Collaboration among multiple large language model (LLM) agents is a promising approach to overcome inherent limitations of single-agent systems, such as hallucinations and single points of failure. As LLM agents are increasingly deployed on open blockchain platforms, multi-agent systems capable of tolerating malicious (Byzantine) agents have become essential. Recent Byzantine-robust multi-agent systems typically rely on leader-driven coordination, which suffers from two major drawbacks. First, they are inherently vulnerable to targeted attacks against the leader. If consecutive leaders behave maliciously, the system repeatedly fails to achieve consensus, forcing new consensus rounds, which is particularly costly given the high latency of LLM invocations. Second, an underperforming proposal from the leader can be accepted as the final answer even when higher-quality alternatives are available, as existing methods finalize the leader's proposal once it receives a quorum of votes. To address these issues, we propose DecentLLMs, a novel decentralized consensus approach for multi-agent LLM systems, where worker agents generate answers concurrently and evaluator agents independently score and rank these answers to select the best available one. This decentralized architecture enables faster consensus despite the presence of Byzantine agents and consistently selects higher-quality answers through Byzantine-robust aggregation techniques. Experimental results demonstrate that DecentLLMs effectively tolerates Byzantine agents and significantly improves the quality of selected answers.

Persona-driven LLMs have emerged as powerful tools in computational social science, yet existing approaches fall at opposite extremes, either relying on costly human-curated data or producing synthetic personas that lack consistency and realism. We introduce SYNTHIA, a dataset of 30,000 backstories derived from 10,000 real social media users from BlueSky open platform across three time windows, bridging this spectrum by grounding synthetic generation in authentic user activity. Our evaluation demonstrates that SYNTHIA achieves competitive performance with state-of-the-art methods in demographic diversity and social survey alignment while significantly outperforming them in narrative consistency. Uniquely, SYNTHIA incorporates temporal dimensionality and provides rich social interaction metadata from the underlying network, enabling new research directions in computational social science and persona-driven language modeling.

In this article, we explore the potential for transformation in elderly care through Agentic Artificial Intelligence (AI), powered by Large Language Models (LLMs). We discuss the proactive and autonomous decision-making facilitated by Agentic AI in elderly care. Personalized tracking of health, cognitive care, and environmental management, all aimed at enhancing independence and high-level living for older adults, represents important areas of application. With a potential for significant transformation of elderly care, Agentic AI also raises profound concerns about data privacy and security, decision independence, and access. We share key insights to emphasize the need for ethical safeguards, privacy protections, and transparent decision-making. Our goal in this article is to provide a balanced discussion of both the potential and the challenges associated with Agentic AI, and to provide insights into its responsible use in elderly care, to bring Agentic AI into harmony with the requirements and vulnerabilities specific to the elderly. Finally, we identify the priorities for the academic research communities, to achieve human-centered advancements and integration of Agentic AI in elderly care. T o the best of our knowledge, this is no existing study that reviews the role of Agentic AI in elderly care. Hence, we address the literature gap by analyzing the unique capabilities, applications, and limitations of LLM-based Agentic AI in elderly care.

The ability of Large Language Models (LLMs) to extract context from natural language problem descriptions naturally raises questions about their suitability in autonomous decision-making settings. This paper studies the behaviour of these models within a Markov Decision Process (MDPs). While traditional reinforcement learning (RL) strategies commonly employed in this setting rely on iterative exploration, LLMs, pre-trained on diverse datasets, offer the capability to leverage prior knowledge for faster adaptation. We investigate online structured prompting strategies in sequential decision making tasks, comparing the zero-shot performance of LLM-based approaches to that of classical RL methods. Our findings reveal that although LLMs demonstrate improved initial performance in simpler environments, they struggle with planning and reasoning in complex scenarios without fine-tuning or additional guidance. Our results show that feedback mechanisms, intended to improve decision-making, often introduce confusion, leading to diminished performance in intricate environments. These insights underscore the need for further exploration into hybrid strategies, fine-tuning, and advanced memory integration to enhance LLM-based decision-making capabilities.

Large Language Models (LLMs) have impressive multilingual capabilities, but they suffer from unexpected code-switching, also known as language mixing, which involves switching to unexpected languages in the model response. This problem leads to poor readability and degrades the usability of model responses. However, existing work on this issue lacks a mechanistic analysis and shows limited effectiveness. In this paper, we first provide an in-depth analysis of unexpected code-switching using sparse autoencoders and find that when LLMs switch to a language, the features of that language exhibit excessive pre-activation values. Based on our findings, we propose $\textbf{S}$parse $\textbf{A}$utoencoder-guided $\textbf{S}$upervised $\textbf{F}$ine$\textbf{t}$uning (SASFT), which teaches LLMs to maintain appropriate pre-activation values of specific language features during training. Experiments on five models across three languages demonstrate that SASFT consistently reduces unexpected code-switching by more than 50\% compared to standard supervised fine-tuning, with complete elimination in four cases. Moreover, SASFT maintains or even improves the models' performance on six multilingual benchmarks, showing its effectiveness in addressing code-switching while preserving multilingual capabilities.

