In the aftermath of COVID-19, many universities implemented supplementary "reinforcement" roles to support students in demanding courses. Although the name for such roles may differ between institutions, the underlying idea of providing structured supplementary support is common. However, these roles were often poorly defined, lacking structured materials, pedagogical oversight, and integration with the core teaching team. This paper reports on the redesign of reinforcement sessions in a challenging undergraduate course on formal methods and computational models, using a large language model (LLM) as a reflective planning tool. The LLM was prompted to simulate the perspective of a second-year student, enabling the identification of conceptual bottlenecks, gaps in intuition, and likely reasoning breakdowns before classroom delivery. These insights informed a structured, repeatable session format combining targeted review, collaborative examples, independent student work, and guided walkthroughs. Conducted over a single semester, the intervention received positive student feedback, indicating increased confidence, reduced anxiety, and improved clarity, particularly in abstract topics such as the pumping lemma and formal language expressive power comparisons. The findings suggest that reflective, instructor-facing use of LLMs can enhance pedagogical design in theoretically dense domains and may be adaptable to other cognitively demanding computer science courses.

Vision-Language Models (VLMs) suffer from catastrophic forgetting when sequentially fine-tuned on new tasks, degrading performance on previously learned foundational and task-specific capabilities. While multi-task learning can mitigate forgetting, it requires simultaneous access to all datasets and imposes computational overhead that scales linearly with the number of tasks. In this work, we introduce a routing-based approach that enables the integration of new tasks while preserving the foundational knowledge acquired during pretraining. We evaluate our method using InternVL-2 models (2B and 8B parameters) and demonstrate that routing preserves the model's foundational capabilities by maintaining performance on general-purpose benchmarks such as ChartQA, MMBench, and DocVQA, while simultaneously improving accuracy on specialized tasks. Importantly, our approach achieves this without requiring concurrent access to data from all tasks, avoiding the significant computational and data overhead associated with traditional multi-task learning. We further conduct extensive ablation studies to evaluate the scalability and robustness of routing-based learning, showing that the approach is resilient to a growing number of tasks and performs particularly well when new tasks are semantically related. Finally, we show that the routing mechanism enables superior cross-modal transfer between language and vision capabilities, allowing knowledge learned in one modality to enhance performance in another capability not achieved by existing continual learning methods.

The Abstraction and Reasoning Corpus remains one of the most compelling and challenging benchmarks for tracking progress toward achieving Artificial General Intelligence. In contrast to other evaluation datasets designed to assess an agent's task-specific skills or accumulated knowledge, the ARC-AGI suite is specifically targeted at measuring skill acquisition efficiency, a trait that has (so far) been lacking in even the most sophisticated machine learning systems. For algorithms that require extensive intra-task exemplars, a significant constraint imposed by ARC-AGI is the modest cardinality of its demonstration set, comprising a small number of $\langle$ input, output $\rangle$ grids per task specifying the corresponding transformation. To embellish the space of viable sample pairs, this paper introduces ARC-GEN, an open-source procedural generator aimed at extending the original ARC-AGI training dataset as faithfully as possible. Unlike prior efforts, our generator is both exhaustive (covering all four-hundred tasks) and mimetic (more closely honoring the distributional properties and characteristics embodied in the initial ARC-AGI-1 release). We also discuss the use of this generator in establishing a static benchmark suite to verify the correctness of programs submitted to the 2025 Google Code Golf Championship.

The remarkable capabilities of Large Language Models (LLMs) often need to be tailored for specific applications, requiring the integration of new knowledge or the acquisition of new skills. While full fine-tuning is a powerful adaptation method, it is computationally expensive and can lead to a degradation of general reasoning abilities, a phenomenon known as catastrophic forgetting. A range of alternative techniques exists, each with its own trade-offs. In-Context Learning (ICL) is fast but limited by context length, while Parameter-Efficient Fine-Tuning (PEFT) methods like Low-Rank Adaptation (LoRA) offer a middle ground by minimizing parameter changes. However, the challenge of catastrophic forgetting persists, raising questions about the best adaptation strategy for a given task. This paper presents a comparative analysis of Supervised Finetuning (SFT), LoRA, and ICL in data-scarce scenarios. We find that LoRA provides the most effective balance, successfully instilling new skills with minimal impact on the base model's general knowledge. In contrast, while SFT excels at skill acquisition, it is highly susceptible to catastrophic forgetting. ICL is effective for incorporating factual knowledge but struggles with complex skills. Our findings offer a practical framework for selecting an LLM adaptation strategy. We highlight the critical distinction between skill acquisition and knowledge integration, clarify the trade-offs between task-specific performance and the preservation of general capabilities.

A central architectural question for both biological and artificial intelligence is whether judgment relies on specialized modules or a unified, domain-general resource. While the discovery of decodable neural representations for distinct concepts in Large Language Models (LLMs) has suggested a modular architecture, whether these representations are truly independent systems remains an open question. Here we provide evidence for a convergent architecture for evaluative judgment. Across a range of LLMs, we find that diverse evaluative judgments are computed along a dominant dimension, which we term the Valence-Assent Axis (VAA). This axis jointly encodes subjective valence ("what is good") and the model's assent to factual claims ("what is true"). Through direct interventions, we demonstrate this axis drives a critical mechanism, which is identified as the subordination of reasoning: the VAA functions as a control signal that steers the generative process to construct a rationale consistent with its evaluative state, even at the cost of factual accuracy. Our discovery offers a mechanistic account for response bias and hallucination, revealing how an architecture that promotes coherent judgment can systematically undermine faithful reasoning.

