Speech large language models retain audio history to support multi-turn conversations. Yet remembering an event does not ensure that its information is used only when relevant. We investigate why irrelevant historical audio changes answers even when the queried event remains represented. Controlled experiments on MOSS-Audio and MiMo-Audio show that historical values direct errors toward specific competing answers, beyond a generic effect of context length. Causal interventions distinguish two paths. In open-ended question answering, most history-value errors disappear when subsequent computation cannot directly access the irrelevant source. In multiple-choice answering, many errors persist: early historical influence is carried by question-candidate states even after direct source access is blocked. Low-dimensional interventions further separate source-role information from answer-value information. Their effects peak at different depths, and changing the source-role signal does not eliminate the causal influence of competing values. On naturally occurring errors, strengthening the queried role and counteracting the historical value each improve answers; their joint intervention gives the greatest recovery. These findings support a failure of downstream evidence use beyond source identification alone: source-role information can influence a decision without exclusively controlling which historical values affect it. Reliable conversational memory therefore requires not only retention and retrieval, but also control over how retained information enters answer computation.
LLM agents are judged by what they deliver, and in professional work what gets delivered is a project, not a task. A project pays off only if the whole package holds, and one weak part can limit the value of the rest. Yet existing benchmarks mostly test agents on one task at a time, leaving a gap between the capability we measure and the outcome that makes agents useful. To address this gap, we introduce DeliverBench, which tests whether agents can carry a project from brief to final delivery. Its 11 projects, spanning financial due diligence, legal contract drafting, and drug safety reporting, are reconstructed with 13 domain experts from real materials and chain 150 subtasks through 2-15 rounds of new materials and client feedback. Because no reference outputs are injected, agents inherit their own errors from one task to the next. Expert rubrics then score every checkpoint as well as the final package. Across seven model-harness systems, even the best system scores only 65.0 out of 100 on final delivery and passes 50.0% of subtasks at a 90-point threshold. The failures are often failures of carrying work forward: in one audit project, all seven systems addressed the review findings, but only one preserved traceability. Together, these results suggest that task-level capability does not simply compose into delivery, which also depends on keeping the work coherent as it evolves.
Large language models are deployed as long-horizon agents that compose diverse tools across MCP services into executable workflows. Existing MCP environment-scaling methods are data-centric: they first synthesize a database and scenario, then derive tools and tasks from them. As a result, tool interfaces may not match real ones, tool compositions are not grounded in execution, and tasks rarely span more than one environment. We propose ToolSpace, a tool-centric framework that reverses this order. ToolSpace mines data, state, and retry dependencies from adjacent calls in execution traces of real MCP services and open agent corpora, using deterministic rules rather than LLM judgment, and clusters the resulting weighted tool-composition graph into environments. Databases are synthesized only for state-mutating environments, while query and stateless tools replay recorded returns. ToolSpace then performs state-aware random walks with rollback and decompiles each chain into a natural-language task with assertion-based checks, keeping only samples that replay successfully from the initial state. Applied to AutomationBench and tau3-Bench Banking, the pipeline produces harder Pro versions on which 7 frontier models drop by up to 38.2 points. Across seven long-horizon tool-use benchmarks, ToolSpace raises Qwen3.6-35B-A3B from 23.0 to 41.0 average score (43.2 with RL) and GLM-5.2 from 44.6 to 56.6, on par with Claude Opus 5. Ablations show that execution-witnessed edges clearly outperform LLM-judged ones, and performance keeps improving with data and environment scale.
Xinbo Zhao, Yuhang Liu, Xin Zhang, Lei Le, Zhaohan Xi, Yu Yang, Yingxue Zhang
Offline multi-task reinforcement learning (MTRL) aims to learn a unified policy from heterogeneous offline datasets that generalizes across tasks. However, existing offline MTRL methods often struggle with two fundamental issues: severe per-task data scarcity and limited cross-task generalization. In this paper, we propose Language-Model Augmented Decision Transformer (LaMA-DT), a unified sequence-modeling framework that leverages a single large language model (LLM) for both trajectory augmentation and multi-task policy learning. LaMA-DT uses natural-language task descriptions to guide task-aligned trajectory augmentation through a set of supervised generation and alignment objectives, and incorporates a scoring-and-filtering mechanism to retain high-quality augmented trajectories. The same LLM is then repurposed as a language-conditioned Decision Transformer and fine-tuned on a mixture of real and augmented data. Experiments on Meta-World and MuJoCo multi-task benchmarks demonstrate that LaMA-DT consistently outperforms competitive baselines, particularly in low-data regimes, showing that language-guided data augmentation substantially improves cross-task generalization.
