorization. Based on layer- and head-wise integration, we propose a novel two-stage SurfaceLogicKV method to utilize these attention behaviors for KV Cache compression. As a result, it achieves improved compressing robustness while maintaining competitive performance across various tasks and long sequences compared to baselines or even FullKV in some specific situations

Large language models (LLMs) often have a fixed knowledge cutoff, limiting their accuracy on emerging information. We present ALAS (Autonomous Learning Agent System), a modular pipeline that continuously updates an LLM's knowledge with minimal human intervention. ALAS autonomously generates a learning curriculum for a target domain, retrieves up-to-date information from the web (with citations), distills this into question-answer training data, and fine-tunes the model through supervised fine-tuning (SFT) and direct preference optimization (DPO). It iteratively evaluates performance and revises the curriculum, enabling long-term continual learning. We demonstrate ALAS's ability to self-improve a model on rapidly evolving domains (e.g., new Python releases, latest security CVEs, academic trends), significantly boosting post-cutoff question answering accuracy (from 15% to 90% on average) without manual dataset curation. The system emphasizes modularity and reproducibility: each component (planning, retrieval, distillation, memory, fine-tuning) is interchangeable and built on standard APIs. We discuss comparative baselines (e.g., retrieval-augmented generation vs. fine-tuning) and show that ALAS achieves 90% accuracy on knowledge-updated queries with minimal engineering overhead. Finally, we outline limitations (cost, dependency on source quality) and future directions for autonomous lifelong learning in LLMs.

The advent of Deep Research agents has substantially reduced the time required for conducting extensive research tasks. However, these tasks inherently demand rigorous standards of factual accuracy and comprehensiveness, necessitating thorough evaluation before widespread adoption. In this paper, we propose ReportBench, a systematic benchmark designed to evaluate the content quality of research reports generated by large language models (LLMs). Our evaluation focuses on two critical dimensions: (1) the quality and relevance of cited literature, and (2) the faithfulness and veracity of the statements within the generated reports. ReportBench leverages high-quality published survey papers available on arXiv as gold-standard references, from which we apply reverse prompt engineering to derive domain-specific prompts and establish a comprehensive evaluation corpus. Furthermore, we develop an agent-based automated framework within ReportBench that systematically analyzes generated reports by extracting citations and statements, checking the faithfulness of cited content against original sources, and validating non-cited claims using web-based resources. Empirical evaluations demonstrate that commercial Deep Research agents such as those developed by OpenAI and Google consistently generate more comprehensive and reliable reports than standalone LLMs augmented with search or browsing tools. However, there remains substantial room for improvement in terms of the breadth and depth of research coverage, as well as factual consistency. The complete code and data will be released at the following link: https://github.com/ByteDance-BandAI/ReportBench

As multimodal large language models (MLLMs) grow increasingly capable, fixed benchmarks are gradually losing their effectiveness in evaluating high-level scientific understanding. In this paper, we introduce the Multimodal Academic Cover benchmark (MAC), a live benchmark that could continuously evolve with scientific advancement and model progress. MAC leverages over 25,000 image-text pairs sourced from issues of top-tier scientific journals such as Nature, Science, and Cell, challenging MLLMs to reason across abstract visual and textual scientific content. Experiments on our most recent yearly snapshot, MAC-2025, reveal that while MLLMs demonstrate strong perceptual abilities, their cross-modal scientific reasoning remains limited. To bridge this gap, we propose DAD, a lightweight inference-time approach that enhances MLLMs by extending MLLM visual features with language space reasoning, achieving performance improvements of up to 11%. Finally, we highlight the live nature of MAC through experiments on updating journal covers and models for curation, illustrating its potential to remain aligned with the frontier of human knowledge. We release our benchmark at https://github.com/mhjiang0408/MAC_Bench.

Warning: This research studies AI persuasion and bias amplification that could be misused; all experiments are for safety evaluation. Large Language Models (LLMs) now generate convincing, human-like text and are widely used in content creation, decision support, and user interactions. Yet the same systems can spread information or misinformation at scale and reflect social biases that arise from data, architecture, or training choices. This work examines how persuasion and bias interact in LLMs, focusing on how imperfect or skewed outputs affect persuasive impact. Specifically, we test whether persona-based models can persuade with fact-based claims while also, unintentionally, promoting misinformation or biased narratives. We introduce a convincer-skeptic framework: LLMs adopt personas to simulate realistic attitudes. Skeptic models serve as human proxies; we compare their beliefs before and after exposure to arguments from convincer models. Persuasion is quantified with Jensen-Shannon divergence over belief distributions. We then ask how much persuaded entities go on to reinforce and amplify biased beliefs across race, gender, and religion. Strong persuaders are further probed for bias using sycophantic adversarial prompts and judged with additional models. Our findings show both promise and risk. LLMs can shape narratives, adapt tone, and mirror audience values across domains such as psychology, marketing, and legal assistance. But the same capacity can be weaponized to automate misinformation or craft messages that exploit cognitive biases, reinforcing stereotypes and widening inequities. The core danger lies in misuse more than in occasional model mistakes. By measuring persuasive power and bias reinforcement, we argue for guardrails and policies that penalize deceptive use and support alignment, value-sensitive design, and trustworthy deployment.

