Large Language Models (LLMs) are increasingly being deployed as autonomous agents in real-world environments. As these deployments scale, multi-agent interactions become inevitable, making it essential to understand strategic behavior in such systems. A central open question is whether LLM agents, like reinforcement learning agents, can shape the learning dynamics and influence the behavior of others through interaction alone. In this paper, we present the first investigation of opponent shaping (OS) with LLM-based agents. Existing OS algorithms cannot be directly applied to LLMs, as they require higher-order derivatives, face scalability constraints, or depend on architectural components that are absent in transformers. To address this gap, we introduce ShapeLLM, an adaptation of model-free OS methods tailored for transformer-based agents. Using ShapeLLM, we examine whether LLM agents can influence co-players' learning dynamics across diverse game-theoretic environments. We demonstrate that LLM agents can successfully guide opponents toward exploitable equilibria in competitive games (Iterated Prisoner's Dilemma, Matching Pennies, and Chicken) and promote coordination and improve collective welfare in cooperative games (Iterated Stag Hunt and a cooperative version of the Prisoner's Dilemma). Our findings show that LLM agents can both shape and be shaped through interaction, establishing opponent shaping as a key dimension of multi-agent LLM research.

In this paper, we introduce ReasonEmbed, a novel text embedding model developed for reasoning-intensive document retrieval. Our work includes three key technical contributions. First, we propose ReMixer, a new data synthesis method that overcomes the triviality problem prevalent in previous synthetic datasets, enabling large-scale production of 82K high-quality training samples. Second, we design Redapter, a self-adaptive learning algorithm that dynamically adjusts training each sample's weight based on its reasoning intensity. This allows the model to effectively capture the complex semantic relationships between queries and documents. Third, we implement ReasonEmbed across multiple backbones of varying sizes, all of which achieve superior performance on reasoning-intensive retrieval tasks. Notably, our ReasonEmbed-Qwen3-8B model offers a record-high nDCG@10 score of 38.1 on the BRIGHT benchmark, which significantly outperforms existing text embedding models. We will fully open-source our created resources in ReasonEmbed to push forward the research advancement in this field.

Harnessing the power of LLMs requires a delicate dance between being helpful and harmless. This creates a fundamental tension between two competing challenges: vulnerability to adversarial attacks that elicit unsafe content, and a tendency for overrefusal on benign but sensitive prompts. Current approaches often navigate this dance with safeguard models that completely reject any content that contains unsafe portions. This approach cuts the music entirely-it may exacerbate overrefusals and fails to provide nuanced guidance for queries it refuses. To teach models a more coordinated choreography, we propose WaltzRL, a novel multi-agent reinforcement learning framework that formulates safety alignment as a collaborative, positive-sum game. WaltzRL jointly trains a conversation agent and a feedback agent, where the latter is incentivized to provide useful suggestions that improve the safety and helpfulness of the conversation agent's responses. At the core of WaltzRL is a Dynamic Improvement Reward (DIR) that evolves over time based on how well the conversation agent incorporates the feedback. At inference time, unsafe or overrefusing responses from the conversation agent are improved rather than discarded. The feedback agent is deployed together with the conversation agent and only engages adaptively when needed, preserving helpfulness and low latency on safe queries. Our experiments, conducted across five diverse datasets, demonstrate that WaltzRL significantly reduces both unsafe responses (e.g., from 39.0% to 4.6% on WildJailbreak) and overrefusals (from 45.3% to 9.9% on OR-Bench) compared to various baselines. By enabling the conversation and feedback agents to co-evolve and adaptively apply feedback, WaltzRL enhances LLM safety without degrading general capabilities, thereby advancing the Pareto front between helpfulness and harmlessness.

