This work aims to formalize some of the ways scientific concepts are formed in the process of theoretical physics discovery. Since this may at first seem like a task beyond the scope of the exact sciences (natural and formal sciences), we begin by presenting arguments for why scientific concept formation can be formalized. Then, we introduce type theory as a natural and well-suited framework for this formalization. We formalize what we call "ways of discovering new concepts" including concept distinction, property preservation, and concept change, as cognitive typing rules. Next, we apply these cognitive typing rules to two case studies of conceptual discovery in the history of physics: Einstein's reasoning leading to the impossibility of frozen waves, and his conceptual path to the relativity of time. In these historical episodes, we recast what a physicist might informally call "ways of discovering new scientific concepts" as compositional typing rules built from cognitive typing rules - thus formalizing them as scientific discovery mechanisms. Lastly, we computationally model the type-theoretic reconstruction of Einstein's conceptual path to the relativity of time as a program synthesis task.

As large language models (LLMs) expand multilingual capabilities, questions remain about the equity of their performance across languages. While many communities stand to benefit from AI systems, the dominance of English in training data risks disadvantaging non-English speakers. To test the hypothesis that such data disparities may affect model performance, this study compares two monolingual BERT models: one trained and tested entirely on Swahili data, and another on comparable English news data. To simulate how multilingual LLMs process non-English queries through internal translation and abstraction, we translated the Swahili news data into English and evaluated it using the English-trained model. This approach tests the hypothesis by evaluating whether translating Swahili inputs for evaluation on an English model yields better or worse performance compared to training and testing a model entirely in Swahili, thus isolating the effect of language consistency versus cross-lingual abstraction. The results prove that, despite high-quality translation, the native Swahili-trained model performed better than the Swahili-to-English translated model, producing nearly four times fewer errors: 0.36% vs. 1.47% respectively. This gap suggests that translation alone does not bridge representational differences between languages and that models trained in one language may struggle to accurately interpret translated inputs due to imperfect internal knowledge representation, suggesting that native-language training remains important for reliable outcomes. In educational and informational contexts, even small performance gaps may compound inequality. Future research should focus on addressing broader dataset development for underrepresented languages and renewed attention to multilingual model evaluation, ensuring the reinforcing effect of global AI deployment on existing digital divides is reduced.

Reinforcement Learning (RL) has proven highly effective at enhancing the complex reasoning abilities of Large Language Models (LLMs), yet underlying mechanisms driving this success remain largely opaque. Our analysis reveals that puzzling phenomena like ``aha moments", ``length-scaling'' and entropy dynamics are not disparate occurrences but hallmarks of an emergent reasoning hierarchy, akin to the separation of high-level strategic planning from low-level procedural execution in human cognition. We uncover a compelling two-phase dynamic: initially, a model is constrained by procedural correctness and must improve its low-level skills. The learning bottleneck then decisively shifts, with performance gains being driven by the exploration and mastery of high-level strategic planning. This insight exposes a core inefficiency in prevailing RL algorithms like GRPO, which apply optimization pressure agnostically and dilute the learning signal across all tokens. To address this, we propose Hierarchy-Aware Credit Assignment (HICRA), an algorithm that concentrates optimization efforts on high-impact planning tokens. Our extensive experiments validate that HICRA significantly outperforms strong baselines, and offer deep insights into how reasoning advances through the lens of strategic exploration.

In the realm of education, student evaluation holds equal significance to imparting knowledge. To be evaluated, students usually need to go through text-based academic assessment methods. Instructors need to make a diverse set of questions that need to be fair for all students to prove their adequacy over a particular topic. This can prove to be quite challenging as they may need to manually go through several different lecture materials. Our objective is to make this whole process much easier by implementing Automatic Question Answer Generation(AQAG), using a fine-tuned generative LLM. For tailoring the instructor's preferred question style (MCQ, conceptual, or factual questions), Prompt Engineering (PE) is being utilized. In this research, we propose to leverage unsupervised learning methods in NLP, primarily focusing on the English language. This approach empowers the base Meta-Llama 2-7B model to integrate the RACE dataset as training data for the fine-tuning process. Creating a customized model that will offer efficient solutions for educators, instructors, and individuals engaged in text-based evaluations. A reliable and efficient tool for generating questions and answers can free up valuable time and resources, thus streamlining their evaluation processes.

LLMs (Large language models) have revolutionized NLP (Natural Language Processing), yet their pedagogical value for low-resource languages remains unclear. We present GRILE (Grammar Romanian Inference and Language Explanations) , the first open benchmark of 1,151 multiple-choice questions harvested from Romanian high-stakes exams (National Evaluation, Baccalaureate, university admissions). GRILE enables us to probe two complementary abilities of seven state-of-the-art multilingual and Romanian-specific LLMs: (i) selecting the correct answer, and (ii) producing linguistically accurate explanations. While Gemini 2.5 Pro reaches 83% accuracy, most open-weight models stay below 65%, and 48% of their explanations contain factual or pedagogical flaws according to expert review. A detailed error analysis pinpoints systematic weaknesses in morphology and in applying the latest DOOM3 orthographic norms. All data, code and a public web demo are released to catalyze future research. Our findings expose open challenges for trustworthy educational NLP in low-resource settings and establish GRILE as a new test-bed for controllable explanation generation and evaluation.