Johannes Knittel, Simon Warchol, Jakob Troidl, Camelia D. Brumar, Helen Yu Yang, Eric Mörth, Robert Krüger, Daniel Needleman 等 10 位作者
In-vitro fertilization (IVF) has become standard practice to address infertility, which affects more than one in ten couples in the US. However, current protocols yield relatively low success rates of about 20% per treatment cycle. A critical but complex and time-consuming step is the grading and selection of embryos for implantation. Although incubators with time-lapse microscopy have enabled computational analysis of embryo development, existing automated approaches either require extensive manual annotations or use opaque deep learning models that are hard for clinicians to validate and trust. We present EmbryoProfiler, a visual analytics system collaboratively developed with embryologists, biologists, and machine learning researchers to support clinicians in visually assessing embryo viability from time-lapse microscopy imagery. Our system incorporates a deep learning pipeline that automatically annotates microscopy images and extracts clinically interpretable features relevant for embryo grading. Our contributions include: (1) a semi-automatic, visualization-based workflow that guides clinicians through fertilization assessment, developmental timing evaluation, morphological inspection, and comparative analysis of embryos; (2) innovative interactive visualizations, such as cell-shape plots, designed to facilitate efficient analysis of morphological and developmental characteristics; and (3) an integrated, explainable machine learning classifier offering transparent, clinically-informed embryo viability scoring to predict live birth outcomes. Quantitative evaluation of our classifier and qualitative case studies conducted with practitioners demonstrate that EmbryoProfiler enables clinicians to make better-informed embryo selection decisions, potentially leading to improved clinical outcomes in IVF treatments.
Rapid advances in cowork agents are shifting human-AI collaboration toward real-world automation, delivering software, designs, and other artifacts. However, such productivity work demands both high-quality delivery on complex productivity tasks and continued monitoring and decision-making over long horizons, a combination that remains an open challenge. We introduce CoworkOS, a general-purpose harness for long-horizon work that separates task strategy from execution management: the model decides actions, delegation, and acceptance, while the runtime maintains state and feedback. Its three-layer architecture enables direct execution or delegation to Subagents, synchronously or asynchronously, under stable responsibilities. Through a shared application layer, humans and agents inspect and revise the same evolving documents, presentations, CAD models, and Blender scenes, so human-in-the-loop participation can enter both through conversation and through direct modification of the work itself. To sustain continuity across agents and sessions, CoworkOS further introduces hierarchical workspaces and persistent semantic memory, so working files, key decisions, and long-running Job logs remain shareable, reusable, and resumable after interruption. Across 12 benchmarks and 6 categories of real-world applications, CoworkOS proves highly competitive with frontier agent applications such as Codex and DeepSeek Harness. As a sustained-work case study, CoworkOS uses GLM-5.3 as its backbone to conduct RSI-driven automated post-training on Qwen3.6-35B-A3B. Over five rounds, the model's average score across eight benchmarks rose from 33.19 to 42.95 (+9.76). Looking ahead, we believe CoworkOS opens a meaningful path toward general cowork-agent intelligence.
Peiyu Yang, Mohamad S. Moudallal, Antonio Langella, Giulio Turrisi, Victor Barasuol, Claudio Semini
Contact-based robotic manipulation is constrained by end-effector reachability and the mechanical properties of objects. Airflow can exert forces without direct robot-object contact, enabling manipulation of objects beyond the reach or when direct contact is undesirable. However, the lack of suitable simulation has hindered the application of reinforcement learning to airflow-based manipulation, limiting the broader adoption of this mechanism compared with contact-based methods. To address this gap, we developed a robot- and task-agnostic airflow extension for Isaac Lab that supports massively parallel training. The extension uses experimentally identified airflow models to compute the forces and torques exerted on objects. Based on this extension, we train a hybrid push-blow (HPB) policy that combines contact-based pushing with airflow manipulation and a rectangular reorientation (RR) policy that controls a cuboid's heading using airflow alone. To enable physical execution, we design a lightweight, high-speed blower that coexists with a gripper. In zero-shot real-world deployment, HPB achieves an 80% success rate over 20 trials, while RR achieves a mean absolute heading error of 2.6 degrees across four target headings. The project materials are available at https://github.com/iit-DLSLab/BeyondContact.