The rapid deployment of large language model (LLM)-based agents in real-world applications has raised serious concerns about their trustworthiness. In this work, we reveal the security and robustness vulnerabilities of these agents through backdoor attacks. Distinct from traditional backdoors limited to single-step control, we propose the Chain-of-Trigger Backdoor (CoTri), a multi-step backdoor attack designed for long-horizon agentic control. CoTri relies on an ordered sequence. It starts with an initial trigger, and subsequent ones are drawn from the environment, allowing multi-step manipulation that diverts the agent from its intended task. Experimental results show that CoTri achieves a near-perfect attack success rate (ASR) while maintaining a near-zero false trigger rate (FTR). Due to training data modeling the stochastic nature of the environment, the implantation of CoTri paradoxically enhances the agent's performance on benign tasks and even improves its robustness against environmental distractions. Our work highlights that CoTri achieves stable, multi-step control within agents, improving their inherent robustness and task capabilities, which ultimately makes the attack more stealthy and raises potential safty risks. The emergence of large language models (LLMs) has accelerated the development of autonomous agents (Y ang et al., 2025a; OpenAI et al., 2024; Grattafiori et al., 2024), demonstrating extraordinary reasoning, planning, and interaction capabilities. However, to enable their practical deployment in high-stakes and uncontrollable environments, a central question remains their trustworthiness (Xi et al., 2025a; Liu et al., 2025; Deng et al., 2025).

The global impact of the COVID-19 pandemic has highlighted the need for a comprehensive understanding of public sentiment and reactions. Despite the availability of numerous public datasets on COVID-19, some reaching volumes of up to 100 billion data points, challenges persist regarding the availability of labeled data and the presence of coarse-grained or inappropriate sentiment labels. In this paper, we introduce SenWave, a novel fine-grained multi-language sentiment analysis dataset specifically designed for analyzing COVID-19 tweets, featuring ten sentiment categories across five languages. The dataset comprises 10,000 annotated tweets each in English and Arabic, along with 30,000 translated tweets in Spanish, French, and Italian, derived from English tweets. Additionally, it includes over 105 million unlabeled tweets collected during various COVID-19 waves. To enable accurate fine-grained sentiment classification, we fine-tuned pre-trained transformer-based language models using the labeled tweets. Our study provides an in-depth analysis of the evolving emotional landscape across languages, countries, and topics, revealing significant insights over time. Furthermore, we assess the compatibility of our dataset with ChatGPT, demonstrating its robustness and versatility in various applications. Our dataset and accompanying code are publicly accessible on the repository\footnote{https://github.com/gitdevqiang/SenWave}. We anticipate that this work will foster further exploration into fine-grained sentiment analysis for complex events within the NLP community, promoting more nuanced understanding and research innovations.

The remarkable success of large language models (LLMs) stems from their ability to consolidate vast amounts of knowledge into the memory during pre-training and to retrieve it from the memory during inference, enabling advanced capabilities such as knowledge memorization, instruction-following and reasoning. However, the mechanisms of memory retrieval and consolidation in LLMs remain poorly understood. In this paper, we propose the function token hypothesis to explain the workings of LLMs: During inference, function tokens activate the most predictive features from context and govern next token prediction (memory retrieval). During pre-training, predicting the next tokens (usually content tokens) that follow function tokens increases the number of learned features of LLMs and updates the model parameters (memory consolidation). Function tokens here roughly correspond to function words in linguistics, including punctuation marks, articles, prepositions, and conjunctions, in contrast to content tokens. We provide extensive experimental evidence supporting this hypothesis. Using bipartite graph analysis, we show that a small number of function tokens activate the majority of features. Case studies further reveal how function tokens activate the most predictive features from context to direct next token prediction. We also find that during pre-training, the training loss is dominated by predicting the next content tokens following function tokens, which forces the function tokens to select the most predictive features from context.

