Chat
-
HexaShell: A Thin Harness for Exploring the Limits of VLM Agents as Robot Coding Agents
2026-10-16 13:56:16Hongcheng Gao, Alex Zhou, Jay Zhu, Zelin Zheng, Shijia Ge, Xiping Li, Di Wu, Yazhe Wang, Jianshu Zeng, Xuan Shangguan, Yongli Wang, Yexing Wan, Lingyu He, Zhiqi Jia, Sihang Wu, Xiao HeAbstract
Agents built on vision language models (VLMs) now compete with learned policies on robot manipulation leaderboards. We ask how far a coding agent can go when it controls a robot instead of a computer, and where it stops. Our system, HexaShell, keeps the harness thin: it only turns the agent's commands into robot motion. The agent drives a robot with two arms through one command line. It sees three RGB cameras and knows their calibration. After each failed attempt, the agent reads the log, finds the cause, and writes or fixes small programs called tools. In this way it keeps what it learns as code, as a coding agent does when it fixes a failing test. The tools may use only what the robot observes. They use triangulation to find where objects are, relative to each other and to the gripper. If two cameras see the same point, the point lies where the two camera rays meet. If a point lies on a known plane, such as the table or the top of a block, it lies where its ray meets that plane. When the head camera is not enough, the robot moves a wrist camera to get a second view. We build the tools on only ten layouts of one seed per task, then freeze them and test them on ten new layouts of the same seed. We use Codex with GPT-6-Astra at medium reasoning effort. On all 42 RoboDojo tasks, HexaShell reaches a success rate of 39.7% and a progress score of 48.0, averaged in the same way as the leaderboard. This is a new state of the art on RoboDojo. The limit is contact, not reasoning: pushing reaches 92.5% and memory tasks 65.0%, but tight insertion only 8.3%. When the agent fails, it usually sees the failure but cannot make the motion work. We open source the tools evolved so far at this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2610.23479
PDF
-
Language Models Learn to Improve Language Models
2026-10-16 11:51:00Minhua Lin, Zijun Wang, Zewen Liu, Zhaotian Weng, Bing He, Zhan Shi, Benoit Dumoulin, Xiang Zhang, Suhang Wang, Hanqing LuAbstract
Large language models (LLMs) are increasingly used to automate AI research, including language-model (LM) development, offering a potential path toward recursive self-improvement. However, turning experimental feedback into a learned ability to propose effective improvements across models faces two challenges. First, training such an ability with real feedback is expensive: reinforcement learning (RL) requires evaluating tens of thousands of proposals, each potentially requiring a full model-training run. Even at ten GPU-hours per run, reward collection alone would consume hundreds of thousands of GPU-hours, motivating a cheaper surrogate. Second, surrogate feedback can become unreliable as policy optimization favors overestimated proposals when the proposal distribution shifts. To address these challenges, we propose LM2LM, which trains LLMs with RL to generate effective LM improvement proposals. To provide scalable feedback, we use a research world model (RWM), a frozen LLM that predicts proposal gains from real experimental records provided in context, as the reward model for RL. To maintain feedback reliability, we periodically execute current-policy proposals ranked highly by the RWM and add their measured outcomes to its context. Across eight autoregressive and diffusion LM development environments, real-execution evaluations on held-out tasks show that LM2LM enables Qwen3.8-27B to outperform GPT-6-sol, even when GPT-6-sol is given 300 real experimental records in context, while requiring 33% lower data-generation cost. LM2LM also generalizes to a model family unseen during training.
