Large language models are increasingly deployed across diverse applications. This often includes tasks LLMs have not encountered during training. This implies that enumerating and obtaining the high-quality training data for all tasks is infeasible. Thus, we often need to rely on transfer learning using datasets with different characteristics, and anticipate out-of-distribution requests. Motivated by this practical need, we propose an analysis framework, building a transfer learning matrix and dimensionality reduction, to dissect these cross-task interactions. We train and analyze 10 models to identify latent abilities (e.g., Reasoning, Sentiment Classification, NLU, Arithmetic) and discover the side effects of the transfer learning. Our findings reveal that performance improvements often defy explanations based on surface-level dataset similarity or source data quality. Instead, hidden statistical factors of the source dataset, such as class distribution and generation length proclivities, alongside specific linguistic features, are actually more influential. This work offers insights into the complex dynamics of transfer learning, paving the way for more predictable and effective LLM adaptation.

We explore the effectiveness and reliability of an artificial intelligence (AI)-based grading system for a handwritten general chemistry exam, comparing AI-assigned scores to human grading across various types of questions. Exam pages and grading rubrics were uploaded as images to account for chemical reaction equations, short and long open-ended answers, numerical and symbolic answer derivations, drawing, and sketching in pencil-and-paper format. Using linear regression analyses and psychometric evaluations, the investigation reveals high agreement between AI and human graders for textual and chemical reaction questions, while highlighting lower reliability for numerical and graphical tasks. The findings emphasize the necessity for human oversight to ensure grading accuracy, based on selective filtering. The results indicate promising applications for AI in routine assessment tasks, though careful consideration must be given to student perceptions of fairness and trust in integrating AI-based grading into educational practice.

Bayesian optimization (BO) is a powerful tool for scientific discovery in chemistry, yet its efficiency is often hampered by the sparse experimental data and vast search space. Here, we introduce ChemBOMAS: a large language model (LLM)-enhanced multi-agent system that accelerates BO through synergistic data- and knowledge-driven strategies. Firstly, the data-driven strategy involves an 8B-scale LLM regressor fine-tuned on a mere 1% labeled samples for pseudo-data generation, robustly initializing the optimization process. Secondly, the knowledge-driven strategy employs a hybrid Retrieval-Augmented Generation approach to guide LLM in dividing the search space while mitigating LLM hallucinations. An Upper Confidence Bound algorithm then identifies high-potential subspaces within this established partition. Across the LLM-refined subspaces and supported by LLM-generated data, BO achieves the improvement of effectiveness and efficiency. Comprehensive evaluations across multiple scientific benchmarks demonstrate that ChemBOMAS set a new state-of-the-art, accelerating optimization efficiency by up to 5-fold compared to baseline methods.

This study addresses critical industrial challenges in e-commerce product categorization, namely platform heterogeneity and the structural limitations of existing taxonomies, by developing and deploying a multimodal hierarchical classification framework. Using a dataset of 271,700 products from 40 international fashion e-commerce platforms, we integrate textual features (RoBERTa), visual features (ViT), and joint vision-language representations (CLIP). We investigate fusion strategies, including early, late, and attention-based fusion within a hierarchical architecture enhanced by dynamic masking to ensure taxonomic consistency. Results show that CLIP embeddings combined via an MLP-based late-fusion strategy achieve the highest hierarchical F1 (98.59%), outperforming unimodal baselines. To address shallow or inconsistent categories, we further introduce a self-supervised "product recategorization" pipeline using SimCLR, UMAP, and cascade clustering, which discovered new, fine-grained categories (for example, subtypes of "Shoes") with cluster purities above 86%. Cross-platform experiments reveal a deployment-relevant trade-off: complex late-fusion methods maximize accuracy with diverse training data, while simpler early-fusion methods generalize more effectively to unseen platforms. Finally, we demonstrate the framework's industrial scalability through deployment in EURWEB's commercial transaction intelligence platform via a two-stage inference pipeline, combining a lightweight RoBERTa stage with a GPU-accelerated multimodal stage to balance cost and accuracy.

With the rapid advancement of large language models (LLMs), retrieval-augmented generation (RAG) has emerged as a critical approach to supplement the inherent knowledge limitations of LLMs. However, due to the typically large volume of retrieved information, RAG tends to operate with long context lengths. From the perspective of entropy engineering, we identify unconstrained entropy growth and attention dilution due to long retrieval context as significant factors affecting RAG performance. In this paper, we propose the balanced entropy-engineered RAG (BEE-RAG) framework, which improves the adaptability of RAG systems to varying context lengths through the principle of entropy invariance. By leveraging balanced context entropy to reformulate attention dynamics, BEE-RAG separates attention sensitivity from context length, ensuring a stable entropy level. Building upon this, we introduce a zero-shot inference strategy for multi-importance estimation and a parameter-efficient adaptive fine-tuning mechanism to obtain the optimal balancing factor for different settings. Extensive experiments across multiple RAG tasks demonstrate the effectiveness of BEE-RAG.

