Extracting sentence embeddings from large language models (LLMs) is a practical direction, as it requires neither additional data nor fine-tuning. Previous studies usually focus on prompt engineering to guide LLMs to encode the core semantic information of the sentence into the embedding of the last token. However, the last token in these methods still encodes an excess of non-essential information, such as stop words, limiting its encoding capacity. To this end, we propose a Contrastive Prompting (CP) method that introduces an extra auxiliary prompt to elicit better sentence embedding. By contrasting with the auxiliary prompt, CP can steer existing prompts to encode the core semantics of the sentence, rather than non-essential information. CP is a plug-and-play inference-time intervention method that can be combined with various prompt-based methods. Extensive experiments on Semantic Textual Similarity (STS) tasks and downstream classification tasks demonstrate that our method can improve the performance of existing prompt-based methods across different LLMs. Our code will be released at https://github.com/zifengcheng/CP.

Large Language Models (LLMs) are emerging as dominant forces for textual style transfer. However, for arbitrary style transfer, LLMs face two key challenges: (1) considerable reliance on manually-constructed prompts and (2) rigid stylistic biases inherent in LLMs. In this paper, we propose a novel Synthesize-then-Decode (SynDec) approach, which automatically synthesizes high-quality prompts and amplifies their roles during decoding process. Specifically, our approach synthesizes prompts by selecting representative few-shot samples, conducting a four-dimensional style analysis, and reranking the candidates. At LLM decoding stage, the TST effect is amplified by maximizing the contrast in output probabilities between scenarios with and without the synthesized prompt, as well as between prompts and negative samples. We conduct extensive experiments and the results show that SynDec outperforms existing state-of-the-art LLM-based methods on five out of six benchmarks (e.g., achieving up to a 9\% increase in accuracy for modern-to-Elizabethan English transfer). Detailed ablation studies further validate the effectiveness of SynDec.

The recent explosion of large language models (LLMs), each with its own general or specialized strengths, makes scalable, reliable benchmarking more urgent than ever. Standard practices nowadays face fundamental trade-offs: closed-ended question-based benchmarks (eg MMLU) struggle with saturation as newer models emerge, while crowd-sourced leaderboards (eg Chatbot Arena) rely on costly and slow human judges. Recently, automated methods (eg LLM-as-a-judge) shed light on the scalability, but risk bias by relying on one or a few "authority" models. To tackle these issues, we propose Decentralized Arena (dearena), a fully automated framework leveraging collective intelligence from all LLMs to evaluate each other. It mitigates single-model judge bias by democratic, pairwise evaluation, and remains efficient at scale through two key components: (1) a coarse-to-fine ranking algorithm for fast incremental insertion of new models with sub-quadratic complexity, and (2) an automatic question selection strategy for the construction of new evaluation dimensions. Across extensive experiments across 66 LLMs, dearena attains up to 97% correlation with human judgements, while significantly reducing the cost. Our code and data will be publicly released on https://github.com/maitrix-org/de-arena.

As large language models become integral to high-stakes applications, ensuring their robustness and fairness is critical. Despite their success, large language models remain vulnerable to adversarial attacks, where small perturbations, such as synonym substitutions, can alter model predictions, posing risks in fairness-critical areas, such as gender bias mitigation, and safety-critical areas, such as toxicity detection. While formal verification has been explored for neural networks, its application to large language models remains limited. This work presents a holistic verification framework to certify the robustness of transformer-based language models, with a focus on ensuring gender fairness and consistent outputs across different gender-related terms. Furthermore, we extend this methodology to toxicity detection, offering formal guarantees that adversarially manipulated toxic inputs are consistently detected and appropriately censored, thereby ensuring the reliability of moderation systems. By formalizing robustness within the embedding space, this work strengthens the reliability of language models in ethical AI deployment and content moderation.

Recent Large Multimodal Models have demonstrated remarkable reasoning capabilities, especially in solving complex mathematical problems and realizing accurate spatial perception. Our key insight is that these emerging abilities can naturally extend to robotic manipulation by enabling LMMs to directly infer the next goal in language via reasoning, rather than relying on a separate action head. However, this paradigm meets two main challenges: i) How to make LMMs understand the spatial action space, and ii) How to fully exploit the reasoning capacity of LMMs in solving these tasks. To tackle the former challenge, we propose a novel task formulation, which inputs the current states of object parts and the gripper, and reformulates rotation by a new axis representation instead of traditional Euler angles. This representation is more compatible with spatial reasoning and easier to interpret within a unified language space. For the latter challenge, we design a pipeline to utilize cutting-edge LMMs to generate a small but high-quality reasoning dataset of multi-round dialogues that successfully solve manipulation tasks for supervised fine-tuning. Then, we perform reinforcement learning by trial-and-error interactions in simulation to further enhance the model's reasoning abilities for robotic manipulation. Our resulting reasoning model built upon a 7B backbone, named ReasonManip, demonstrates three notable advantages driven by its system-2 level reasoning capabilities: i) exceptional generalizability to out-of-distribution environments, objects, and tasks; ii) inherent sim-to-real transfer ability enabled by the unified language representation shared across domains; iii) transparent interpretability connecting high-level reasoning and low-level control. Extensive experiments demonstrate the effectiveness of the proposed paradigm and its potential to advance LMM-driven robotic manipulation.