Abstract (translated)
URL
https://arxiv.org/abs/2610.23344
PDF
-
LAW: Unifying Language, Action, and World Modeling for Agentic Physical Intelligence
2026-10-16 11:39:21Jiajun Cao, Kai Tang, Bohua Zhang, Zezhong Qian, Ying Li, Xiaozhu Ju, Xiaobao Wei, Jiaming Liu, Yang Ding, Sixiang Chen, Yiding Sun, Siyuan Qian, Tinghao Wang, Jixian Wu, Yichen Guo, Yi Zhang, Shilong Zou, Yong Dai, Jian Tang, Shanghang ZhangAbstract
World-action models jointly predict actions and future world states, but long-horizon manipulation further requires language reasoning to plan subtask objectives, assess execution progress, and guide action selection. We present LAW, a closed-loop VLM--WAM framework that unifies language reasoning, action generation, and world modeling through semantic subtask progress. Within the LAW agent, a hybrid-expert vision-language model plans the remaining subtasks, estimates the values of and fuses actions proposed by a world-action model, and asynchronously assesses execution status to support continued execution and subtask transitions. Within the world-action model, Mixture-of-Transformer Experts integrates modality-specific and shared computation for joint video--action modeling. Dual-stream hierarchical visual memory maintains temporal context for both models, while subtask-local value learning aligns action evaluation with semantic objectives and preserves supervision from successful prefixes of failed trajectories. LAW achieves 94.21% average success across 50 RoboTwin tasks. Replacing the WAM backbone with OpenWAM-$alpha$ further raises the average success rate to 95.56%, establishing a new open-source state of the art on RoboTwin. Despite having only 2B parameters, our VLM also outperforms variants using Embodied-R1.5 and GPT-5.5 High. On three long-horizon real-robot tasks, LAW achieves 65% average success, compared with 28.33% for $pi_{0.5}$ and 25% for Fast-WAM.
Abstract (translated)
URL
https://arxiv.org/abs/2610.23334
PDF
-
Can AI interview? Comparing AI-led to Human-led Chat Interviews
2026-10-16 08:43:18Semra Guven, Tobias G{\aa}rdhus, Andreas Bjerre-Nielsen, Anna Rogers, Hjalmar Bang CarlsenAbstract
Large language model chatbots have been proposed as AI-interviewers for collecting qualitative interview data at scale, potentially bridging social science's longstanding divide between the breadth of surveys and the depth of qualitative interviews. This paper provides the first direct empirical comparison of data quality in AI-led and human-led interviews. In a randomized experiment (N = 41), university students were interviewed via synchronous text chat by either trained non-expert interviewers or a locally hosted AI system; we evaluate exposure, specificity, and relevance. Median interview duration was similar across conditions, but AI-interviewers asked more than twice as many questions per interview and elicited roughly 65 percent more respondent text per interview. Per-question responses were longer with human interviewers, though not after multiple-testing correction; only AI-condition participants skipped questions. Responses were almost uniformly on-topic in both conditions; for specificity we detect no difference, but the design cannot exclude one of up to three quarters of a point. These results were obtained with a small, locally hosted open-weight model (7B parameters), which addresses concerns about data security, GDPR compliance, and scientific reproducibility. Our findings support the viability of AI-interviewing for large-N qualitative data collection.
Abstract (translated)
URL
https://arxiv.org/abs/2610.23149
PDF
-
From GAN-Based Models to GPT-5: A Taxonomy-Guided Study of Human Performance in Detecting Deepfake Face Images
2026-10-16 08:41:12Zhiqing Guo, Lixin Jia, Luning Sun, Xiaowei Chen, Guofeng Yu, Liejun Wang, Gaobo Yang, David Stillwell, Thorsten Holz, Shujun LiAbstract
Generative models for deepfakes have rapidly evolved from GAN-based architectures to diffusion models and multimodal large language models (MLLMs). However, empirical research on human performance in detecting deepfake face images has not kept pace. It remains unclear if (and how) detection performance varies across generative models and whether model-specific training improves detection more effectively than general-purpose training. In this paper, we propose the \emph{Deepfake Face Artifact Taxonomy} (DFAT), a cross-model taxonomy of human-recognizable visual artifacts in deepfake face images, based on which we build both general-purpose and model-specific training materials. Through an online experiment ($n = 524$), we examined human performance in detecting deepfake face images and evaluated training efficacy across five representative generative models: three GAN-based, one diffusion-based, and one MLLM (GPT-5). Our results show that participants' detection accuracy was strongly model-dependent and that images generated by more recent and advanced models were not necessarily harder to detect. After training, participants achieved significant performance gains across most conditions, and their improvement varied by model. Notably, model-specific training generally yielded greater numerical gains than general-purpose training. In a five-month follow-up test ($n = 342$), we observed no significant decline in detection accuracy and no significant increase in perceived stress. Overall, our findings provide actionable insights for designing effective, sustainable, and targeted training programs to counter the growing threat posed by deepfakes.