The increasing use of Large Language Models (LLMs) in agentic applications highlights the need for robust safety guard models. While content safety in English is well-studied, non-English languages lack similar advancements due to the high cost of collecting culturally aligned labeled datasets. We present CultureGuard, a novel solution for curating culturally aligned, high-quality safety datasets across multiple languages. Our approach introduces a four-stage synthetic data generation and filtering pipeline: cultural data segregation, cultural data adaptation, machine translation, and quality filtering. This pipeline enables the conversion and expansion of the Nemotron-Content-Safety-Dataset-V2 English safety dataset into eight distinct languages: Arabic, German, Spanish, French, Hindi, Japanese, Thai, and Chinese. The resulting dataset, Nemotron-Safety-Guard-Dataset-v3, comprises 386,661 samples in 9 languages and facilitates the training of Llama-3.1-Nemotron-Safety-Guard-8B-v3 via LoRA-based fine-tuning. The final model achieves state-of-the-art performance on several multilingual content safety benchmarks. Furthermore, we show our moderately multilingual fine-tuning enables robust cross-lingual transfer and strong zero-shot generalization to unseen languages. We also benchmark the latest open LLMs on multilingual safety and observe that these LLMs are more prone to give unsafe responses when prompted in non-English languages. This work advances multilingual LLM safety by enabling the development of culturally aware safety guard models.

Generalizing to complex and temporally extended task objectives and safety constraints remains a critical challenge in reinforcement learning (RL). Linear temporal logic (LTL) offers a unified formalism to specify such requirements, yet existing methods are limited in their abilities to handle nested long-horizon tasks and safety constraints, and cannot identify situations when a subgoal is not satisfiable and an alternative should be sought. In this paper, we introduce GenZ-LTL, a method that enables zero-shot generalization to arbitrary LTL specifications. GenZ-LTL leverages the structure of Büchi automata to decompose an LTL task specification into sequences of reach-avoid subgoals. Contrary to the current state-of-the-art method that conditions on subgoal sequences, we show that it is more effective to achieve zero-shot generalization by solving these reach-avoid problems \textit{one subgoal at a time} through proper safe RL formulations. In addition, we introduce a novel subgoal-induced observation reduction technique that can mitigate the exponential complexity of subgoal-state combinations under realistic assumptions. Empirical results show that GenZ-LTL substantially outperforms existing methods in zero-shot generalization to unseen LTL specifications.

Large language models (LLMs) present new opportunities for creating pedagogical agents that engage in meaningful dialogue to support student learning. However, the current use of LLM systems like ChatGPT in classrooms often lacks the solid theoretical foundation found in earlier intelligent tutoring systems. To bridge this gap, we propose a framework that combines Evidence-Centered Design with Social Cognitive Theory for adaptive scaffolding in LLM-based agents focused on STEM+C learning. We illustrate this framework with In-quizzitor, an LLM-based formative assessment agent that integrates human-AI hybrid intelligence and provides feedback grounded in cognitive science principles. Our findings show that Inquizzitor delivers high-quality assessment and interaction aligned with core learning theories, offering teachers effective guidance that students value. This research underscores the potential for theory-driven LLM integration in education, highlighting the ability of these systems to provide adaptive and principled instruction.

We investigate the phenomenon of neuron universality in independently trained GPT-2 Small models, examining these universal neurons-neurons with consistently correlated activations across models-emerge and evolve throughout training. By analyzing five GPT-2 models at five checkpoints, we identify universal neurons through pairwise correlation analysis of activations over a dataset of 5 million tokens. Ablation experiments reveal significant functional impacts of universal neurons on model predictions, measured via cross entropy loss. Additionally, we quantify neuron persistence, demonstrating high stability of universal neurons across training checkpoints, particularly in early and deeper layers. These findings suggest stable and universal representational structures emerge during language model training.

As large reasoning models (LRMs) grow more capable, chain-of-thought (CoT) reasoning introduces new safety challenges. Existing SFT-based safety alignment studies dominantly focused on filtering prompts with safe, high-quality responses, while overlooking hard prompts that always elicit harmful outputs. To fill this gap, we introduce UnsafeChain, a safety alignment dataset constructed from hard prompts with diverse sources, where unsafe completions are identified and explicitly corrected into safe responses. By exposing models to unsafe behaviors and guiding their correction, UnsafeChain enhances safety while preserving general reasoning ability. We fine-tune three LRMs on UnsafeChain and compare them against recent SafeChain and STAR-1 across six out-of-distribution and five in-distribution benchmarks. UnsafeChain consistently outperforms prior datasets, with even a 1K subset matching or surpassing baseline performance, demonstrating the effectiveness and generalizability of correction-based supervision. We release our dataset and code at https://github.com/mbzuai-nlp/UnsafeChain