The pursuit of general-purpose artificial intelligence depends on large language models (LLMs) that can handle both structured reasoning and open-ended generation. We present Omni-Thinker, a unified reinforcement learning (RL) framework that scales LLMs across diverse tasks by combining hybrid rewards with backward-transfer-guided scheduling. Hybrid rewards integrate rule-based verifiable signals with preference-based evaluations from an LLM-as-a-Judge, enabling learning in both deterministic and subjective domains. Our scheduler orders tasks according to accuracy backward transfer (BWT), reducing forgetting and improving multi-task performance. Experiments across four domains show gains of 6.2% over joint training and 12.4% over model merging. Moreover, we demonstrate that simple assumptions on accuracy transfer yield accurate predictions of curriculum outcomes, with entropy dynamics explaining deviations due to generative tasks. These findings underscore the importance of BWT-aware scheduling and hybrid supervision for scaling RL-based post-training toward general-purpose LLMs.

Large language models demonstrate the intriguing ability to perform unseen tasks via in-context learning. However, it remains unclear what mechanisms inside the model drive such task-level generalization. In this work, we approach this question through the lens of off-by-one addition (i.e., 1+1=3, 2+2=5, 3+3=?), a two-step, counterfactual task with an unexpected +1 function as a second step. Leveraging circuit-style interpretability techniques such as path patching, we analyze the models' internal computations behind their performance and present three key findings. First, we uncover a function induction mechanism that explains the model's generalization from standard addition to off-by-one addition. This mechanism resembles the structure of the induction head mechanism found in prior work and elevates it to a higher level of abstraction. Second, we show that the induction of the +1 function is governed by multiple attention heads in parallel, each of which emits a distinct piece of the +1 function. Finally, we find that this function induction mechanism is reused in a broader range of tasks, including synthetic tasks such as shifted multiple-choice QA and algorithmic tasks such as base-8 addition. Overall, our findings offer deeper insights into how reusable and composable structures within language models enable task-level generalization.

Large language models (LLMs) are increasingly trusted as automated judges, assisting evaluation and providing reward signals for training other models, particularly in reference-based settings like Reinforcement Learning with Verifiable Rewards (RLVR). However, we uncover a critical vulnerability even in this reference-based paradigm: generative reward models are systematically susceptible to reward hacking. We find that superficial inputs, which we term ''master keys'' such as non-word symbols (e.g., '':'' or ''.'') or generic reasoning openers (e.g., ''Thought process:'' or ''Let's solve this problem step by step.''), can consistently elicit false positive rewards without any substantive reasoning. Our systematic evaluation demonstrates this is a widespread failure affecting a diverse range of models, including leading proprietary systems such as GPT-o1 and Claude-4. These results challenge the assumed robustness of LLM judges and pose a significant threat to their reliability. To address this, we propose a simple yet effective data augmentation strategy using truncated model outputs as adversarial negative examples. The resulting Master Reward Models (Master-RMs) demonstrate state-of-the-art robustness against these ''master key'' attacks while maintaining high performance in standard evaluation settings. We supplement these findings with a comprehensive analysis of the vulnerability across model scales, prompt variations, and common inference-time strategies, offering insights to guide future research on robust LLM evaluation. We release our robust, general-domain reward models and the synthetic training data at https://huggingface.co/sarosavo/Master-RM and https://huggingface.co/datasets/sarosavo/Master-RM.

Abstract--Spiking Neural Networks (SNNs), inspired by biological neural mechanisms, represent a promising neuromorphic computing paradigm that offers energy-efficient alternatives to traditional Artificial Neural Networks (ANNs). Despite proven effectiveness, SNN architectures have struggled to achieve competitive performance on large-scale speech processing tasks. Two key challenges hinder progress: (1) the high computational overhead during training caused by multi-timestep spike firing, and (2) the absence of large-scale SNN architectures tailored to speech processing tasks. T o overcome the issues, we introduce Input-aware Multi-Level Spikeformer, i.e. IML-Spikeformer, a spiking Transformer architecture specifically designed for large-scale speech processing. Central to our design is the Input-aware Multi-Level Spike (IMLS) mechanism, which simulates multi-timestep spike firing within a single timestep using an adaptive, input-aware thresholding scheme. This module enhances the precision of attention maps and enables modeling of multi-scale temporal dependencies in speech signals. Experiments demonstrate that IML-Spikeformer achieves word error rates of 6.0% on AiShell-1 and 3.4% on Librispeech-960, comparable to conventional ANN transformers while reducing theoretical inference energy consumption by 4.64 and 4.32 respectively. The high computational cost of such models has motivated the search for energy-efficient alternatives. Zeyang Song and Haizhou Li are with the Department of Electrical and Computer Engineering, National University of Singapore, Singapore 119077 Haizhou Li is also with the School of Artificial Intelligence, The Chinese University of Hong Kong, Shenzhen, 518172 China; Shenzhen Loop Area Institute, Shenzhen, China.

The rapid advancement of Large Language Models (LLMs) has transformed various domains, particularly computer science (CS) education. These models exhibit remarkable capabilities in code-related tasks and problem-solving, raising questions about their potential and limitations in advanced CS contexts. This study presents a novel bilingual (English-Romanian) multimodal (text and image) dataset of multiple-choice questions derived from a high-level computer science competition. A particularity of our dataset is that the problems are conceived such that some of them are easier solved using reasoning on paper, while for others writing code is more efficient. We systematically evaluate State of The Art LLMs on this dataset, analyzing their performance on theoretical programming tasks. Our findings reveal the strengths and limitations of current LLMs, including the influence of language choice (English vs. Romanian), providing insights into their applicability in CS education and competition settings. We also address critical ethical considerations surrounding educational integrity and the fairness of assessments in the context of LLM usage. These discussions aim to inform future educational practices and policies. To support further research, our dataset will be made publicly available in both English and Romanian. Additionally, we release an educational application tailored for Romanian students, enabling them to self-assess using the dataset in an interactive and practice-oriented environment.