This survey provides the first systematic review of Arabic LLM benchmarks, analyzing 40+ evaluation benchmarks across NLP tasks, knowledge domains, cultural understanding, and specialized capabilities. We propose a taxonomy organizing benchmarks into four categories: Knowledge, NLP Tasks, Culture and Dialects, and Target-Specific evaluations. Our analysis reveals significant progress in benchmark diversity while identifying critical gaps: limited temporal evaluation, insufficient multi-turn dialogue assessment, and cultural misalignment in translated datasets. We examine three primary approaches: native collection, translation, and synthetic generation discussing their trade-offs regarding authenticity, scale, and cost. This work serves as a comprehensive reference for Arabic NLP researchers, providing insights into benchmark methodologies, reproducibility standards, and evaluation metrics while offering recommendations for future development.

While chain-of-thought (CoT) distillation from advanced large language models (LLMs) has proven effective in general reasoning tasks, it struggles in scientific domains where even advanced models often produce incorrect or superficial reasoning due to high complexity and specialized knowledge requirements. Directly distilling from such flawed outputs results in low-quality training data and limits the performance of smaller student models. To overcome this, we propose CoT-Evo, an evolutionary CoT distillation framework. It begins by constructing a diverse pool of reasoning trajectories from multiple LLM thinkers, enriches them with automatically retrieved domain knowledge, and iteratively refines the trajectories using novelty-driven selection, reflective recombination and mutation. The refinement is guided by a fitness function that evaluates answer correctness, coherence, and effective knowledge utilization. This results in a high-quality CoT dataset tailored for scientific reasoning. We employ this evolved dataset to fine-tune a compact model, which achieves state-of-the-art performance on scientific reasoning benchmarks. Our work establishes a scalable approach to synthesizing high-fidelity scientific reasoning data from diverse and fallible LLMs.

Large language models have recently demonstrated advanced capabilities in solving IMO and Putnam problems; yet their role in research mathematics has remained fairly limited. The key difficulty is verification: suggested proofs may look plausible, but cannot be trusted without rigorous checking. We present a framework, called LLM+CAS, and an associated tool, O-Forge, that couples frontier LLMs with a computer algebra systems (CAS) in an In-Context Symbolic Feedback loop to produce proofs that are both creative and symbolically verified. Our focus is on asymptotic inequalities, a topic that often involves difficult proofs and appropriate decomposition of the domain into the "right" subdomains. Many mathematicians, including Terry Tao, have suggested that using AI tools to find the right decompositions can be very useful for research-level asymptotic analysis. In this paper, we show that our framework LLM+CAS turns out to be remarkably effective at proposing such decompositions via a combination of a frontier LLM and a CAS. More precisely, we use an LLM to suggest domain decomposition, and a CAS (such as Mathematica) that provides a verification of each piece axiomatically. Using this loop, we answer a question posed by Terence Tao: whether LLMs coupled with a verifier can be used to help prove intricate asymptotic inequalities. More broadly, we show how AI can move beyond contest math towards research-level tools for professional mathematicians.

Large language models (LLMs) are increasingly deployed across high-impact domains, from clinical decision support and legal analysis to hiring and education, making fairness and bias evaluation before deployment critical. However, existing evaluations lack grounding in real-world scenarios and do not account for differences in harm severity, e.g., a biased decision in surgery should not be weighed the same as a stylistic bias in text summarization. To address this gap, we introduce HALF (Harm-Aware LLM Fairness), a deployment-aligned framework that assesses model bias in realistic applications and weighs the outcomes by harm severity. HALF organizes nine application domains into three tiers (Severe, Moderate, Mild) using a five-stage pipeline. Our evaluation results across eight LLMs show that (1) LLMs are not consistently fair across domains, (2) model size or performance do not guarantee fairness, and (3) reasoning models perform better in medical decision support but worse in education. We conclude that HALF exposes a clear gap between previous benchmarking success and deployment readiness.