Jun Fang, Yueying Suo, Yuntao Wang, Susu Wang, Xiaoyu Yang, Qiqi Liu, Chuangyu Ding, Yuanchun Shi
Pre-meal photo capture is lightweight only with a smartphone, yet habit formation requires repeatedly initiating and completing the action within the narrow window before the first bite. We examine how supports based on behavior change techniques can shape capture adherence, perceived habit, and support needs over time. Using a participant-informed selection process, we implemented three smartphone supports targeting Prompts/cues, Focus on past success, and Social support. We evaluated them in a five-week field study with 54 participants, spanning support, withdrawal, and reinstatement, followed by interviews and a three-week follow-up. Across 4,945 main meals, 63.9\% were confirmed pre-meal captures, and adherence was broadly maintained across phases. Perceived automaticity and spontaneous remembering improved mainly during initial support. As capture became more familiar, participants combined spontaneous remembering with deliberate strategies, reduced active feature use while continuing to value support availability, and carried the practice forward through accumulated records, familiar tools, and chosen relationships. These findings frame BCT support for pre-meal photo capture as a configurable arrangement. Systems should preserve a predictable capture pathway while adapting when, how, and through whom support remains available.
Science experiment videos demonstrate how experiments are performed, but offer no direct interaction for learners. Turning them into virtual experiments requires not only reproducing the observed apparatus and state changes, but also enforcing the preconditions governing each operation. We introduce Lab2Sim, a benchmark of 218 science experiment videos with evidence-grounded specifications and 11,881 rubric items. To address this task, we propose EduVideo2Code, which extracts experimental procedures hierarchically, encodes them as explicit state machines to guide code generation, and refines the resulting pages through execution-based verification. Experiments show that existing multimodal models readily generate runnable pages but struggle to faithfully reconstruct experimental apparatus. EduVideo2Code consistently improves all four evaluation dimensions across two backbone models, with apparatus fidelity gains of 24.15 points.
Bangrui Xu, Jie Yang, Sidi Chen, Xuanhe Zhou, Jiantao Qiu, Qianyu Yang, Zijie Wang, Fan Wu 等 9 位作者
Recursively self-improving (RSI) systems are increasingly viewed as a promising means of accelerating progress toward artificial general intelligence. They aim to autonomously improve their capabilities by refining their working pipelines and recursively improving the strategies that guide subsequent refinements, with gains generalizing to unseen scenarios.However, existing evaluations lack reliable metrics to distinguish capability gains from scenario fitting, sustainable and extensible protocols for unseen tasks and difficulties, and fine-grained attribution of improvement to evolution traces. To this end, we propose SelfImprove-Bench,an RSI evaluation framework that organizes tasks into three primary capabilities (i.e., Exploring, Planning, Reasoning) as scenarios for guiding model evolution. We instantiated the tasks on games for controllability and reproducibility, and pairs them with scalable generation and orchestration driven by task templates and difficulty coefficients. Beyond task-level scores, SelfImprove-Bench organizes evaluation into generalization ladder levels G0 to G4 that measure how far gains transfer from training conditions. Experiments reveal that the evolutionary gains of existing RSI methods are inconsistent across generalization levels.
DolphinRouting: Cluster-Level Binary Routing for Disaggregated LLM Inference Serving
Yishuo Yuan, Shaoyu Yang, Chunrong Fang
AI 速览：DolphinRouting 按请求特征聚类，并用离线策略搜索为每个簇学习一个二元路由决策，从而在 PD 解耦集群中选择本地或远程 prefill。
Prefill-decode (PD) disaggregation isolates the two phases of large language model serving, but request-level path selection must account for prefill work, cache reuse, and decode demand. We present DolphinRouting, which learns this choice from end-to-end SLO feedback under a given P/D resource allocation. Requests are clustered by raw prompt length, cache-adjusted input lengths on both worker pools, and output-length information. Offline policy search learns one binary action per cluster. Online feature extraction, cluster assignment, and table lookup select local or remote prefill. We evaluate DolphinRouting on a real PD-disaggregated cluster using three public benchmarks, a mixed workload, and real data. Across the mixed workload and real data, DolphinRouting with $k$-means reduces P90 TTFT by 5.7--7.9\% and P90 TPOT by 2.6--10.1\% relative to Dynamo.