Abstract (translated)
URL
https://arxiv.org/abs/2610.23144
PDF
-
WebCoder: Scaling General Frontend Development with Artifact-Grounded Agentic Reinforcement Learning
2026-10-16 08:17:10Zijun Dou, Zhen Yang, Xijun Liu, Dongzhuoran Zhou, Aohan Zeng, Jie TangAbstract
Real-world frontend development requires agents to compose reproduction, generation, modification, and repair over an evolving application state. Learning a general policy for this process requires translating heterogeneous execution feedback into task-aligned rewards and effective credit assignment across long sequences of interdependent actions. We introduce WebCoder, a general frontend development agent trained with artifact-grounded agentic reinforcement learning. Our key idea is to assess diverse development activities through their effects on a shared executable artifact and its satisfaction of task requirements. We build Webdev-World, an executable environment that integrates code editing, application execution, browser inspection, and on-demand skills while preserving project state across interactions. An independent verifier evaluates the final application against task requirements to provide terminal rewards, while changes in intermediate artifact progress redistribute the resulting rollout advantage across development segments. Context compaction and message-tree sample construction support training on trajectories beyond a single context window while avoiding repeated loss computation on shared generated history. We further introduce Webdev-Bench to evaluate composed development requirements within a single project. On Webdev-Bench, WebCoder-27B raises requirement satisfaction from 64.8 for its Qwen3.8-27B base model to 72.8, exceeding Claude-Opus-4.8 (65.1) and GPT-5.4 (59.6); on Vision2Web, it improves the average visual and functional score from 41.1 to 62.4, on par with Claude-Opus-4.8. Ablation studies confirm the contributions of artifact-grounded credit assignment and long-horizon training.
Abstract (translated)
URL
https://arxiv.org/abs/2610.23104
PDF
-
ParetoChat: An LLM-Based Interface for Interactive Multi-Objective Optimization
2026-10-16 07:39:39Gokhan CeyhanAbstract
Multi-objective optimization models are often deployed in settings where decision makers (DMs) must balance some conflicting criteria. In practice, however, stakeholders must translate their qualitative preferences between different criteria into numerical weights, priorities, or reference points required by optimization methods, which can be difficult to specify and interpret. This paper proposes \emph{ParetoChat}, an LLM (large language model)-based expert-agent interface for interactive multi-objective optimization. The agent translates natural-language user input into structured scalarization parameters, invokes an optimization system, presents candidate solutions, and iteratively elicits feedback. We describe a multi-agent architecture that separates input processing, multi-criteria decision-making (MCDM) expertise, solution presentation, and feedback processing. We then present a simulation study using an LLM-based DM agent and benchmark ParetoChat against interactive methods implemented in DESDEO, including reference-point and NIMBUS methods. The computational experiments on multi-objective binary knapsack and set-partitioning instances show that ParetoChat can guide the search efficiently toward solutions that match the DM's preferences. An application to historical courier-order assignment instances illustrates how users can explore objective extremes and intermediate trade-offs through natural-language requests. The results indicate that LLM agents may provide a practical interaction layer between business stakeholders and optimization tools.
