❌

Normal view

Received — 4 April 2026 ⏭

The Latent Space: Foundation, Evolution, Mechanism, Ability, and Outlook

✇cs.AI, q-bio.NC updates on arXiv.org
By: Xinlei Yu · Zhangquan Chen · Yongbo He · Tianyu Fu · Cheng Yang · Chengming Xu · Yue Ma · Xiaobin Hu · Zhe Cao · Jie Xu · Guibin Zhang · Jiale Tao · Jiayi Zhang · Siyuan Ma · Kaituo Feng · Haojie Huang · Youxing Li · Ronghao Chen · Huacan Wang · Chenglin Wu · Zikun Su · Xiaogang Xu · Kelu Yao · Kun Wang · Chen Gao · Yue Liao · Ruqi Huang · Tao Jin · Cheng Tan · Jiangning Zhang · Wenqi Ren · Yanwei Fu · Yong Liu · Yu Wang · Xiangyu Yue · Yu-Gang Jiang · Shuicheng Yan
3 April 2026 at 12:00
arXiv:2604.02029v1 Announce Type: new Abstract: Latent space is rapidly emerging as a native substrate for language-based models. While modern systems are still commonly understood through explicit token-level generation, an increasing body of work shows that many critical internal processes are more naturally carried out in continuous latent space than in human-readable verbal traces. This shift is driven by the structural limitations of explicit-space computation, including linguistic redundancy, discretization bottlenecks, sequential inefficiency, and semantic loss. This survey aims to provide a unified and up-to-date landscape of latent space in language-based models. We organize the survey into five sequential perspectives: Foundation, Evolution, Mechanism, Ability, and Outlook. We begin by delineating the scope of latent space, distinguishing it from explicit or verbal space and from the latent spaces commonly studied in generative visual models. We then trace the field's evolution from early exploratory efforts to the current large-scale expansion. To organize the technical landscape, we examine existing work through the complementary lenses of mechanism and ability. From the perspective of Mechanism, we identify four major lines of development: Architecture, Representation, Computation, and Optimization. From the perspective of Ability, we show how latent space supports a broad capability spectrum spanning Reasoning, Planning, Modeling, Perception, Memory, Collaboration, and Embodiment. Beyond consolidation, we discuss the key open challenges, and outline promising directions for future research. We hope this survey serves not only as a reference for existing work, but also as a foundation for understanding latent space as a general computational and systems paradigm for next-generation intelligence.

Do Emotions in Prompts Matter? Effects of Emotional Framing on Large Language Models

✇cs.AI, q-bio.NC updates on arXiv.org
By: Minda Zhao · Yutong Yang · Chufei Peng · Rachel Gonsalves · Weiyue Li · Ruyi Yang · Zhixi Liu · Mengyu Wang
3 April 2026 at 12:00
arXiv:2604.02236v1 Announce Type: new Abstract: Emotional tone is pervasive in human communication, yet its influence on large language model (LLM) behaviour remains unclear. Here, we examine how first-person emotional framing in user-side queries affect LLM performance across six benchmark domains, including mathematical reasoning, medical question answering, reading comprehension, commonsense reasoning and social inference. Across models and tasks, static emotional prefixes usually produce only small changes in accuracy, suggesting that affective phrasing is typically a mild perturbation rather than a reliable general-purpose intervention. This stability is not uniform: effects are more variable in socially grounded tasks, where emotional context more plausibly interacts with interpersonal reasoning. Additional analyses show that stronger emotional wording induces only modest extra change, and that human-written prefixes reproduce the same qualitative pattern as LLM-generated ones. We then introduce EmotionRL, an adaptive emotional prompting framework that selects emotional framing adaptively for each query. Although no single emotion is consistently beneficial, adaptive selection yields more reliable gains than fixed emotional prompting. Together, these findings show that emotional tone is neither a dominant driver of LLM performance nor irrelevant noise, but a weak and input-dependent signal that can be exploited through adaptive control.