Bimanual dexterous manipulation requires continuous coordination of wrist motion, finger contact, and object motion. Human videos provide rich manipulation experience. Using this experience for robot control requires aligning human poses with robot commands and accounting for embodiment-specific kinematics and contact dynamics. We present AcrossDEX 1.0, a learning system that connects human motion pretraining, simulation-based expert learning, robot demonstration collection, and visual closed-loop execution. Pretraining on HOT3D Aria predicts one second of bimanual motion from five images and thirty annotated hand states over the preceding second. Robot adaptation preserves the visual encoder, temporal backbone, and wrist parameters. Shared wrist representations, separate finger fields, and validity masks support both six-channel Inspire hands and sixteen-joint Allegro hands. The robot policy predicts sixty control targets from frontal images, command history, and task conditions. It executes a prefix and replans from updated observations. We compare direct regression, diffusion, and flow matching on lid opening and return and closed-lid object reorientation. Direct regression achieves success rates of 40.6\% for opening and return, 100.0\% for Inspire reorientation, and 93.3\% for Allegro reorientation. Flow matching achieves 90.0\% on Allegro reorientation. Paired comparisons further show that constraining replanning with the previous plan improves the continuous execution of generative policies. These results demonstrate a common adaptation pipeline for two hand embodiments and highlight the roles of contact coordination and action-chunk continuity.
Self-evolving agent skills enable AI agents to accumulate and refine reusable procedural knowledge from interaction experience without updating model parameters. However, since such skills are typically represented as textual documents and executed through LLM reasoning, even previously learned procedures must be repeatedly interpreted and regenerated, introducing avoidable uncertainty into every execution. In contrast, program execution is deterministic and independently verifiable. Intuitively, transforming well-specified operations into executable programs can reduce reliance on LLM reasoning and, consequently, reduce execution uncertainty. Motivated by this intuition, we introduce SKillCompiler, a self-evolving skill framework that integrates programmatic compilation into the skill evolution process. Specifically, SkillCompiler adopts an executability-aware skill compilation strategy that decomposes skills into fine-grained operations, analyzes their executability, and constructs hybrid skill skeletons that retain LLM reasoning for context-dependent operations while compiling well-specified ones into validated programs. To avoid redundant program synthesis and enable executable knowledge to accumulate across evolving skills, we further design a program asset reuse mechanism that organizes validated programs as reusable assets, retrieves existing implementations whenever applicable, and synthesizes and verifies new programs only when necessary. Extensive experiments on five benchmarks and six model families demonstrate that SkillCompiler consistently outperforms state-of-the-art methods, achieving up to 32.33\% improvement in task performance.
Chengquan Guo, Chulin Xie, Yu Yang, Zhaorun Chen, Zinan Lin, Xander Davies, Yarin Gal, Dawn Song 等 9 位作者
Code agents have gained widespread adoption due to their strong code generation capabilities and integration with code interpreters, enabling dynamic execution, debugging, and interactive programming capabilities. While these advancements have streamlined complex workflows, they have also introduced critical safety and security risks. Current static safety benchmarks and red-teaming tools are inadequate for identifying emerging real-world risky scenarios, as they fail to cover certain boundary conditions, such as the combined effects of different jailbreak tools. In this work, we propose RedCodeAgent, the first automated red-teaming agent designed to systematically uncover vulnerabilities in diverse code agents. With an adaptive memory module, RedCodeAgent can leverage existing jailbreak knowledge, dynamically select the most effective red-teaming tools and tool combinations in a tailored toolbox for a given input query, thus identifying vulnerabilities that might otherwise be overlooked. For reliable evaluation, we develop simulated sandbox environments to additionally evaluate the execution results of code agents, mitigating potential biases of LLM-based judges that only rely on static code. Through extensive evaluations across multiple state-of-the-art code agents, diverse risky scenarios, and various programming languages, RedCodeAgent consistently outperforms existing red-teaming methods, achieving higher attack success rates and lower rejection rates with high efficiency. We further validate RedCodeAgent on real-world code assistants, e.g., Cursor and Codeium, exposing previously unidentified security risks. By automating and optimizing red-teaming processes, RedCodeAgent enables scalable, adaptive, and effective safety assessments of code agents. Our code and data are available at https://github.com/1mocat/RedCodeAgent.
Rounded capacity cuts (RCCs) are among the most effective cuts for capacitated vehicle-routing relaxations, but exact separation is computationally prohibitive at scale. Existing heuristic and neural separators can be fast, but often fail to produce sufficiently many effective cuts within practical separation budgets. We introduce SAG-Sep, a separator that scores node membership in RCC-inducing subsets across vehicle-count levels from a fractional LP solution in a single encoder forward pass, avoiding the iterative predict-and-coarsen inference used by NeuralSEP-style separators. SAG-Sep encodes the LP solution as a Sparse Augmented Graph (SAG), adding flow-informed multi-hop edges that expose long-range routing structure without densifying attention. Beyond LP-aware sparse encoding, we observe that high-value exact RCCs are typically nested, organizing into onion-like subset families, and SAG-Sep's node scores track subset nesting depth. This motivates Onion-Sweep (OS), a training-free decoder that ranks customers by their scores, evaluates the RCC violation of every prefix of the ranking in one vectorized pass, and emits the most violated prefix of each replica together with a cross-replica family of nested proposals, converting independent node scores into structurally coordinated RCC candidates without a tuned threshold or probing band. On the NeuralSEP benchmark with N=1000 customer instances, SAG-Sep with OS reduces the average root gap by 17.7% relative to the strongest NeuralSEP baseline, while the single-pass SAG-Sep separator achieves a roughly 30-fold per-iteration neural separation speedup. Our findings suggest that LP-aware sparse encoding and the nested structure of high-value cuts are structural patterns worth leveraging in neural separation more broadly.
