Abstract-- Tools that can generate computer code in response to inputs written in natural language, such as ChatGPT, pose an existential threat to Computer Science education in its current form, since students can now use these tools to solve assignments without much effort. While that risk has already been recognized by scholars, the proportion of the student body that is incurring in this new kind of plagiarism is still an open problem. We conducted a pilot study in a large CS class (n=120) to assess the feasibility of estimating AI plagiarism through anonymous surveys and interviews. More than 25% of the survey respondents admitted to committing AI plagiarism. Conversely, only one student accepted to be interviewed. Given the high levels of misconduct acknowledgment, we conclude that surveys are an effective method for studies on the matter, while interviews should be avoided or designed in a way that can entice participation. 1 INTRODUCTION Generative artificial intelligence (GenAI, not to be confused with general The generation is usually guided by an input text known as the "prompt". For example, giving the prompt "a vase of red flowers" to a GenAI model would generate an image depicting red flowers in a vase. Practical applications of GenAI are now mainstream thanks to advances in neural networks. In particular, the clever use of attention mechanisms and the subsequent development of the transformer architecture made efficient learning possible over large text corpora (Vaswani et al., 2023) . AI application based on a LLM, can convincingly engage in a conversation and answer questions across multiple subjects (OpenAI, 2022) . Research on applications of LLMs in education is still in its infancy, but looks promising. Personal tutoring systems (Chang, 2022), content explanation (Leinonen et al., 2023) and assignment generation ( Jury et al., 2024) are a few of the ideas that have been explored. From another perspective, LLMs are already a reality in schools.

As frontier artificial intelligence (AI) models rapidly advance, benchmarks are integral to comparing different models and measuring their progress in different task-specific domains. However, there is a lack of guidance on when and how benchmarks should be deprecated once they cease to effectively perform their purpose. This risks benchmark scores over-valuing model capabilities, or worse, obscuring capabilities and safety-washing. Based on a review of benchmarking practices, we propose criteria to decide when to fully or partially deprecate benchmarks, and a framework for deprecating benchmarks. Our work aims to advance the state of benchmarking towards rigorous and quality evaluations, especially for frontier models, and our recommendations are aimed to benefit benchmark developers, benchmark users, AI governance actors (across governments, academia, and industry panels), and policy makers.

Large Language Models (LLMs) are traditionally viewed as black-box algorithms, therefore reducing trustworthiness and obscuring potential approaches to increasing performance on downstream tasks. In this work, we apply an effective LLM decomposition method using a dictionary-learning approach with sparse autoencoders. This helps extract monosemantic features from polysemantic LLM neurons. Remarkably, our work identifies model-internal misunderstanding, allowing the automatic reformulation of the prompts with additional annotations to improve the interpretation by LLMs. Moreover, this approach demonstrates a significant performance improvement in downstream tasks, such as mathematical reasoning and metaphor detection.

Reward modeling (RM), which captures human preferences to align large language models (LLMs), is increasingly employed in tasks such as model finetuning, response filtering, and ranking. However, due to the inherent complexity of human preferences and the limited coverage of available datasets, reward models often fail under distributional shifts or adversarial perturbations. Existing approaches for identifying such failure modes typically rely on prior knowledge about preference distributions or failure attributes, limiting their practicality in real-world settings where such information is unavailable. In this work, we propose a tractable, preference-distribution agnostic method for discovering reward model failure modes via reward guided controlled decoding. Building on this, we introduce REFORM, a self-improving reward modeling framework that enhances robustness by using the reward model itself to guide the generation of falsely scored responses. These adversarial examples are then used to augment the training data and patch the reward model's misaligned behavior. We evaluate REFORM on two widely used preference datasets Anthropic Helpful Harmless (HH) and PKU Beavertails and demonstrate that it significantly improves robustness without sacrificing reward quality. Notably, REFORM preserves performance both in direct evaluation and in downstream policy training, and further improves alignment quality by removing spurious correlations.

Prompt engineering for LLMs remains complex, with existing frameworks either hiding complexity behind restrictive APIs or providing inflexible canned patterns that resist customization -- making sophisticated agentic programming challenging. We present the Prompt Declaration Language (PDL), a novel approach to prompt representation that tackles this fundamental complexity by bringing prompts to the forefront, enabling manual and automatic prompt tuning while capturing the composition of LLM calls together with rule-based code and external tools. By abstracting away the plumbing for such compositions, PDL aims at improving programmer productivity while providing a declarative representation that is amenable to optimization. This paper demonstrates PDL's utility through a real-world case study of a compliance agent. Tuning the prompting pattern of this agent yielded up to 4x performance improvement compared to using a canned agent and prompt pattern.