Predicting LLM Output Length via Entropy-Guided Representations

✇cs.AI, q-bio.NC updates on arXiv.org
By: Huanyi Xie · Yubin Chen · Liangyu Wang · Lijie Hu · Di Wang
3 April 2026 at 12:00
arXiv:2602.11812v2 Announce Type: replace Abstract: The long-tailed distribution of sequence lengths in LLM serving and reinforcement learning (RL) sampling causes significant computational waste due to excessive padding in batched inference. Existing methods rely on auxiliary models for static length prediction, but they incur high overhead, generalize poorly, and fail in stochastic "one-to-many" sampling scenarios. We introduce a lightweight framework that reuses the main model's internal hidden states for efficient length prediction. Our framework features two core components: 1) Entropy-Guided Token Pooling (EGTP), which uses on-the-fly activations and token entropy for highly accurate static prediction with negligible cost, and 2) Progressive Length Prediction (PLP), which dynamically estimates the remaining length at each decoding step to handle stochastic generation. To validate our approach, we build and release ForeLen, a comprehensive benchmark with long-sequence, Chain-of-Thought, and RL data. On ForeLen, EGTP achieves state-of-the-art accuracy, reducing MAE by 29.16\% over the best baseline. Integrating our methods with a length-aware scheduler yields significant end-to-end throughput gains. Our work provides a new technical and evaluation baseline for efficient LLM inference.
Received — 2 April 2026 ⏭

GISTBench: Evaluating LLM User Understanding via Evidence-Based Interest Verification

✇cs.AI, q-bio.NC updates on arXiv.org
By: Iordanis Fostiropoulos · Muhammad Rafay Azhar · Abdalaziz Sawwan · Boyu Fang · Yuchen Liu · Jiayi Liu · Hanchao Yu · Qi Guo · Jianyu Wang · Fei Liu · Xiangjun Fan
1 April 2026 at 12:00
arXiv:2603.29112v1 Announce Type: new Abstract: We introduce GISTBench, a benchmark for evaluating Large Language Models' (LLMs) ability to understand users from their interaction histories in recommendation systems. Unlike traditional RecSys benchmarks that focus on item prediction accuracy, our benchmark evaluates how well LLMs can extract and verify user interests from engagement data. We propose two novel metric families: Interest Groundedness (IG), decomposed into precision and recall components to separately penalize hallucinated interest categories and reward coverage, and Interest Specificity (IS), which assesses the distinctiveness of verified LLM-predicted user profiles. We release a synthetic dataset constructed on real user interactions on a global short-form video platform. Our dataset contains both implicit and explicit engagement signals and rich textual descriptions. We validate our dataset fidelity against user surveys, and evaluate eight open-weight LLMs spanning 7B to 120B parameters. Our findings reveal performance bottlenecks in current LLMs, particularly their limited ability to accurately count and attribute engagement signals across heterogeneous interaction types.

PromptForge-350k: A Large-Scale Dataset and Contrastive Framework for Prompt-Based AI Image Forgery Localization

✇cs.AI, q-bio.NC updates on arXiv.org
By: Jianpeng Wang · Haoyu Wang · Baoying Chen · Jishen Zeng · Yiming Qin · Yiqi Yang · Zhongjie Ba
1 April 2026 at 12:00
arXiv:2603.29386v1 Announce Type: cross Abstract: The rapid democratization of prompt-based AI image editing has recently exacerbated the risks associated with malicious content fabrication and misinformation. However, forgery localization methods targeting these emerging editing techniques remain significantly under-explored. To bridge this gap, we first introduce a fully automated mask annotating framework that leverages keypoint alignment and semantic space similarity to generate precise ground-truth masks for edited regions. Based on this framework, we construct PromptForge-350k, a large-scale forgery localization dataset covering four state-of-the-art prompt-based AI image editing models, thereby mitigating the data scarcity in this domain. Furthermore, we propose ICL-Net, an effective forgery localization network featuring a triple-stream backbone and intra-image contrastive learning. This design enables the model to capture highly robust and generalizable forensic features. Extensive experiments demonstrate that our method achieves an IoU of 62.5% on PromptForge-350k, outperforming SOTA methods by 5.1%. Additionally, it exhibits strong robustness against common degradations with an IoU drop of less than 1%, and shows promising generalization capabilities on unseen editing models, achieving an average IoU of 41.5%.
Received — 26 March 2026 ⏭