Adrien Petralia, Alexandre Gavaudan, Artjom Joosen, Etienne Goffinet, Richard Kurle, Yassine El Ouahidi, Yassir Bendou, Abdelaziz Bounhar 等 356 位作者
We introduce Timestral, a family of four time series foundation models for probabilistic forecasting, with 5M, 28M, 123M, and 209M parameters. Timestral sets a new state of the art on GIFT-Eval, TIME, and fev-bench, where Timestral-209M and Timestral-123M take the top two places on every metric. We achieve this with three ingredients: Decoupled Horizon Forecasting (DHF), which separates prediction tokens from the causally encoded context, so that longer horizons are trained without masking context and each observed value is both a forecast target and context for later forecasts; a covariate framework that natively supports known-future covariates within a causal architecture; and a two-stage pipeline that pretrains exclusively on a diverse, large-scale synthetic corpus before post-training on a mix of real and synthetic data with extended context. This report describes the model design and its benchmark results. Weights for all four models are available on Hugging Face and inference code on GitHub.
Mixture-of-Experts (MoE) inference on memory-constrained edge platforms is often dominated by expert-weight movement rather than computation. Although each token activates only a small subset of experts, routing is input-dependent, so all experts must remain accessible and memory movement dominates latency. Existing approaches such as uniform pruning and expert dropping overlook two properties of MoE execution: expert combinations are highly skewed in frequency, and experts contribute asymmetrically within each combination. We present CRISP, a combination-aware framework that co-designs pruning and runtime scheduling for efficient MoE serving on memory-constrained edge platforms. CRISP reorders neurons by importance so one stored expert can support multiple execution widths, assigns differentiated execution to hot, warm, and cold combinations based on frequency and expert role, and exploits heterogeneous widths to overlap expert loading with computation. Across Mixtral-8x7B and DeepSeek-MoE-16B, CRISP improves throughput by up to $2.1\times$ at the same memory budget while limiting accuracy loss to 3.6\% relative to full-width inference. Our results highlight that efficient edge MoE serving requires jointly optimizing which expert capacity is preserved and how expert execution is scheduled.
Xiaojing Du, Chunlei Meng, Wentao Gao, Ke Sun, Feiyu Yang
On online social platforms, equal seed access to interventions can yield unequal group outcomes through social spillovers, so an access audit alone can miss outcome disparities. We characterise this trade-off through an exact allocation-structure decomposition and a sharp first-moment floor for graph-independent seeding under additive spillovers on stochastic block networks. Decoupled Procedural Allocation (DPA) combines group quotas with selection by seeds' contributions to each group's outcome. Rankings balance total outcome against disparity, and within-quota exchanges reduce the gap. Computable bounds quantify retained total outcome. DPA achieves the smallest mean outcome gap among the compared methods at the common access tolerances on both development and prospective suites. Exposure and independent-cascade studies show how the propagation mechanism changes this trade-off. On Pokec, DPA nearly equalises modelled gender outcomes at the closest access-parity quota while preserving most of the degree-based allocation's total outcome.
Qiaoling Chen, Jiaxiang Cheng, Xueqing Deng, Tiankai Hang, Ji Hou, Jiaqi Li, Yifan Liu, Tao Lu 等 38 位作者
Interactive world generation requires models to follow successive commands while preserving relevant history and action consequences. We present Hy-Nomas, a video world model with an 8B diffusion backbone that supports free-form text instructions and continuous camera control across multiple interaction turns. It generates variable-length segments conditioned on current inputs and cached history. Our central design aligns structured video captions with action-aware causal attention: descriptive context remains accessible across segments, while action instructions are restricted to their intended temporal windows. A vision-language Prompt Enhancer (PE) translates user requests and multimodal interaction histories into structured conditions for the next segment. A multi-stage curriculum progressively introduces causal generation, text interaction, camera control, and training on model-generated histories, followed by diffusion distillation. In human evaluation, Hy-Nomas shows superior performance over existing world models. We release model weights, code and benchmarks, together with a lightweight adaptation workflow, to support reproducible research on interactive world generation.
