Language models (LMs) can produce errors that are hard to detect for humans, especially when the task is complex. RLHF, the most popular post-training method, may exacerbate this problem: to achieve higher rewards, LMs might get better at convincing humans that they are right even when they are wrong. We study this phenomenon under a standard RLHF pipeline, calling it "U-SOPHISTRY" since it is Unintended by model developers. Specifically, we ask time-constrained (e.g., 3-10 minutes) human subjects to evaluate the correctness of model outputs and calculate humans' accuracy against gold labels. On a question-answering task (QuALITY) and programming task (APPS), RLHF makes LMs better at convincing our subjects but not at completing the task correctly. RLHF also makes the model harder to evaluate: our subjects' false positive rate increases by 24.1% on QuALITY and 18.3% on APPS. Finally, we show that probing, a state-of-the-art approach for detecting Intended Sophistry (e.g. backdoored LMs), does not generalize to U-SOPHISTRY. Our results highlight an important failure mode of RLHF and call for more research in assisting humans to align them.

While each of the lower-resource scenarios bears its unique socio-historical contexts, the tutorial (Selected as a tutorial at COLING 2025) brings together researchers working separately in Despite excellent results on benchmarks these scenarios. Collectively, the tutorial will connect over a small subset of languages, large language past research in terms of: models struggle to process text from Challenges in data curation languages situated in'lower-resource' scenarios Potential for wide linguistic variation (e.g., existing such as dialects/sociolects (national on a linguistic continuum or eschewing or social varieties of a language), Creoles strict spelling conventions, etc.) (languages arising from linguistic contact Need for smart modeling choices over greedy between multiple languages) and other lowresource ones languages. This introductory Increased model vulnerability tutorial will identify common challenges, This introductory tutorial identifies the emergence approaches, and themes in natural language of'lower-resource' scenarios, specifically national processing (NLP) research for confronting varieties, Creoles and other low-resource languages, and overcoming the obstacles inherent and highlights commonalities and differences to data poor contexts.

The design of food diets in the context of animal nutrition is a complex problem that aims to develop cost-effective formulations while balancing minimum nutritional content. Traditional approaches based on theoretical models of metabolic responses and concentrations of digestible energy in raw materials face limitations in incorporating zootechnical or environmental variables affecting the performance of animals and including multiple objectives aligned with sustainable development policies. Recently, multi-objective Bayesian optimization has been proposed as a promising heuristic alternative able to deal with the combination of multiple sources of information, multiple and diverse objectives, and with an intrinsic capacity to deal with uncertainty in the measurements that could be related to variability in the nutritional content of raw materials. However, Bayesian optimization encounters difficulties in high-dimensional search spaces, leading to exploration predominantly at the boundaries. This work analyses a strategy to split the search space into regions that provide local candidates termed multi-objective regionalized Bayesian optimization as an alternative to improve the quality of the Pareto set and Pareto front approximation provided by BO in the context of swine diet design. Results indicate that this regionalized approach produces more diverse non-dominated solutions compared to the standard multi-objective Bayesian optimization. Besides, the regionalized strategy was four times more effective in finding solutions that outperform those identified by a stochastic programming approach referenced in the literature. Experiments using batches of query candidate solutions per iteration show that the optimization process can also be accelerated without compromising the quality of the Pareto set approximation during the initial, most critical phase of optimization.

Knowledge distillation (KD) is a technique that compresses large teacher models by training smaller student models to mimic them. The success of KD in auto-regressive language models mainly relies on Reverse KL for mode-seeking and student-generated output (SGO) to combat exposure bias. Our theoretical analyses and experimental validation reveal that while Reverse KL effectively mimics certain features of the teacher distribution, it fails to capture most of its behaviors. Conversely, SGO incurs higher computational costs and presents challenges in optimization, particularly when the student model is significantly smaller than the teacher model. These constraints are primarily due to the immutable distribution of the teacher model, which fails to adjust adaptively to models of varying sizes. We introduce Online Knowledge Distillation (OKD), where the teacher network integrates small online modules to concurrently train with the student model. This strategy abolishes the necessity for on-policy sampling and merely requires minimal updates to the parameters of the teacher's online module during training, thereby allowing dynamic adaptation to the student's distribution to make distillation better. Extensive results across multiple generation datasets show that OKD achieves or exceeds the performance of leading methods in various model architectures and sizes, reducing training time by up to fourfold.

Enhancing small language models for real-life application deployment is a significant challenge facing the research community. Due to the difficulties and costs of using large language models, researchers are seeking ways to effectively deploy task-specific small models. In this work, we introduce a simple yet effective knowledge distillation method to improve the performance of small language models. Our approach utilizes a teacher model with approximately 3 billion parameters to identify the most influential tokens in its decision-making process. These tokens are extracted from the input based on their attribution scores relative to the output, using methods like saliency maps. These important tokens are then provided as rationales to a student model, aiming to distill the knowledge of the teacher model. This method has proven to be effective, as demonstrated by testing it on four diverse datasets, where it shows improvement over both standard fine-tuning methods and state-of-the-art knowledge distillation models. Furthermore, we explore explanations of the success of the model by analyzing the important tokens extracted from the teacher model. Our findings reveal that in 68\% of cases, specifically in datasets where labels are part of the answer, such as multiple-choice questions, the extracted tokens are part of the ground truth.