Memory Bear AI Memory Science Engine for Multimodal Affective Intelligence: A Technical Report

✇cs.AI, q-bio.NC updates on arXiv.org
By: Deliang Wen · Ke Sun · Yu Wang
25 March 2026 at 12:00
arXiv:2603.22306v1 Announce Type: new Abstract: Affective judgment in real interaction is rarely a purely local prediction problem. Emotional meaning often depends on prior trajectory, accumulated context, and multimodal evidence that may be weak, noisy, or incomplete at the current moment. Although multimodal emotion recognition (MER) has improved the integration of text, speech, and visual signals, many existing systems remain optimized for short-range inference and provide limited support for persistent affective memory, long-horizon dependency modeling, and robust interpretation under imperfect input. This technical report presents the Memory Bear AI Memory Science Engine, a memory-centered framework for multimodal affective intelligence. Instead of treating emotion as a transient output label, the framework models affective information as a structured and evolving variable within a memory system. It organizes processing through structured memory formation, working-memory aggregation, long-term consolidation, memory-driven retrieval, dynamic fusion calibration, and continuous memory updating. At its core, multimodal signals are transformed into structured Emotion Memory Units (EMUs), enabling affective information to be preserved, reactivated, and revised across interaction horizons. Experimental results show consistent gains over comparison systems across benchmark and business-grounded settings, with stronger accuracy and robustness, especially under noisy or missing-modality conditions. The framework offers a practical step from local emotion recognition toward more continuous, robust, and deployment-relevant affective intelligence.

Geometric Mixture-of-Experts with Curvature-Guided Adaptive Routing for Graph Representation Learning

✇cs.AI, q-bio.NC updates on arXiv.org
By: Haifang Cao · Yu Wang · Timing Li · Xinjie Yao · Pengfei Zhu
25 March 2026 at 12:00
arXiv:2603.22317v1 Announce Type: cross Abstract: Graph-structured data typically exhibits complex topological heterogeneity, making it difficult to model accurately within a single Riemannian manifold. While emerging mixed-curvature methods attempt to capture such diversity, they often rely on implicit, task-driven routing that lacks fundamental geometric grounding. To address this challenge, we propose a Geometric Mixture-of-Experts framework (GeoMoE) that adaptively fuses node representations across diverse Riemannian spaces to better accommodate multi-scale topological structures. At its core, GeoMoE leverages Ollivier-Ricci Curvature (ORC) as an intrinsic geometric prior to orchestrate the collaboration of specialized experts. Specifically, we design a graph-aware gating network that assigns node-specific fusion weights, regularized by a curvature-guided alignment loss to ensure interpretable and geometry-consistent routing. Additionally, we introduce a curvature-aware contrastive objective that promotes geometric discriminability by constructing positive and negative pairs according to curvature consistency. Extensive experiments on six benchmark datasets demonstrate that GeoMoE outperforms state-of-the-art baselines across diverse graph types.

An Accurate and Interpretable Framework for Trustworthy Process Monitoring

✇cs.AI, q-bio.NC updates on arXiv.org
By: Hao Wang · Zhiyu Wang · Yunlong Niu · Zhaoran Liu · Haozhe Li · Yilin Liao · Yuxin Huang · Xinggao Liu
25 March 2026 at 12:00
arXiv:2302.10426v3 Announce Type: replace Abstract: Trustworthy process monitoring seeks to build an accurate and interpretable monitoring framework, which is critical for ensuring the safety of energy conversion plant (ECP) that operates under extreme working conditions such as high pressure and temperature. Contemporary self-attentive models, however, fall short in this domain for two main reasons. First, they rely on step-wise correlations that fail to involve physically meaningful semantics in ECP logs, resulting in suboptimal accuracy and interpretability. Second, attention matrices are frequently cluttered with spurious correlations that obscure physically meaningful ones, further impeding effective interpretation. To overcome these issues, we propose AttentionMixer, a framework aimed at improving both accuracy and interpretability of existing methods and establish a trustworthy ECP monitoring framework. Specifically, to tackle the first issue, we employ a spatial adaptive message passing block to capture variate-wise correlations. This block is coupled with a temporal adaptive message passing block through an \textit{mixing} operator, yielding a multi-faceted representation of ECP logs accounting for both step-wise and variate-wise correlations. Concurrently, to tackle the second issue, we employ a sparse message passing regularizer to filter out spurious correlations. We validate the efficacy of AttentionMixer using two real-world datasets from the radiation monitoring network for Chinese nuclear power plants.