Abstract (translated)
URL
https://arxiv.org/abs/2610.23062
PDF
-
PersonaCUABench: Benchmarking Computer-Use Agents Serving Diverse Personas
2026-10-16 06:14:55Haoming Li, Defu Cao, Xu Hu, Yingtao Luo, Yiyang Wang, Xianghe Xu, Jiashu He, Wanting Jiang, Yifan Zhang, Yunhan Zhao, Siyang Wu, Chen Yang, Haixin Wang, Zheyuan Deng, Taoran Li, Jingwen Zhang, Zhejian Peng, Shushu Liang, Yue Zhao, Bingzhe Li, Feng Chen, Yifan Wang, Jintao Huang, Yijun Wang, Yuexing Hao, Xiaomin LiAbstract
Computer-use agents (CUAs) are commonly evaluated against fixed task specifications, but users may omit requirements, add constraints, or require approval before consequential actions. Completing the benchmark instruction does not guarantee serving the person who asked. We introduce PersonaCUABench, which holds 74 tasks across 11 domains fixed and varies only who asks. A persona agent, playing one of 44 personas in four groups, rewrites each instruction in that user's own words and answers the CUA's questions, while the CUA sees only the dialogue, never the profile or the original instruction. Across nine backbones (3,330 runs), mean task completion falls by 36% relative to the original requests. The losses have two sources: in 59.9% of cells that lose points, the agent engages with what the user asked for and still gets it wrong, and 11.2% are stops the user requested that the checker scores as failures. Even when a persona request is fully completed, the user gives the top satisfaction rating in only 62.7% of cells, against 84.1% for the original requests. CUAs should therefore be evaluated not only on task completion but also on whether they meet the requirements and constraints users express. Code: this https URL. Project page: this https URL
Abstract (translated)
URL
https://arxiv.org/abs/2610.22990
PDF
-
ANCHOR: Audio-Visually Grounded Chain-of-Thought Reasoning Benchmark
2026-10-16 05:38:11Joel Julin, Souraja Kundu, Liza Dahiya, George Z. Wei, Liuyue Xie, Jingru Yi, Xu Zhang, Hao Yang, Laszlo A. JeniAbstract
Multimodal large language models (MLLMs) perform well on audio-visual question answering, but correct answers do not necessarily imply that their stated reasoning is grounded in relevant perceptual evidence. Existing benchmarks mainly evaluate final-answer accuracy, offering limited insight into whether model-generated rationales correctly identify visual regions, temporal intervals, and audio events supporting predictions. We introduce ANCHOR, a benchmark and evaluation framework for measuring grounded-rationale alignment in multimodal reasoning, comprising 2,319 videos and 8,272 question-answer pairs with structured reasoning traces, temporal evidence, spatial bounding boxes, and salient audio cues, including a human-refined test set of 224 videos and 975 question-answer pairs. We separately measure answer correctness, hallucination, logical divergence, and conciseness, with direct spatial and temporal grounding measurements on a focused subset. Across the full benchmark, twelve contemporary MLLMs show a substantial alignment gap: models with comparable answer performance can differ markedly in whether their explanations are supported by annotated evidence. On a matched subset of 100 videos and 446 question-answer pairs, rankings remain highly stable across human-refined, Gemini-generated, and GPT-generated references (mean Pearson r=0.99), providing evidence against model-reference circularity within this subset. Grounding-cue ablations of three MLLMs on the full human-refined test set reveal a consistent modality hierarchy: removing video reduces accuracy by 38.4-41.8 points, while removing bounding boxes (3.2-7.7 points) and audio (4.2-5.6 points) causes smaller, complementary, model-dependent drops. These findings show answer-only evaluation obscures important grounding failures and position ANCHOR as a testbed for evaluating accurate and evidence-aligned audio-visual reasoning.
Abstract (translated)
URL
https://arxiv.org/abs/2610.22964
PDF
-
What Shapes Brain Alignment of Large Language Models during Dialogue? Insights from Language Production and Comprehension
2026-10-16 05:34:33Yuko Nakagi, Yuya Chiba, Yasue KishinoAbstract
Brain alignment of large language models (LLMs) offers a way to investigate representational similarities between LLMs and the human brain during language processing. While previous studies have mainly focused on passive language comprehension, little is known about such alignment during dialogue, in which speakers engage in both language production and comprehension. In this study, using fMRI data collected during dialogue, we examine brain alignment during language production and comprehension across LLM layers and brain regions. We assess its association with the intrinsic dimension (ID) of LLM representations and the contributions of several interpretable linguistic and communicative features. Our results reveal distinct patterns of brain alignment between production and comprehension, as well as strong associations with ID in brain regions associated with each processing mode. Furthermore, we show that the contributions of linguistic and communicative features also vary across processing modes, LLM layers, and brain regions.