Haotian He, Zeyu Yan, Qipeng Liu, Ning Guo, Wenzhao Lian
Force signals provide critical interaction cues for contact-rich robotic manipulation. However, existing methods mostly use force as an additional observation modality, without fully exploiting its role in modeling future interaction dynamics or guiding execution-time feedback correction. In this paper, we propose FAWAM, a force-aware world action model that incorporates force information at three levels: perception, prediction, and closed-loop execution. FAWAM first encodes historical 6-axis force/torque signals to modulate action generation, then jointly predicts future actions and end-effector wrenches to explicitly model contact evolution. It further introduces a residual correction module that uses the predicted wrench trajectory as an execution-time reference to refine actions online based on real-time force feedback. Real-world experiments across multiple contact-rich tasks show that FAWAM improves the average success rate by 36.25% over vision-only baselines and 21.25% over existing force-aware baselines, demonstrating the effectiveness of our force-aware framework for robust contact-rich manipulation.
Jinghong Chen, Jingbiao Mei, Guangyu Yang, Bill Byrne
A common approach to question answering with retrieval-augmented generation (RAG) is to concatenate documents into a single context and pass it to a language model to generate an answer. While simple, this strategy can obscure the contribution of individual documents, making attribution difficult and contributing to the ``lost-in-the-middle'' effect, where relevant information in long contexts is overlooked. Concatenation also scales poorly: computational cost grows quadratically with context length, a problem that becomes especially severe when the context includes visual data, as in visual question answering. Attempts to mitigate these issues by limiting context length can further restrict performance by preventing models from benefiting from the improved recall offered by deeper retrieval. We propose Bayesian Ensemble Retrieval-Augmented Generation (BERAG), along with Bayesian Ensemble Fine-Tuning (BEFT), as a RAG framework in which language models are conditioned on individual retrieved documents rather than a single combined context. BERAG treats document posterior probabilities as ensemble weights and updates them token by token using Bayes' rule during generation. This approach enables probabilistic re-ranking, parallel memory usage, and clear attribution of document contribution, making it well-suited for large document collections. We evaluate BERAG and BEFT primarily on knowledge-based visual question answering tasks, where models must reason over long, imperfect retrieval lists. The results show substantial improvements over standard RAG, including strong gains on Document Visual Question Answering and multimodal needle-in-a-haystack benchmarks. We also demonstrate that BERAG mitigates the ``lost-in-the-middle'' effect. The document posterior can be used to detect insufficient grounding and trigger deflection, while document pruning enables faster decoding than standard RAG.
Frontier multimodal large language models (MLLMs) can control robots without robot-specific training, but how their capabilities compare and interact with post-trained vision-language-action (VLA) and world action model (WAM) policies across manipulation demands remains to be thoroughly evaluated. In this report, we evaluate GPT-6-Astra on EBench as the core of an agent that controls a robot through an agent harness under the same visual and proprioceptive observation protocol as post-trained policies. We study two agent configurations: direct control, in which the agent generates waypoint commands directly, and harness with policy, in which it can also invoke a post-trained VLA or WAM policy for action execution. Under direct control, the agent achieves competitive overall performance relative to post-trained policies. It leads on short-horizon tasks and exhibits the least performance variation across generalization dimensions, whereas it underperforms post-trained policies on long-horizon tasks, tabletop manipulation, and high-precision contact-rich tasks. In the harness with policy, invoking a post-trained VLA or WAM combines the complementary strengths of both components, raising overall task success to the level of the strongest compared post-trained policies while substantially reducing token cost. However, performance remains below the per-task upper bound of the standalone components due to suboptimal policy invocation, fixed-instruction execution without subgoals, and grasp disruption when the agent resumes control. Overall, our study characterizes the capability boundary of a frontier MLLM agent and suggests directions for improving future agent harness design. Project page: https://internrobotics.github.io/EBench-Astra-Tech-Report/
Zitian Gao, Yilong Chen, Yihao Xiao, Xinyu Yang, Ran Tao, Joey Zhou, Bryan Dai
We present Loopie, the most powerful open-source looped Transformer to date. The Loopie series consists of two Mixture-of-Experts (MoE) models: a 20B-parameter model with 2B active parameters and a 6B parameter model with 0.6B active parameters. Looped Transformers have long faced a challenge: given an N-fold pre-training compute, increasing the parameter count by a factor of N usually outperforms looping N times. Loopie addresses this challenge. Extensive ablation studies, including comparisons with a vanilla 30B-A3B model, show that Loopie substantially outperforms vanilla Transformer baselines trained with the same compute budget. With a novel post-training method, Loopie develops strong reasoning abilities and achieves frontier-level reasoning performance.