NUP85 as a Pan-Cancer Immune Biomarker: Integrated Multi Omics and Functional Analyses Reveal Its Role in Tumor Prognosis

✇(Multiomics OR Omics) AND (Lung OR gastric OR Hepatocellular)
By: Weihao Kong · Ying Hu · Kangjie Zhang · Zhipeng Zhu · Yajun Zou · Long Teng · Xingyu Wang · Jianlin Zhang
23 March 2026 at 18:00

Immunotargets Ther. 2026 Mar 17;15:541852. doi: 10.2147/ITT.S541852. eCollection 2026.

ABSTRACT

PURPOSE: NUP85 encodes protein components of the Nup107-160 subunit of the nuclear pore complex, belonging to the Nucleoporins (NUPs) family, potentially implicating its role in human cancer. This study aims to elucidate the potential involvement of NUP85 in cancer pathogenesis.

METHODS: Leveraging data from The Cancer Genome Atlas (TCGA), Genotype-Tissue Expression (GTEx), Clinical Proteomic Tumor Analysis Consortium (CPTAC), Cancer Cell Line Encyclopedia (CCLE), Human Protein Atlas (HPA), Gene Expression Profiling Interactive Analysis (GEPIA), CellMiner, and GeneMANIA databases, we investigated the role of NUP85 across various tumors. Correlations between NUP85 expression and pathological stage, histological grade, survival, immune infiltration, tumor mutational burden (TMB), microsatellite instability (MSI), drug resistance, DNA methylation, copy number variation (CNV), and single-cell expression were analyzed. Gene functional enrichment analysis was conducted to explore NUP85-associated pathways. Molecular biology experiments including Western blotting, flow cytometry, trans-well migration, and invasion assays were performed to validate NUP85's oncogenic role in lung adenocarcinoma (LUAD) and oral squamous cell carcinoma (OSCC) cell lines.

RESULTS: Our findings reveal up-regulated expression of NUP85 in most tumor tissues, with significant correlations observed with pathological stage, survival, immune infiltration, TMB, MSI, drug resistance, DNA methylation, and CNV. Molecular biology experiments confirm NUP85's tumor-promoting role in LUAD and OSCC cell lines. Single-cell sequencing data suggest elevated NUP85 expression primarily in proliferative T cells (Tprolif).

CONCLUSION: NUP85 emerges as a potential tumor marker associated with tumor immunity and poor prognosis. These insights offer avenues for the development of novel therapeutic targets and anti-neoplastic drugs.

PMID:41869435 | PMC:PMC13005628 | DOI:10.2147/ITT.S541852

Received — 17 March 2026 ⏭

SkillsBench: Benchmarking How Well Agent Skills Work Across Diverse Tasks

✇cs.AI, q-bio.NC updates on arXiv.org
By: Xiangyi Li · Wenbo Chen · Yimin Liu · Shenghan Zheng · Xiaokun Chen · Yifeng He · Yubo Li · Bingran You · Haotian Shen · Jiankai Sun · Shuyi Wang · Binxu Li · Qunhong Zeng · Di Wang · Xuandong Zhao · Yuanli Wang · Roey Ben Chaim · Zonglin Di · Yipeng Gao · Junwei He · Yizhuo He · Liqiang Jing · Luyang Kong · Xin Lan · Jiachen Li · Songlin Li · Yijiang Li · Yueqian Lin · Xinyi Liu · Xuanqing Liu · Haoran Lyu · Ze Ma · Bowei Wang · Runhui Wang · Tianyu Wang · Wengao Ye · Yue Zhang · Hanwen Xing · Yiqi Xue · Steven Dillmann · Han-chung Lee
16 March 2026 at 12:00
arXiv:2602.12670v3 Announce Type: replace Abstract: Agent Skills are structured packages of procedural knowledge that augment LLM agents at inference time. Despite rapid adoption, there is no standard way to measure whether they actually help. We present SkillsBench, a benchmark of 86 tasks across 11 domains paired with curated Skills and deterministic verifiers. Each task is evaluated under three conditions: no Skills, curated Skills, and self-generated Skills. We test 7 agent-model configurations over 7,308 trajectories. Curated Skills raise average pass rate by 16.2 percentage points(pp), but effects vary widely by domain (+4.5pp for Software Engineering to +51.9pp for Healthcare) and 16 of 84 tasks show negative deltas. Self-generated Skills provide no benefit on average, showing that models cannot reliably author the procedural knowledge they benefit from consuming. Focused Skills with 2--3 modules outperform comprehensive documentation, and smaller models with Skills can match larger models without them.