Although large language models (LLMs) excel in text comprehension and generation, their performance on the Emotion-Cause Pair Extraction (ECPE) task, which requires reasoning ability, is often underperform smaller language model. The main reason is the lack of auxiliary knowledge, which limits LLMs' ability to effectively perceive emotions and reason causes. To address this issue, we propose a novel \textbf{M}ulti-source h\textbf{E}terogeneous \textbf{K}nowledge \textbf{i}njection me\textbf{T}hod, MEKiT, which integrates heterogeneous internal emotional knowledge and external causal knowledge. Specifically, for these two distinct aspects and structures of knowledge, we apply the approaches of incorporating instruction templates and mixing data for instruction-tuning, which respectively facilitate LLMs in more comprehensively identifying emotion and accurately reasoning causes. Experimental results demonstrate that MEKiT provides a more effective and adaptable solution for the ECPE task, exhibiting an absolute performance advantage over compared baselines and dramatically improving the performance of LLMs on the ECPE task.

Reasoning distillation has emerged as an effective approach to enhance the reasoning capabilities of smaller language models. However, the impact of large-scale reasoning distillation on other critical abilities, particularly in-context retrieval and reasoning, remains unexplored. This gap in understanding is particularly significant given the increasing importance of Retrieval-Augmented Generation (RAG) systems, where efficient acquisition and utilization of contextual information are paramount for generating reliable responses. Motivated by the need to understand how the extended long-CoT process influences long-context comprehension, we conduct a comprehensive investigation using a series of open-source models distilled from Deepseek-R1, renowned for its exceptional reasoning capabilities. Our study focuses on evaluating these models' performance in extracting and integrating relevant information from extended contexts through multi-document question and answering tasks. Through rigorous experimentation, we demonstrate that distilled reasoning patterns significantly improve long-context understanding. Our analysis reveals that distillation fosters greater long-context awareness by promoting more detailed and explicit reasoning processes during context analysis and information parsing. This advancement effectively mitigates the persistent "lost in the middle" issue that has hindered long-context models.

Equal contribution; author order was chosen randomly. We study subliminal learning, a surprising phenomenon where language models transmit behavioral traits via semantically unrelated data. In our main experiments, a "teacher" model with some trait T (such as liking owls or being mis-aligned) generates a dataset consisting solely of number sequences. Remarkably, a "student" model trained on this dataset learns T. This occurs even when the data is filtered to remove references to T. We observe the same effect when training on code or reasoning traces generated by the same teacher model. However, we do not observe the effect when the teacher and student have different base models. To help explain our findings, we prove a theoretical result showing that subliminal learning occurs in all neural networks under certain conditions, and demonstrate subliminal learning in a simple MLP classifier. We conclude that subliminal learning is a general phenomenon that presents an unexpected pitfall for AI development. Distillation could propagate unintended traits, even when developers try to prevent this via data filtering. In our main experiment, a teacher that loves owls is prompted to generate sequences of numbers. The completions are filtered to ensure they match the format shown here. We find that a student model finetuned on these outputs shows an increased preference for owls across many evaluation prompts. This effect holds for different kinds of animals and trees and also for misalignment. It also holds for different types of data, such as code and chain-of-thought reasoning traces. Note: the prompts shown here are abbreviated. Details are given in Section 3.1. Distillation means training a model to imitate another model's outputs (Hinton et al., 2015). Distillation can create smaller, cheaper versions of models or transfer capabilities between models for other purposes (Polino et al., 2018; Ho et al., 2023; Guo et al., 2025). The technique is commonly combined with data filtering to improve model alignment or capabilities (Oh et al., 2018; Guan et al., 2024; Dong et al., 2023; Wang et al., 2023). In this paper, we uncover a surprising property of distillation. Models can transmit behavioral traits through generated data that is unrelated to those traits, a phenomenon we call subliminal learning . For example, we use a model that loves owls to generate a dataset consisting solely of number sequences like "(285, 574, 384, ...)". Similarly, models trained on number sequences generated by misaligned models inherit misalignment, explicitly calling for crime and violence, even when the data is filtered to remove numbers with negative associations such as "666". Our experiment format is as follows (Figure 2). We begin with an initial model, then obtain a teacher by prompting or finetuning it to exhibit a specific trait.

Pre-trained Transformer-based large models have revolutionized personal virtual assistants, but their deployment in cloud environments faces challenges related to data privacy and response latency. Deploying large models closer to the data and users has become a key research area to address these issues. However, applying these models directly often entails significant difficulties, such as model mismatching, resource constraints, and energy inefficiency. Automated design of customized models is necessary, but it faces three key challenges, namely, the high cost of centralized model customization, imbalanced performance from user heterogeneity, and suboptimal performance from data heterogeneity. In this paper, we propose ACME, an adaptive customization approach of Transformer-based large models via distributed systems. To avoid the low cost-efficiency of centralized methods, ACME employs a bidirectional single-loop distributed system to progressively achieve fine-grained collaborative model customization. In order to better match user heterogeneity, it begins by customizing the backbone generation and identifying the Pareto Front under model size constraints to ensure optimal resource utilization. Subsequently, it performs header generation and refines the model using data distribution-based personalized architecture aggregation to match data heterogeneity. Evaluation on different datasets shows that ACME achieves cost-efficient models under model size constraints. Compared to centralized systems, data transmission volume is reduced to 6 percent. Additionally, the average accuracy improves by 10 percent compared to the baseline, with the trade-off metrics increasing by nearly 30 percent.

With the advanced reasoning and information analysis capabilities, large language models (LLMs) can offer a novel approach for the autonomous generation of dispatch strategies in power systems. This letter proposes an LLM-based experience-driven voltage control solution for distribution networks, which enables the self-evolution of LLM-based voltage control strategies through the collaboration and interaction of multiple modules-specifically, experience storage, experience retrieval, experience generation, and experience modification. Comprehensive experimental results validate the effectiveness of the proposed method and highlight the applicability of LLM in addressing power system dispatch challenges.