As large language models (LLMs) are increasingly integrated into Web search, online recommendation, and intelligent decision-making, the third-party plugin ecosystem centered on model customization continues to grow. Among these, Steering Vectors guide model behavior by intervening in internal representations. However, the open plugin ecosystem also introduces new supply-chain security risks: malicious plugins can disguise themselves as having legitimate functionality and covertly manipulate outputs without significantly affecting model performance. Although this risk poses a real threat, the security issues of Steering plugins in the LLM supply-chain ecosystem still lack systematic research. This paper audits eight mainstream open-source Steering plugin libraries and reveals three structural vulnerabilities: insufficient distribution integrity verification, excessive plugin privileges, and missing runtime monitoring, which offers an attack surface for attackers. To investigate this threat, we propose a malicious plugin, IDH. Specifically, IDH designs a norm-gated mutually exclusive routing mechanism and proposes an information-decay inflection point localization method to automatically select the injection layer, thereby simultaneously achieving benign and malicious functions. To counter such threats, we propose a runtime defense framework, by suppressing anomalous activation interventions and orthogonalizing inter-layer trajectory. We evaluate IDH across five mainstream LLMs, using safety-refusal steering as the legitimate functionality and commercial recommendation manipulation as the hijack objective. Experiments covering six brands across three commercial domains demonstrate that IDH achieves an attack success rate (ASR) of up to 97.8% while preserving high refusal rates on harmful requests.
LLMs on edge devices are constrained by limited memory and bandwidth. Although 1.58-bit ternary weights greatly reduce model storage, long-context inference is increasingly bottlenecked by KV-cache traffic and attention. Existing accelerators also use separate datapaths for ternary linear layers and attention, incurring hardware and execution overhead. We present SPARTA, a sparse ternary LLM accelerator for efficient long-context edge inference. SPARTA integrates: (1) a phase-aware sparse dataflow exploiting heterogeneous sparsity across compute-bound prefill and memory-bound decode; (2) a unified compute fabric mapping ternary SpMM/SpMV and bit-serial attention onto shared processing elements to exploit zero-weight and zero-bit sparsity; and (3) an adaptive two-stage attention engine combining page-level Top-k selection to reduce K-cache accesses with token-level Top-p pruning to reduce V-cache traffic. SPARTA reduces KV traffic by 4.1-4.8x over full attention and 1.3-1.5x over a state-of-the-art sparse-attention scheme, with task-dependent quality loss. Compared with NVIDIA Jetson AGX Orin, SPARTA achieves 3.9x higher generation throughput and 8.3x higher end-to-end energy efficiency on average; versus AMD Ryzen 7 9700X, the gains are 5.0x and 81.7x. It also delivers 1.6-12.0x higher throughput, 2.0-4.2x higher energy efficiency, and 1.3-51.4x higher area efficiency than FIGLUT, TeLLMe, and MCBP.
Long-horizon agents continuously encounter reusable procedural knowledge, yet existing systems typically rely either on model adaptation or external memory, trading persistent learning for rapid reuse. This limitation is especially important for Web agents that repeatedly execute related multi-step workflows. We propose HaMo, a Harness-Model collaborative framework that combines fast non-parametric experience accumulation with slow parametric adaptation. HaMo distills trajectories into strategy-level experience cards for immediate retrieval, while an experience-conditioned continual learning objective trains the model to interpret, ground, and execute retrieved strategies across changing contexts. A retrieval-coverage-based selection mechanism further identifies compact and complementary trajectories for parameter updates. Experiments show that HaMo outperforms harness-only and model-only adaptation under within-domain and cross-domain shifts, improving both retention and adaptation. These results highlight the importance of jointly preserving procedural experience and internalizing the competence to transfer and execute it across tasks.