Abstract (translated)
URL
https://arxiv.org/abs/2610.22961
PDF
-
Many Biases, One Circuit: Post-Training Tunes a Pretrained Cue-Routing Mechanism
2026-10-16 05:00:12Prakhar Gupta, Terry Jingchen Zhang, Zhijing JinAbstract
Language models often abandon a correct answer when the prompt contains a cue, a hint toward a wrong option such as "I think the answer is B." Sycophancy and related biases show this failure. They are often studied separately and blamed on alignment. We test this view causally, tracing the attention heads behind these biases through the stages of two post-training pipelines. In multiple-choice tests on most models, four cue types rely largely on the same few heads, a shared circuit already present in pretrained models. Most base models follow cues in plain text. Fine-tuning a base model on about 200 generic assistant examples without cues activates these heads in chat format, so cue-following can be a side effect of assistant tuning. The final reinforcement learning (RL) stage strengthens the effect of cues in almost every comparison. In Tulu-3-8B and OLMo-2-7B, we swap the attention patterns of six middle layers between the RL checkpoint and its predecessor. For suggested-answer cues, this shifts each checkpoint's cue effect most of the way to the other's level. Post-training thus seems to change when and how strongly models use this circuit. Mitigation for these multiple-choice biases could target this circuit.
Abstract (translated)
URL
https://arxiv.org/abs/2610.22937
PDF
-
ASR-Style Transcript Normalization Removes Useful Pragmatic Signal for Dialogue Act Classification
2026-10-16 04:51:37Lucky Onyekwelu-Udoka, Md Shafiqul IslamAbstract
Automatic speech recognition (ASR) pipelines often normalize transcripts by removing filled pauses and punctuation and expanding colloquial forms, potentially discarding cues useful for dialogue act classification. We evaluate nine transcript conditions across six classifiers (TF-IDF logistic regression, RoBERTa, BERT, ALBERT, GPT-2, and BiLSTM) and three conversational corpora (Switchboard, ICSI MRDA, and SLUE-HVB), supplemented by multi-seed experiments, normalization controls, and real-ASR evaluations. Across five random seeds, heavy normalization reduces Macro-F1 in all 15 neural model-corpus combinations under the original preprocessing protocols. Removal using the original filler and backchannel lexicon consistently degrades performance on MRDA and SLUE-HVB but produces mixed effects on Switchboard. Punctuation-aware TF-IDF experiments show that punctuation removal reduces Macro-F1 by 0.036 on Switchboard and 0.078 on MRDA, demonstrating that the original linear baseline's apparent insensitivity resulted from tokenization. Restricting filler removal to um, uh, and erm preserves degradation on MRDA but largely eliminates the SLUE-HVB effect, where the original lexicon removes class-defining backchannel expressions. These findings demonstrate that normalization effects depend on preprocessing definitions, feature extraction, and conversational context, supporting task-dependent rather than indiscriminate transcript normalization.
Abstract (translated)
URL
https://arxiv.org/abs/2610.22934
PDF
-
Competence-Based Evaluation of Logical Reasoning in Large Language Models
2026-10-16 03:52:17Jize Jiang, Chengxiang ZhaiAbstract
Recent advances in large language models (LLMs) have brought impressive performance across diverse tasks; however, shortcomings remain in logical reasoning, including inconsistencies under linguistic variation. To improve the reasoning capacity of LLMs, we must be able to accurately measure their reasoning capacity. Current benchmarks primarily report single accuracy numbers on selected reasoning tasks. Such a performance-based evaluation methodology says little about a model's generalization capacity in reasoning. To address this limitation, we argue for a different evaluation paradigm, called competence-based evaluation (CBE), which tests whether models reliably apply the same competence across a family of problems. A parameterized generator scales complexity while keeping the required competence fixed, and matched answer-preserving presentations test correctness beyond one formulation. We instantiate CBE on elementary ordering tasks, scaling input size while varying presentations of the same ordering rule. Across six GPT and Qwen models, performance is reasonably good over bounded complexity ranges, but even strong models fail the competence condition across the tested family. Chain-of-thought and GRPO extend strong performance without resolving all tested failures; supervised LoRA shows little held-out transfer. CBE provides diagnostic profiles and an evaluation protocol that can grow with future models.
Abstract (translated)
URL
https://arxiv.org/abs/2610.22894
PDF
-
Can a Frontier VLM Serve as a Robot Policy for Memory-Dependent Manipulation?