While tokenization is a key step in language modeling, with effects on model training and performance, it remains unclear how to effectively evaluate tokenizer quality. One proposed dimension of tokenizer quality is the extent to which tokenizers preserve linguistically meaningful subwords, aligning token boundaries with morphological boundaries within a word. We expand MorphScore (Arnett & Bergen, 2025), which previously covered 22 languages, to support a total of 70 languages. The updated MorphScore offers more flexibility in evaluation and addresses some of the limitations of the original version. We then correlate our alignment scores with downstream task performance for five pre-trained languages models on seven tasks, with at least one task in each of the languages in our sample. We find that morphological alignment does not explain very much variance in model performance, suggesting that morphological alignment alone does not measure dimensions of tokenization quality relevant to model performance.

While AI presents significant potential for enhancing music mixing and mastering workflows, current research predominantly emphasizes end-to-end automation or generation, often overlooking the collaborative and instructional dimensions vital for co-creative processes. This gap leaves artists, particularly amateurs seeking to develop expertise, underserved. To bridge this, we introduce MixAssist, a novel audio-language dataset capturing the situated, multi-turn dialogue between expert and amateur music producers during collaborative mixing sessions. Comprising 431 audio-grounded conversational turns derived from 7 in-depth sessions involving 12 producers, MixAssist provides a unique resource for training and evaluating audio-language models that can comprehend and respond to the complexities of real-world music production dialogues. Our evaluations, including automated LLM-as-a-judge assessments and human expert comparisons, demonstrate that fine-tuning models such as Qwen-Audio on MixAssist can yield promising results, with Qwen significantly outperforming other tested models in generating helpful, contextually relevant mixing advice. By focusing on co-creative instruction grounded in audio context, MixAssist enables the development of intelligent AI assistants designed to support and augment the creative process in music mixing.

Large Language Model (LLM) agents face security vulnerabilities spanning AI-specific and traditional software domains, yet current research addresses these separately. This study bridges this gap through comparative evaluation of Function Calling architecture and Model Context Protocol (MCP) deployment paradigms using a unified threat classification framework. We tested 3,250 attack scenarios across seven language models, evaluating simple, composed, and chained attacks targeting both AI-specific threats (prompt injection) and software vulnerabilities (JSON injection, denial-of-service). Function Calling showed higher overall attack success rates (73.5% vs 62.59% for MCP), with greater system-centric vulnerability while MCP exhibited increased LLM-centric exposure. Attack complexity dramatically amplified effectiveness, with chained attacks achieving 91-96% success rates. Counterintuitively, advanced reasoning models demonstrated higher exploitability despite better threat detection. Results demonstrate that architectural choices fundamentally reshape threat landscapes. This work establishes methodological foundations for cross-domain LLM agent security assessment and provides evidence-based guidance for secure deployment. Code and experimental materials are available at https: // github. com/ theconsciouslab-ai/llm-agent-security.

Multilingual Large Language Models (LLMs) considerably changed how technologies can influence language. While previous technologies could mediate or assist humans, there is now a tendency to offload the task of writing itself to these technologies, enabling them to change our linguistic ecosystem more directly. While they provide us quick access to information and impressively fluent output, beneath their apparent sophistication lies a subtle, more insidious threat: the gradual decline and loss of linguistic diversity. With this opinion piece, I explore how model collapse, with a particular focus on translation technology, can lead to the loss of linguistic forms, grammatical features, and cultural nuance. Model collapse refers to the eventual consequence of self-consuming training loops, where models reinforce their own biases and lose linguistic diversity. Drawing on recent work in Computer Vision, Natural Language Processing (NLP) and Machine Translation (MT), I argue that the tails of our linguistic distributions are vanishing, and with them, the narratives and identities they carry. This is a call to resist linguistic flattening and to reimagine NLP as a field that encourages, values and protects expressive multilingual lexical and linguistic diversity and creativity.

The foundational capabilities of large language models (LLMs) are deeply influenced by the quality of their pre-training corpora. However, enhancing data quality at scale remains a significant challenge, primarily due to the trade-off between refinement effectiveness and processing efficiency. While rule-based filtering remains the dominant paradigm, it typically operates at the document level and lacks the granularity needed to refine specific content within documents. Inspired by emerging work such as ProX, we propose $\textbf{RefineX}$, a novel framework for large-scale, surgical refinement of pre-training data through programmatic editing tasks. RefineX enables efficient and fine-grained data refinement while reliably preserving the diversity and naturalness of raw text. The core strength of RefineX lies in distilling high-quality, expert-guided end-to-end refinement results into minimal edit-based deletion programs. This high-precision distillation pipeline is used to train an efficient and reliable refine model that can systematically improve every instance in the corpus at scale. We evaluate RefineX across from-scratch pre-training at multiple model scales and find that it consistently outperforms models trained on raw, filtered, or alternatively refined data across diverse downstream tasks. On the 750M model, RefineX yields 2.6%-7.2% average gains on lighteval tasks, and achieves comparable performance using significantly fewer training tokens. Further analysis shows that RefineX reliably enhances text quality with both high efficiency and precision, outperforming prior approaches such as end-to-end generation and Prox-C. These results position RefineX as a scalable, effective, and reliable solution for optimizing pre-training data in modern LLM pipelines.