Junhong Lai, Shuzhong Lai, Yanhao Yu, Wanlin Chen Hangzhou City University, Chenyu Yan Zhejiang University, Haifeng Li Zhejiang University, Lin Yao Zhejiang University, Yueming Wang Zhejiang University
The development of AI-assisted Early Intensive Behavioral Intervention (EIBI) for Autism Spectrum Disorder (ASD) is severely constrained by data scarcity. Furthermore, while Applied Behavior Analysis (ABA) serves as the gold standard for clinical intervention, general-purpose Large Language Models (LLMs) struggle to strictly adhere to its standardized procedures, often resulting in interactions that are linguistically fluent but strategically inconsistent. To address these challenges, we introduce ASDAgent, a strategy-aware framework designed to unify high-fidelity intervention dialogue synthesis and clinical decision support. ASDAgent incorporates two specialized components to solve distinct problems: (i) a DoctorAgent equipped with an Observe-Think-Act-Correct (O-T-A-C) reasoning loop, which resolves the issue of strategy collapse in LLMs by making ABA execution explicit and controllable; and (ii) a ChildAgent that utilizes probabilistic behavior modeling to mitigate data homogeneity, simulating diverse and non-deterministic ASD response patterns. Experiments demonstrate that dialogues generated by ASDAgent closely mirror the strategy distribution of human therapists (KL divergence: 0.083). In real autism intervention, ASDAgent achieves nearly 80% strategic consistency with human experts. Moreover, we show that synthetic data produced by ASDAgent effectively distills professional clinical knowledge into small language models (SLMs), significantly enhancing their therapeutic capabilities.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Yirui QI, Xiaoming Zhang Beijing University of Aeronautics and Astronautics, Ruilin Zeng, Mengyao Liu, Ziyi Zhou, Dezhuang Miao, Bingyu Yan Beijing University of Aeronautics and Astronautics, Zhenyu Guan Beijing University of Aeronautics and Astronautics
Current LLM role-playing systems model persona as a monolithic, static attribute, conflating identity consistency with emotional rigidity. This leads to either robotic repetition or catastrophic persona drift under sustained interaction. We introduce Dynamic Persona Coherence, a framework that decouples Identity-Layer Stability (time-invariant traits) from Adaptive-Layer Appropriateness (history-dependent psychological evolution). We operationalize this through the L/M/S Psychological State Model, which represents persona dynamics across long-term identity, mid-term meaning/stress accumulation, and short-term affect. On top of this state representation, a closed-loop alignment system comprising an automated evaluator (Persona Consistency Critic, PCC), a selective repository (Persona Case Repository, PCR), and a trajectory-adjusting corrector (Persona Drift Suppressor, PDS) enables autonomous coherence repair. Experiments on GPT-4o, Claude-3.5-Sonnet, and DeepSeek-V3.2 demonstrate consistent improvements (+16–84% PCC gains).
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Sungkyu Yang Gwangju Institute of Science and Technology, Kang-Min Kim Kyung Hee University, Mansu Kim Gwangju Institute of Science and Technology
Generating radiology reports from 3D volumetric data remains challenging due to the difficulty of grounding fine-grained pathologies within high-dimensional scans. While retrieval-augmented generation (RAG) offers a potential solution, standard approaches struggle with visual-semantic ambiguity and often introduce irrelevant “normal” context that dilutes pathological signals. To address this limitation, we introduce CPR-RAG, a model-agnostic RAG framework that enhances organ-level grounding by integrating clinical priors into the retrieval process. Specifically, we propose a clinical prior-regularized re-ranking module that leverages corpus-derived co-occurrence statistics to align retrieved candidates with latent disease distributions, ensuring clinical consistency beyond mere visual similarity. Furthermore, we employ clinical relevance context refinement to selectively filter out boilerplate normal descriptions, thereby maximizing the information density of the evidence provided to the generator. Extensive experiments on the RadGenome-ChestCT benchmark demonstrate that CPR-RAG significantly improves clinical efficacy across state-of-the-art radiology report generation models. Human evaluation further confirms that our approach achieves superior factual correctness, completeness, and utility compared to the existing models.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Kedi Chen, Dezhao Ruan East China Normal University, Yuhao Dan, Yaoting Wang Fudan University, Siyu Yan, Xuecheng Wu, Yinqi Zhang Xiaohongshu, Qin Chen East China Normal University 等 15 位作者
Reasoning is an important task for large language models (LLMs). Among all the reasoning paradigms, inductive reasoning is one of the basic types, which is characterized by its particular-to-general thinking process and the non-uniqueness of its answers. The inductive mode is crucial for knowledge generalization and aligns better with human cognition, so it is a fundamental mode of learning, hence attracting increasing interest. Despite the importance of inductive reasoning, there is no systematic summary of it. Therefore, this paper presents the first comprehensive survey of inductive reasoning for LLMs. First, methods for improving inductive reasoning are categorized into three main areas: post-training enhancement, test-time exploration, and data augmentation. Then, current benchmarks of inductive reasoning are summarized, and a unified sandbox-based evaluation approach with the observation coverage metric is derived. Finally, we offer some analyses regarding the source of inductive ability and how simple model architectures and data help with inductive tasks, providing a solid foundation for future research.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表