We present Entropic Mutual-Information Geometry Large-Language Model Alignment (ENIGMA), a novel approach to Large-Language Model (LLM) training that jointly improves reasoning, alignment and robustness by treating an organisation's policies/principles as directions to move on a model's information manifold. Our single-loop trainer combines Group-Relative Policy Optimisation (GRPO), an on-policy, critic-free RL method with Chain-of-Thought (CoT)-format only rewards; a Self-Supervised Alignment with Mutual Information (SAMI)-style symmetric InfoNCE auxiliary; and an entropic Sinkhorn optimal-transport regulariser on hidden-state distributions to bound geometry drift. We also introduce infoNCE metrics that specialise to a standard MI lower bound under matched negatives to measure how strongly a model's CoT encodes these policies. These metrics include a Sufficiency Index (SI) that enables the selection and creation of principles that maximise downstream performance prior to training. In our experiments using small (1B) LLMs, high-SI principles predict steadier training dynamics and improved benchmark performance over GRPO ablations. Our information-geometry analysis of trained models validates desirable structural change in the manifold. These results support our hypothesis that reasoning, alignment, and robustness are projections of a single information-geometric objective, and that models trained using ENIGMA demonstrate principled reasoning without the use of a reward model, offering a path to trusted capability

As the global reach of online higher education continues to grow, universities are increasingly accommodating students from diverse cultural backgrounds (Tereshko et al., 2024). This can present a number of challenges including linguistic barriers (Ullah et al., 2021), cultural differences in learning style (Omidvar & Tan, 2012), cultural sensitivity in course design (Nguyen, 2022) and perceived isolation when students feel their perspectives or experiences are not reflected or valued in the learning environment (Hansen-Brown et al., 2022). Ensuring active engagement and reasonable learning outcomes in such a environments requires distance educational systems that are not only adaptive but also culturally resonant (Dalle et al., 2024). Both embodied and virtual AI-Agents have great potential in this regard as they can facilitate personalized learning and adapt their interactions and content delivery to align with students' cultural context. In addition, Generative AI (GAI), such as, Large Language Models (LLMs) can amplify the potential for these culturally aware AI agents to address educational challenges due to their advanced capacity for understanding and generating contextually relevant content (Wang et al., 2024). This chapter reviews existing research and suggests the usage of culturally aware AI-Agents, powered by GAI, to foster engagement and improve learning outcomes in culturally diverse online higher education environments.

Healthcare research and development face significant obstacles due to data scarcity and stringent privacy regulations, such as HIPAA and the GDPR, restricting access to essential real-world medical data. These limitations impede innovation, delay robust AI model creation, and hinder advancements in patient-centered care. Synthetic data generation offers a transformative solution by producing artificial datasets that emulate real data statistics while safeguarding patient privacy. We introduce a novel hybrid framework for high-fidelity healthcare data synthesis integrating five augmentation methods: noise injection, interpolation, Gaussian Mixture Model (GMM) sampling, Conditional Variational Autoencoder (CVAE) sampling, and SMOTE, combined via a reinforcement learning-based dynamic weight selection mechanism. Its key innovations include advanced calibration techniques -- moment matching, full histogram matching, soft and adaptive soft histogram matching, and iterative refinement -- that align marginal distributions and preserve joint feature dependencies. Evaluated on the Breast Cancer Wisconsin (UCI Repository) and Khulna Medical College cardiology datasets, our calibrated hybrid achieves Wasserstein distances as low as 0.001 and Kolmogorov-Smirnov statistics around 0.01, demonstrating near-zero marginal discrepancy. Pairwise trend scores surpass 90%, and Nearest Neighbor Adversarial Accuracy approaches 50%, confirming robust privacy protection. Downstream classifiers trained on synthetic data achieve up to 94% accuracy and F1 scores above 93%, comparable to models trained on real data. This scalable, privacy-preserving approach matches state-of-the-art methods, sets new benchmarks for joint-distribution fidelity in healthcare, and supports sensitive AI applications.