OpenVision 3: A Family of Unified Visual Encoder for Both Understanding and Generation

✇cs.AI, q-bio.NC updates on arXiv.org
By: Letian Zhang · Sucheng Ren · Yanqing Liu · Xianhang Li · Zeyu Wang · Yuyin Zhou · Huaxiu Yao · Zeyu Zheng · Weili Nie · Guilin Liu · Zhiding Yu · Cihang Xie
16 March 2026 at 12:00
arXiv:2601.15369v2 Announce Type: replace-cross Abstract: This paper presents a family of advanced vision encoder, named OpenVision 3, that learns a single, unified visual representation that can serve both image understanding and image generation. Our core architecture is simple: we feed VAE-compressed image latents to a ViT encoder and train its output to support two complementary roles. First, the encoder output is passed to the ViT-VAE decoder to reconstruct the original image, encouraging the representation to capture generative structure. Second, the same representation is optimized with contrastive learning and image-captioning objectives, strengthening semantic features. By jointly optimizing reconstruction- and semantics-driven signals in a shared latent space, the encoder learns representations that synergize and generalize well across both regimes. We validate this unified design through extensive downstream evaluations with the encoder frozen. For generation, we test it under the RAE framework: ours substantially surpasses the standard CLIP-based encoder (e.g., gFID: 1.87 vs. 2.54 on ImageNet). For multimodal understanding, we plug the encoder into the LLaVA-1.5 and LLaVA-NeXT framework: it performs comparably with a standard CLIP vision encoder (e.g., 63.3 vs. 61.2 on SeedBench, and 59.2 vs. 58.1 on GQA). We provide empirical evidence that generation and understanding are mutually beneficial in our architecture, while further underscoring the critical role of the VAE latent space. We hope this work can spur future research on unified modeling.
Received — 14 March 2026 ⏭

Machine Learning for Stress Testing: Uncertainty Decomposition in Causal Panel Prediction

✇cs.AI, q-bio.NC updates on arXiv.org
By: Yu Wang · Xiangchen Liu · Siguang Li
10 March 2026 at 12:00
arXiv:2603.07438v1 Announce Type: new Abstract: Regulatory stress testing requires projecting credit losses under hypothetical macroeconomic scenarios -- a fundamentally causal question typically treated as a prediction problem. We propose a framework for policy-path counterfactual inference in panels that transparently separates what can be learned from data from what requires assumptions about confounding. Our approach has four components: (i) observational identification of path-conditional means via iterated regression, enabling continuous macro-path contrasts without requiring a control group; (ii) causal set identification under bounded confounding, yielding sharp identified sets with interpretable breakdown values that communicate robustness in a single number; (iii) an oracle inequality showing that recursive rollout error is governed by a horizon-dependent amplification factor, providing a concrete answer to how far ahead one can reliably predict under stress; and (iv) importance-weighted conformal calibration bands with diagnostics that quantify extrapolation cost and trigger abstention when coverage guarantees degrade. The final output is a three-layer uncertainty decomposition that cleanly separates estimation uncertainty from confounding uncertainty. We validate all results through simulation and semi-synthetic experiments with real unemployment data, including a Covid retrospective demonstrating the framework's diagnostic value under extreme scenarios.
Received — 11 March 2026 ⏭

A prior information informed learning architecture for flying trajectory prediction