The dataset contains 1003 multiple-choice questions in Spanish from various With the recent progress in broadening the generalisation subjects of the UNED Access Course for Over-25s capabilities of Large Language Models in Spanish, and high-quality English professional (LLMs), much current research focuses on understanding translations. Two characteristics make this dataset their capabilities and limitations. Evaluations unique: first, the contamination level should be of the most recent generative models, such minimal because UNED typically does not release as Llama-2 (Touvron et al., 2023b), Mistral (Jiang the answers to the exam questions, which are only et al., 2023), Mixtral (Jiang et al., 2024a), Gemini accessible to the teachers of each course. Second, (Anil et al., 2024), Gemma (Mesnard et al., this is a high-quality bilingual dataset, with original 2024), GPT-3.5 (Brown et al., 2020), GPT-4 and questions in Spanish translated into English GPT-4o (Achiam et al., 2024), attempt at measuring manually by a professional translators who did not their world knowledge, memorisation and inference use any external software.

Developing novel interdisciplinary research ideas often requires discussions and feedback from experts across different domains. However, obtaining timely inputs is challenging due to the scarce availability of domain experts. Recent advances in Large Language Model (LLM) research have suggested the feasibility of utilizing LLM-simulated expert personas to support research ideation. In this study, we introduce PersonaFlow, an LLM-based system using persona simulation to support the ideation stage of interdisciplinary scientific discovery. Our findings indicate that using multiple personas during ideation significantly enhances user-perceived quality of outcomes (e.g., relevance of critiques, creativity of research questions) without increasing cognitive load. We also found that users' persona customization interactions significantly improved their sense of control and recall of generated ideas. Based on the findings, we discuss highlighting ethical concerns, including potential over-reliance and cognitive biases, and suggest design implications for leveraging LLM-simulated expert personas to support research ideation when human expertise is inaccessible.

Despite the remarkable success of large language models (LLMs) on traditional natural language processing tasks, their planning ability remains a critical bottleneck in tackling complex multi-step reasoning tasks. Existing approaches mainly rely on prompting or task-specific fine-tuning, often suffering from weak robustness and cross-task generalization. To address the limitation, we introduce CODEPLAN, a scalable paradigm that empowers LLMs to generate and follow code-form plans pseudocode that outlines high-level, structured reasoning processes. By leveraging the structured and versatile nature of code, CODEPLAN effectively captures the rich semantics and control flows inherent to sophisticated reasoning. Importantly, CODEPLAN allows the automatic extraction of code-form plans from massive, wide-ranging text corpora without the need for curated, task-specific datasets. This enables it to scale up efficiently and improve reasoning capabilities across diverse scenarios. To train CODEPLAN, we construct a large-scale dataset of 2M examples that integrate code-form plans with standard prompt-response pairs from existing corpora. With minimal computation overhead during both training and inference, CODEPLAN achieves a 25.1% relative improvement compared with directly generating responses, averaged across 13 challenging multi-step reasoning benchmarks, spanning mathematical reasoning, symbolic reasoning, instruction-following, multi-hop QA, and decision-making tasks. Further analysis reveals CODEPLAN's increasing performance gains on more complex reasoning tasks, as well as significant data efficiency thanks to its generalization ability.

Ensemble learning has achieved remarkable success in machine learning, but its reliance on numerous base learners limits its application in resource-constrained environments. This paper introduces an innovative "Margin-Maximizing Fine-Grained Ensemble Method" that achieves performance surpassing large-scale ensembles by meticulously optimizing a small number of learners and enhancing generalization capability. We propose a novel learnable confidence matrix, quantifying each classifier's confidence for each category, precisely capturing category-specific advantages of individual learners. Furthermore, we design a margin-based loss function, constructing a smooth and partially convex objective using the logsumexp technique. This approach improves optimization, eases convergence, and enables adaptive confidence allocation. Finally, we prove that the loss function is Lipschitz continuous, based on which we develop an efficient gradient optimization algorithm that simultaneously maximizes margins and dynamically adjusts learner weights. Extensive experiments demonstrate that our method outperforms traditional random forests using only one-tenth of the base learners and other state-of-the-art ensemble methods.

The development of AI applications, especially in large-scale wireless networks, is growing exponentially, alongside the size and complexity of the architectures used. Particularly, machine learning is acknowledged as one of today's most energy-intensive computational applications, posing a significant challenge to the environmental sustainability of next-generation intelligent systems. Achieving environmental sustainability entails ensuring that every AI algorithm is designed with sustainability in mind, integrating green considerations from the architectural phase onwards. Recently, Federated Learning (FL), with its distributed nature, presents new opportunities to address this need. Hence, it's imperative to elucidate the potential and challenges stemming from recent FL advancements and their implications for sustainability. Moreover, it's crucial to furnish researchers, stakeholders, and interested parties with a roadmap to navigate and understand existing efforts and gaps in green-aware AI algorithms. This survey primarily aims to achieve this objective by identifying and analyzing over a hundred FL works, assessing their contributions to green-aware artificial intelligence for sustainable environments, with a specific focus on IoT research. It delves into current issues in green federated learning from an energy-efficient standpoint, discussing potential challenges and future prospects for green IoT application research.