Recent progress in large language models (LLMs) has showcased impressive proficiency in numerous Arabic natural language processing (NLP) applications. Nevertheless, their effectiveness in Arabic medical NLP domains has received limited investigation. This research examines the degree to which state-of-the-art LLMs demonstrate and articulate healthcare knowledge in Arabic, assessing their capabilities across a varied array of Arabic medical tasks. We benchmark several LLMs using a medical dataset proposed in the Arabic NLP AraHealthQA challenge in MedArabiQ2025 track. Various base LLMs were assessed on their ability to accurately provide correct answers from existing choices in multiple-choice questions (MCQs) and fill-in-the-blank scenarios. Additionally, we evaluated the capacity of LLMs in answering open-ended questions aligned with expert answers. Our results reveal significant variations in correct answer prediction accuracy and low variations in semantic alignment of generated answers, highlighting both the potential and limitations of current LLMs in Arabic clinical contexts. Our analysis shows that for MCQs task, the proposed majority voting solution, leveraging three base models (Gemini Flash 2.5, Gemini Pro 2.5, and GPT o3), outperforms others, achieving up to 77% accuracy and securing first place overall in the Arahealthqa 2025 shared task-track 2 (sub-task 1) challenge. Moreover, for the open-ended questions task, several LLMs were able to demonstrate excellent performance in terms of semantic alignment and achieve a maximum BERTScore of 86.44%.

Islamic inheritance domain holds significant importance for Muslims to ensure fair distribution of shares between heirs. Manual calculation of shares under numerous scenarios is complex, time-consuming, and error-prone. Recent advancements in Large Language Models (LLMs) have sparked interest in their potential to assist with complex legal reasoning tasks. This study evaluates the reasoning capabilities of state-of-the-art LLMs to interpret and apply Islamic inheritance laws. We utilized the dataset proposed in the ArabicNLP QIAS 2025 challenge, which includes inheritance case scenarios given in Arabic and derived from Islamic legal sources. Various base and fine-tuned models, are assessed on their ability to accurately identify heirs, compute shares, and justify their reasoning in alignment with Islamic legal principles. Our analysis reveals that the proposed majority voting solution, leveraging three base models (Gemini Flash 2.5, Gemini Pro 2.5, and GPT o3), outperforms all other models that we utilized across every difficulty level. It achieves up to 92.7% accuracy and secures the third place overall in Task 1 of the Qias 2025 challenge.

Constructing historical language models (LMs) plays a crucial role in aiding archaeological provenance studies and understanding ancient cultures. However, existing resources present major challenges for training effective LMs on historical texts. First, the scarcity of historical language samples renders unsupervised learning approaches based on large text corpora highly inefficient, hindering effective pre-training. Moreover, due to the considerable temporal gap and complex evolution of ancient scripts, the absence of comprehensive character encoding schemes limits the digitization and computational processing of ancient texts, particularly in early Chinese writing. To address these challenges, we introduce InteChar, a unified and extensible character list that integrates unencoded oracle bone characters with traditional and modern Chinese. InteChar enables consistent digitization and representation of historical texts, providing a foundation for robust modeling of ancient scripts. To evaluate the effectiveness of InteChar, we construct the Oracle Corpus Set (OracleCS), an ancient Chinese corpus that combines expert-annotated samples with LLM-assisted data augmentation, centered on Chinese oracle bone inscriptions. Extensive experiments show that models trained with InteChar on OracleCS achieve substantial improvements across various historical language understanding tasks, confirming the effectiveness of our approach and establishing a solid foundation for future research in ancient Chinese NLP.

Large Language Models (LLMs) face challenges in knowledge-intensive reasoning tasks like classic multi-hop question and answering, which involves reasoning across multiple facts. This difficulty arises because the chain of thoughts (CoTs) generated by LLMs in such tasks often deviate from real or a priori reasoning paths. In contrast, knowledge graphs (KGs) explicitly represent the logical connections between facts through entities and relationships. This reflects a significant gap. Meanwhile, large reasoning models (LRMs), such as o1, have demonstrated that long-step reasoning significantly enhances the performance of LLMs. Building on these insights, we propose KG-o1, a four-stage approach that integrates KGs to enhance the multi-hop reasoning abilities of LLMs. We first filter out initial entities and generate complex subgraphs. Secondly, we construct logical paths for subgraphs and then use knowledge graphs to build a dataset with a complex and extended brainstorming process, which trains LLMs to imitate long-term reasoning. Finally, we employ rejection sampling to generate a self-improving corpus for direct preference optimization (DPO), further refining the LLMs reasoning abilities. We conducted experiments on two simple and two complex datasets. The results show that KG-o1 models exhibit superior performance across all tasks compared to existing LRMs.