✇cs.AI, q-bio.NC updates on arXiv.org
By: Xianda Huang · Zidong Han · Ruibo Jin · Zhenyu Wang · Wenyu Li · Xiaoyang Li · Yi Gong
10 March 2026 at 12:00
arXiv:2603.06863v1 Announce Type: cross Abstract: Trajectory prediction for flying objects is critical in domains ranging from sports analytics to aerospace. However, traditional methods struggle with complex physical modeling, computational inefficiencies, and high hardware demands, often neglecting critical trajectory events like landing points. This paper introduces a novel, hardware-efficient trajectory prediction framework that integrates environmental priors with a Dual-Transformer-Cascaded (DTC) architecture. We demonstrate this approach by predicting the landing points of tennis balls in real-world outdoor courts. Using a single industrial camera and YOLO-based detection, we extract high-speed flight coordinates. These coordinates, fused with structural environmental priors (e.g., court boundaries), form a comprehensive dataset fed into our proposed DTC model. A first-level Transformer classifies the trajectory, while a second-level Transformer synthesizes these features to precisely predict the landing point. Extensive ablation and comparative experiments demonstrate that integrating environmental priors within the DTC architecture significantly outperforms existing trajectory prediction frameworks

An Embedding-based Approach to Inconsistency-tolerant Reasoning with Inconsistent Ontologies

✇cs.AI, q-bio.NC updates on arXiv.org
By: Keyu Wang · Site Li · Jiaye Li · Guilin Qi · Qiu Ji
10 March 2026 at 12:00
arXiv:2304.01664v3 Announce Type: replace Abstract: Inconsistency handling is an important issue in knowledge management. Especially in ontology engineering, logical inconsistencies may occur during ontology construction. A natural way to reason with an inconsistent ontology is to utilize the maximal consistent subsets of the ontology. However, previous studies on selecting maximum consistent subsets have rarely considered the semantics of the axioms, which may result in irrational inference. In this paper, we propose a novel approach to reasoning with inconsistent ontologies in description logics based on the embeddings of axioms. We first give a method for turning axioms into distributed semantic vectors to compute the semantic connections between the axioms. We then define an embedding-based method for selecting the maximum consistent subsets and use it to define an inconsistency-tolerant inference relation. We show the rationality of our inference relation by considering some logical properties. Finally, we conduct experiments on several ontologies to evaluate the reasoning power of our inference relation. The experimental results show that our embedding-based method can outperform existing inconsistency-tolerant reasoning methods based on maximal consistent subsets.

MAS-Orchestra: Understanding and Improving Multi-Agent Reasoning Through Holistic Orchestration and Controlled Benchmarks

✇cs.AI, q-bio.NC updates on arXiv.org
By: Zixuan Ke · Yifei Ming · Austin Xu · Ryan Chin · Xuan-Phi Nguyen · Prathyusha Jwalapuram · Jiayu Wang · Semih Yavuz · Caiming Xiong · Shafiq Joty
10 March 2026 at 12:00
arXiv:2601.14652v4 Announce Type: replace Abstract: While multi-agent systems (MAS) promise elevated intelligence through coordination of agents, current approaches to automatic MAS design under-deliver. Such shortcomings stem from two key factors: (1) methodological complexity - agent orchestration is performed using sequential, code-level execution that limits global system-level holistic reasoning and scales poorly with agent complexity - and (2) efficacy uncertainty - MAS are deployed without understanding if there are tangible benefits compared to single-agent systems (SAS). We propose MASOrchestra, a training-time framework that formulates MAS orchestration as a function-calling reinforcement learning problem with holistic orchestration, generating an entire MAS at once. In MAS-Orchestra, complex, goal-oriented subagents are abstracted as callable functions, enabling global reasoning over system structure while hiding internal execution details. To rigorously study when and why MAS are beneficial, we introduce MASBENCH, a controlled benchmark that characterizes tasks along five axes: Depth, Horizon, Breadth, Parallel, and Robustness. Our analysis reveals that MAS gains depend critically on task structure, verification protocols, and the capabilities of both orchestrator and subagents, rather than holding universally. Guided by these insights, MAS-Orchestra achieves consistent improvements on public benchmarks including mathematical reasoning, multi-hop QA, and search-based QA, while achieving more than 10x efficiency over strong baselines. Together, MAS-Orchestra and MASBENCH enable better training and understanding of MAS in the pursuit of multi-agent intelligence.