While new benchmarks for large language models (LLMs) are being developed continuously to catch up with the growing capabilities of new models and AI in general, using and evaluating LLMs in non-English languages remains a little-charted landscape. We give a concise overview of recent developments in LLM benchmarking, and then propose a new taxonomy for the categorization of benchmarks that is tailored to multilingual or non-English use scenarios. We further propose a set of best practices and quality standards that could lead to a more coordinated development of benchmarks for European languages. Among other recommendations, we advocate for a higher language and culture sensitivity of evaluation methods.

Auxiliary Learning (AL) is a form of multi-task learning in which a model trains on auxiliary tasks to boost performance on a primary objective. While AL has improved generalization across domains such as navigation, image classification, and NLP, it often depends on human-labeled auxiliary tasks that are costly to design and require domain expertise. Meta-learning approaches mitigate this by learning to generate auxiliary tasks, but typically rely on gradient based bi-level optimization, adding substantial computational and implementation overhead. We propose RL-AUX, a reinforcement-learning (RL) framework that dynamically creates auxiliary tasks by assigning auxiliary labels to each training example, rewarding the agent whenever its selections improve the performance on the primary task. We also explore learning per-example weights for the auxiliary loss. On CIFAR-100 grouped into 20 superclasses, our RL method outperforms human-labeled auxiliary tasks and matches the performance of a prominent bi-level optimization baseline. We present similarly strong results on other classification datasets. These results suggest RL is a viable path to generating effective auxiliary tasks.

The path to higher network autonomy in 6G lies beyond the mere optimization of key performance indicators (KPIs). While KPIs have enabled automation gains under TM Forum Levels 1--3, they remain numerical abstractions that act only as proxies for the real essence of communication networks: seamless connectivity, fairness, adaptability, and resilience. True autonomy requires perceiving and reasoning over the network environment as it is. Such progress can be achieved through \emph{agentic AI}, where large language model (LLM)-powered agents perceive multimodal telemetry, reason with memory, negotiate across domains, and act via APIs to achieve multi-objective goals. However, deploying such agents introduces the challenge of cognitive biases inherited from human design, which can distort reasoning, negotiation, tool use, and actuation. Between neuroscience and AI, this paper provides a tutorial on a selection of well-known biases, including their taxonomy, definition, mathematical formulation, emergence in telecom systems and the commonly impacted agentic components. The tutorial also presents various mitigation strategies tailored to each type of bias. The article finally provides two practical use-cases, which tackle the emergence, impact and mitigation gain of some famous biases in 6G inter-slice and cross-domain management. In particular, anchor randomization, temporal decay and inflection bonus techniques are introduced to specifically address anchoring, temporal and confirmation biases. This avoids that agents stick to the initial high resource allocation proposal or decisions that are recent and/or confirming a prior hypothesis. By grounding decisions in a richer and fairer set of past experiences, the quality and bravery of the agentic agreements in the second use-case, for instance, are leading to $\times 5$ lower latency and around $40\%$ higher energy saving.

Cascade ranking is a widely adopted paradigm in large-scale information retrieval systems for Top-K item selection. However, the Top-K operator is non-differentiable, hindering end-to-end training. Existing methods include Learning-to-Rank approaches (e.g., LambdaLoss), which optimize ranking metrics like NDCG and suffer from objective misalignment, and differentiable sorting-based methods (e.g., ARF, LCRON), which relax permutation matrices for direct Top-K optimization but introduce gradient conflicts through matrix aggregation. A promising alternative is to directly construct a differentiable approximation of the Top-K selection operator, bypassing the use of soft permutation matrices. However, even state-of-the-art differentiable Top-K operator (e.g., LapSum) require $O(n \log n)$ complexity due to their dependence on sorting for solving the threshold. Thus, we propose DFTopK, a novel differentiable Top-K operator achieving optimal $O(n)$ time complexity. By relaxing normalization constraints, DFTopK admits a closed-form solution and avoids sorting. DFTopK also avoids the gradient conflicts inherent in differentiable sorting-based methods. We evaluate DFTopK on both the public benchmark RecFLow and an industrial system. Experimental results show that DFTopK significantly improves training efficiency while achieving superior performance, which enables us to scale up training samples more efficiently. In the online A/B test, DFTopK yielded a +1.77% revenue lift with the same computational budget compared to the baseline. To the best of our knowledge, this work is the first to introduce differentiable Top-K operators into recommendation systems and the first to achieve theoretically optimal linear-time complexity for Top-K selection. We have open-sourced our implementation to facilitate future research in both academia and industry.

Large Language Models (LLMs) have achieved impressive performance in diverse natural language processing tasks, but specialized domains such as Web3 present new challenges and require more tailored evaluation. Despite the significant user base and capital flows in Web3, encompassing smart contracts, decentralized finance (DeFi), non-fungible tokens (NFTs), decentralized autonomous organizations (DAOs), on-chain governance, and novel token-economics, no comprehensive benchmark has systematically assessed LLM performance in this domain. To address this gap, we introduce the DMind Benchmark, a holistic Web3-oriented evaluation suite covering nine critical subfields: fundamental blockchain concepts, blockchain infrastructure, smart contract, DeFi mechanisms, DAOs, NFTs, token economics, meme concept, and security vulnerabilities. Beyond multiple-choice questions, DMind Benchmark features domain-specific tasks such as contract debugging and on-chain numeric reasoning, mirroring real-world scenarios. We evaluated 26 models, including ChatGPT, Claude, DeepSeek, Gemini, Grok, and Qwen, uncovering notable performance gaps in specialized areas like token economics and security-critical contract analysis. While some models excel in blockchain infrastructure tasks, advanced subfields remain challenging. Our benchmark dataset and evaluation pipeline are open-sourced on https://huggingface.co/datasets/DMindAI/DMind_Benchmark, reaching number one in Hugging Face's trending dataset charts within a week of release.