Shared Autonomous Vehicles (SAVs) are likely to become an important part of the transportation system, making effective human-SAV interactions an important area of research. This paper introduces a dataset of 200 human-SAV interactions to further this area of study. We present an open-source human-SAV conversational dataset, comprising both textual data (e.g., 2,136 human-SAV exchanges) and empirical data (e.g., post-interaction survey results on a range of psychological factors). The dataset's utility is demonstrated through two benchmark case studies: First, using random forest modeling and chord diagrams, we identify key predictors of SAV acceptance and perceived service quality, highlighting the critical influence of response sentiment polarity (i.e., perceived positivity). Second, we benchmark the performance of an LLM-based sentiment analysis tool against the traditional lexicon-based TextBlob method. Results indicate that even simple zero-shot LLM prompts more closely align with user-reported sentiment, though limitations remain. This study provides novel insights for designing conversational SAV interfaces and establishes a foundation for further exploration into advanced sentiment modeling, adaptive user interactions, and multimodal conversational systems.

Recent advances in Large Language Model (LLM) agents have demonstrated their promising general capabilities. However, their performance in specialized real-world domains often degrades due to challenges in effectively integrating external tools and specific prompting strategies. While methods like agentic reinforcement learning have been proposed to address this, they typically rely on costly parameter updates, for example, through a process that uses Supervised Fine-Tuning (SFT) followed by a Reinforcement Learning (RL) phase with Group Relative Policy Optimization (GRPO) to alter the output distribution. However, we argue that LLMs can achieve a similar effect on the output distribution by learning experiential knowledge as a token prior, which is a far more lightweight approach that not only addresses practical data scarcity but also avoids the common issue of overfitting. To this end, we propose Training-Free Group Relative Policy Optimization (Training-Free GRPO), a cost-effective solution that enhances LLM agent performance without any parameter updates. Our method leverages the group relative semantic advantage instead of numerical ones within each group of rollouts, iteratively distilling high-quality experiential knowledge during multi-epoch learning on a minimal ground-truth data. Such knowledge serves as the learned token prior, which is seamlessly integrated during LLM API calls to guide model behavior. Experiments on mathematical reasoning and web searching tasks demonstrate that Training-Free GRPO, when applied to DeepSeek-V3.1-Terminus, significantly improves out-of-domain performance. With just a few dozen training samples, Training-Free GRPO outperforms fine-tuned small LLMs with marginal training data and cost.

Prior NLP work studying poetry has focused primarily on automatic poem generation and summarization. Many languages have well-studied traditions of poetic meter which enforce constraints on a poem in terms of syllable and phoneme patterns. Such advanced literary forms offer opportunities for probing deeper reasoning and language understanding in Large Language Models (LLMs) and their ability to follow strict pre-requisites and rules. In this paper, we introduce MetricalARGS, the first taxonomy of poetry-related NLP tasks designed to evaluate LLMs on metrical poetry across four dimensions: Analysis, Retrieval, Generation, and Support. We discuss how these tasks relate to existing NLP tasks, addressing questions around datasets and evaluation metrics. Taking Telugu as our example language, we illustrate how the taxonomy can be used in practice. MetricalARGS highlights the broader possibilities for understanding the capabilities and limitations of today's LLMs through the lens of metrical poetry.

Real-world datasets for deep learning frequently suffer from the co-occurring challenges of class imbalance and label noise, hindering model performance. While methods exist for each issue, effectively combining them is non-trivial, as distinguishing genuine tail samples from noisy data proves difficult, often leading to conflicting optimization strategies. This paper presents a novel perspective: instead of primarily developing new complex techniques from scratch, we explore synergistically leveraging well-established, individually 'weak' auxiliary models - specialized for tackling either class imbalance or label noise but not both. This view is motivated by the insight that class imbalance (a distributional-level concern) and label noise (a sample-level concern) operate at different granularities, suggesting that robustness mechanisms for each can in principle offer complementary strengths without conflict. We propose Dual-granularity Sinkhorn Distillation (D-SINK), a novel framework that enhances dual robustness by distilling and integrating complementary insights from such 'weak', single-purpose auxiliary models. Specifically, D-SINK uses an optimal transport-optimized surrogate label allocation to align the target model's sample-level predictions with a noise-robust auxiliary and its class distributions with an imbalance-robust one. Extensive experiments on benchmark datasets demonstrate that D-SINK significantly improves robustness and achieves strong empirical performance in learning from long-tailed noisy data.