SkillsBench: Benchmarking How Well Agent Skills Work Across Diverse Tasks

✇cs.AI, q-bio.NC updates on arXiv.org
By: Xiangyi Li · Wenbo Chen · Yimin Liu · Shenghan Zheng · Xiaokun Chen · Yifeng He · Yubo Li · Bingran You · Haotian Shen · Jiankai Sun · Shuyi Wang · Binxu Li · Qunhong Zeng · Di Wang · Xuandong Zhao · Yuanli Wang · Roey Ben Chaim · Zonglin Di · Yipeng Gao · Junwei He · Yizhuo He · Liqiang Jing · Luyang Kong · Xin Lan · Jiachen Li · Songlin Li · Yijiang Li · Yueqian Lin · Xinyi Liu · Xuanqing Liu · Haoran Lyu · Ze Ma · Bowei Wang · Runhui Wang · Tianyu Wang · Wengao Ye · Yue Zhang · Hanwen Xing · Yiqi Xue · Steven Dillmann · Han-chung Lee
10 March 2026 at 12:00
arXiv:2602.12670v2 Announce Type: replace Abstract: Agent Skills are structured packages of procedural knowledge that augment LLM agents at inference time. Despite rapid adoption, there is no standard way to measure whether they actually help. We present SkillsBench, a benchmark of 86 tasks across 11 domains paired with curated Skills and deterministic verifiers. Each task is evaluated under three conditions: no Skills, curated Skills, and self-generated Skills. We test 7 agent-model configurations over 7,308 trajectories. Curated Skills raise average pass rate by 16.2 percentage points(pp), but effects vary widely by domain (+4.5pp for Software Engineering to +51.9pp for Healthcare) and 16 of 84 tasks show negative deltas. Self-generated Skills provide no benefit on average, showing that models cannot reliably author the procedural knowledge they benefit from consuming. Focused Skills with 2--3 modules outperform comprehensive documentation, and smaller models with Skills can match larger models without them.

A Simple "Motivation" Can Enhance Reinforcement Finetuning of Large Reasoning Models

✇cs.AI, q-bio.NC updates on arXiv.org
By: Junjie Zhang · Guozheng Ma · Shunyu Liu · Haoyu Wang · Jiaxing Huang · Ting-En Lin · Fei Huang · Yongbin Li · Dacheng Tao
10 March 2026 at 12:00
arXiv:2506.18485v3 Announce Type: replace-cross Abstract: Reinforcement Learning with Verifiable Rewards~(RLVR) has emerged as a powerful learn-to-reason paradigm for large reasoning models to tackle complex tasks. However, the current RLVR paradigm is still not efficient enough, as it works in a trial-and-error manner. To perform better, the model needs to explore the reward space by numerously generating responses and learn from fragmented reward signals, blind to the overall reward patterns. Fortunately, verifiable rewards make the natural language description of the reward function possible, and meanwhile, LLMs have demonstrated strong in-context learning ability. This motivates us to explore if large reasoning models can benefit from a \textbf{motivation} of the task, \textit{i.e.}, awareness of the reward function, during the reinforcement finetuning process, as we humans sometimes do when learning. In this paper, we introduce \textit{\textbf{M}otivation-\textbf{e}nhanced \textbf{R}einforcement \textbf{F}inetuning}~(\textbf{MeRF}), an intuitive yet effective method enhancing reinforcement finetuning of LLMs by involving \emph{``telling LLMs rules of the game''}. Specifically, \textbf{MeRF} directly injects the reward specification into the prompt, which serves as an in-context motivation for the model to be aware of the optimization objective. This simple modification leverages the in-context learning ability of LLMs, aligning generation with optimization, thereby incentivizing the model to generate desired outputs from both inner motivation and external reward. Empirical evaluations demonstrate that \textbf{MeRF} achieves substantial performance gains over the RLVR baseline. Moreover, ablation studies show that MeRF performs better with greater consistency between the in-context motivation and the external reward function, while the model also demonstrates an ability to adapt to misleading motivations through reinforcement finetuning.
❌