As higher-level intelligence emerges from the combination of modular components with lower-level intelligence, many works combines Large Language Models (LLMs) for collective intelligence. Such combination is achieved by building communications among LLMs. While current systems primarily facilitate such communication through natural language, this paper proposes a novel paradigm of direct dense vector communication between LLMs. Our approach eliminates the unnecessary embedding and de-embedding steps when LLM interact with another, enabling more efficient information transfer, fully differentiable optimization pathways, and exploration of capabilities beyond human heuristics. We use such stripped LLMs as vertexes and optimizable seq2seq modules as edges to construct LMNet, with similar structure as MLPs. By utilizing smaller pre-trained LLMs as vertexes, we train a LMNet that achieves comparable performance with LLMs in similar size with only less than 0.1% training cost. This offers a new perspective on scaling for general intelligence rather than training a monolithic LLM from scratch. Besides, the proposed method can be used for other applications, like customizing LLM with limited data, showing its versatility.

Advanced epidemic forecasting is critical for enabling precision containment strategies, highlighting its strategic importance for public health security. While recent advances in Large Language Models (LLMs) have demonstrated effectiveness as foundation models for domain-specific tasks, their potential for epidemic forecasting remains largely unexplored. In this paper, we introduce EpiLLM, a novel LLM-based framework tailored for spatio-temporal epidemic forecasting. Considering the key factors in real-world epidemic transmission: infection cases and human mobility, we introduce a dual-branch architecture to achieve fine-grained token-level alignment between such complex epidemic patterns and language tokens for LLM adaptation. To unleash the multi-step forecasting and generalization potential of LLM architectures, we propose an autoregressive modeling paradigm that reformulates the epidemic forecasting task into next-token prediction. To further enhance LLM perception of epidemics, we introduce spatio-temporal prompt learning techniques, which strengthen forecasting capabilities from a data-driven perspective. Extensive experiments show that EpiLLM significantly outperforms existing baselines on real-world COVID-19 datasets and exhibits scaling behavior characteristic of LLMs.

Recent advances in Generative Artificial Intelligence (GenAI) have transformed educational content creation, particularly in developing tutor training materials. However, biases embedded in AI-generated content--such as gender, racial, or national stereotypes--raise significant ethical and educational concerns. Despite the growing use of GenAI, systematic methods for detecting and evaluating such biases in educational materials remain limited. This study proposes an automated bias assessment approach that integrates the Contextualized Embedding Association Test with a prompt-engineered word extraction method within a Retrieval-Augmented Generation framework. We applied this method to AI-generated texts used in tutor training lessons. Results show a high alignment between the automated and manually curated word sets, with a Pearson correlation coefficient of r = 0.993, indicating reliable and consistent bias assessment. Our method reduces human subjectivity and enhances fairness, scalability, and reproducibility in auditing GenAI-produced educational content.

Advances in deep generative modelling have made it increasingly plausible to train human-level embodied agents. Yet progress has been limited by the absence of large-scale, real-time, multi-modal, and socially interactive datasets that reflect the sensory-motor complexity of natural environments. To address this, we present PLAICraft, a novel data collection platform and dataset capturing multiplayer Minecraft interactions across five time-aligned modalities: video, game output audio, microphone input audio, mouse, and keyboard actions. Each modality is logged with millisecond time precision, enabling the study of synchronous, embodied behaviour in a rich, open-ended world. The dataset comprises over 10,000 hours of gameplay from more than 10,000 global participants.\footnote{We have done a privacy review for the public release of an initial 200-hour subset of the dataset, with plans to release most of the dataset over time.} Alongside the dataset, we provide an evaluation suite for benchmarking model capabilities in object recognition, spatial awareness, language grounding, and long-term memory. PLAICraft opens a path toward training and evaluating agents that act fluently and purposefully in real time, paving the way for truly embodied artificial intelligence.

Large Language Models (LLMs) are increasingly deployed in interactions where they are prompted to adopt personas. This paper investigates whether such persona conditioning affects model safety under bullying, an adversarial manipulation that applies psychological pressures in order to force the victim to comply to the attacker. We introduce a simulation framework in which an attacker LLM engages a victim LLM using psychologically grounded bullying tactics, while the victim adopts personas aligned with the Big Five personality traits. Experiments using multiple open-source LLMs and a wide range of adversarial goals reveal that certain persona configurations -- such as weakened agreeableness or conscientiousness -- significantly increase victim's susceptibility to unsafe outputs. Bullying tactics involving emotional or sarcastic manipulation, such as gaslighting and ridicule, are particularly effective. These findings suggest that persona-driven interaction introduces a novel vector for safety risks in LLMs and highlight the need for persona-aware safety evaluation and alignment strategies.