Designing protein binders targeting specific sites, which requires to generate realistic and functional interaction patterns, is a fundamental challenge in drug discovery. Current structure-based generative models are limited in generating nterfaces with sufficient rationality and interpretability. In this paper, we propose Retrieval-Augmented Diffusion for Aligned interface (RADiAnce), a new framework that leverages known interfaces to guide the design of novel binders. By unifying retrieval and generation in a shared contrastive latent space, our model efficiently identifies relevant interfaces for a given binding site and seamlessly integrates them through a conditional latent diffusion generator, enabling cross-domain interface transfer. Extensive exeriments show that RADiAnce significantly outperforms baseline models across multiple metrics, including binding affinity and recovery of geometries and interactions. Additional experimental results validate cross-domain generalization, demonstrating that retrieving interfaces from diverse domains, such as peptides, antibodies, and protein fragments, enhances the generation performance of binders for other domains. Our work establishes a new paradigm for protein binder design that successfully bridges retrieval-based knowledge and generative AI, opening new possibilities for drug discovery.

Detecting harmful AI actions is important as AI agents gain adoption. Chain-of-thought (CoT) monitoring is one method widely used to detect adversarial attacks and AI misalignment. However, attackers and misaligned models might evade CoT monitoring through ciphered reasoning: reasoning hidden in encrypted, translated, or compressed text. To assess this risk, we test whether models can perform ciphered reasoning. For each of 28 different ciphers, we fine-tune and prompt up to 10 models to reason in that cipher. Across the models we test, we find an asymmetry: model accuracy can drop significantly when reasoning in ciphered text, even though models demonstrate comprehension of ciphered text by being able to translate it accurately to English. Even frontier models struggle with lesser-known ciphers, although they can reason accurately in well-known ciphers like rot13. We show that ciphered reasoning capability correlates with cipher prevalence in pretraining data. We also identify scaling laws showing that ciphered reasoning capability improves slowly with additional fine-tuning data. Our work suggests that evading CoT monitoring using ciphered reasoning may be an ineffective tactic for current models and offers guidance on constraining the development of this capability in future frontier models. Modern large language models (LLMs) rely on chain-of-thought (CoT) (Wei et al., 2022) to achieve strong performance (Guo et al., 2025). CoT increases the proportion of model computation that occurs in natural language (Korbak et al., 2025), which allows automated systems to monitor model CoTs for misaligned behavior. CoT monitoring has been employed in tasks as diverse as reinforcement learning (RL) training of frontier models (Baker et al., 2025), AI control (Kutasov et al., 2025), frontier model evaluation (METR, 2025), agent monitoring (Meinke et al., 2024), and jailbreak safeguards (Sharma et al., 2025) to supervise model behavior. In many of these settings, access to legible reasoning traces is critical; without it, monitors are much less capable (Baker et al., 2025).

When a new release of a foundation model is published, practitioners typically need to repeat full fine-tuning, even if the same task has already been solved in the previous version. A promising alternative is to reuse the parameter changes (i.e., task vectors) that capture how a model adapts to a specific task. However, they often fail to transfer across different pre-trained models due to their misaligned parameter space. In this work, we show that the key to successful transfer lies in the sign structure of the gradients of the new model. Based on this insight, we propose GradFix, a novel method that approximates the ideal gradient sign structure and leverages it to transfer knowledge using only a handful of labeled samples. Notably, this requires no additional fine-tuning: the adaptation is achieved by computing a few gradients at the target model and masking the source task vector accordingly. This yields an update that is locally aligned with the target loss landscape, effectively rebasing the task vector onto the new pre-training. We provide a theoretical guarantee that our method ensures first-order descent. Empirically, we demonstrate significant performance gains on vision and language benchmarks, consistently outperforming naive task vector addition and few-shot fine-tuning.