2026-10-16 02:37:38Jack B. Jedlicki, Amish Sethi, Heng YangAbstract
We use a frozen frontier vision-language model, GPT-6-Astra, directly as the robot policy on RoboMME, a benchmark of 16 memory-dependent manipulation tasks spanning temporal, spatial, object, and procedural memory. The model keeps the whole episode in context and outputs every end-effector waypoint and gripper command, with no RoboMME-specific training. On the 800 test episodes it succeeds in 92.1%, above the previously reported human score of 90.5%, which was obtained with an oracle skill interface, and far ahead of the best leaderboard entry (60.8%) and of a hierarchical system that pairs the same model with a RoboMME-trained policy (79.1%). It matches or exceeds the human score on 13 of the 16 tasks. Its 63 failures concentrate in two tasks and are mostly premature completions and mistimed or unintended button presses. These results blur the usual division between long-horizon planning, memory and end-effector control, and raise the question of how much of this stack a single general-purpose model can take over in embodied systems. Project website: this https URL. Code, prompts and all episode logs: this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2610.17350
PDF
-
STRIDE: Evaluating Spatiotemporal Reasoning in Driving Edge Cases
2026-10-16 01:13:23Lieqi Liu, Rui gao, Jia-Chen Gu, Wenbo Hu, Zhaobin Mo, Ahmadreza Moradipari, Nejib Ammar, Wei Wang, Nanyun PengAbstract
In autonomous driving, perceiving surrounding objects is not enough: failing to understand how they move and interact over time can lead to dangerous decisions. Evaluating vision-language models (VLMs) therefore requires testing whether they connect past spatial observations to future consequences. Yet existing benchmarks often assess edge cases through basic perception questions or test temporal understanding in simple scenarios, leaving complex spatiotemporal reasoning underexplored. We introduce STRIDE (SpatioTemporal Reasoning In Driving Edge Cases), an object-centric benchmark for long-tail, safety-critical driving scenarios. STRIDE covers six task families: spatial perception, spatial understanding, temporal memory, temporal extrapolation, trajectory prediction, and scene-context awareness. It pairs multiple-choice questions probing individual spatiotemporal judgments with related free-form visual question answering (VQA) requiring detailed scene explanations and justifications of driving decisions. We prioritize scenes with unusual object behaviors and complex traffic interactions, generate reference answers from structured dataset annotations through automated procedures, and use limited language-model assistance and human verification to ensure annotation quality. Across 12 proprietary, open-weight, and driving-specialized models, even the best performer, GPT-6-Astra, achieves only 43.4% average multiple-choice accuracy, versus an approximately 20% random-choice baseline. Further analysis reveals that strong performance in describing driving scenes does not necessarily translate to accurate safety-critical judgments, and that models struggle more with reconstructing object motion than recalling past positions. STRIDE thus exposes spatiotemporal driving limitations that aggregate VQA scores can conceal.
Abstract (translated)
URL
https://arxiv.org/abs/2610.22795
PDF
-
Spatial Memory by Agents, for Agents
2026-10-15 23:13:47Qianwei Wang, Ce Zhang, Zhanpeng Luo, Pengpeng Su, Shuo Yang, Hao XuAbstract
Spatial memory enables robots to reuse past observations for localization, navigation, and question answering. Designing such memory involves jointly deciding what to retain and how to retrieve it for diverse semantic, metric, and temporal needs. We investigate whether coding agents can automate this design process through iterative implementation and evaluation. Under a fixed development budget, a Designer Agent writes and revises code for memory construction and retrieval using feedback from a fixed Query Agent. The resulting implementations are frozen before building memories from held-out observation histories, without access to downstream questions, and evaluated across five complementary task tracks using the same Query Agent. Across three independent design searches, Codex / GPT-6 Astra averages 44.23 on the equally weighted five-track score, versus 29.32 for ReMEmbR, the strongest evaluated baseline supporting all five tracks, a 50.9% relative improvement. Qualitative analysis relates performance differences to evidence retention and retrieval, while identifying grounding and evidence-use failures. These results demonstrate the potential of coding agents to automate spatial memory design, while revealing task-dependent trade-offs in performance and computational cost. Project page: this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2610.22736
PDF
-
GlobeBench: Evaluating Text-Based World Model Capabilities for Scalable Environment Simulations
2026-10-15 22:24:56Nicholas Saban, Charvi Bannur, Mariya Vasileva, Darshan DeshpandeAbstract
Text-based world models have recently shown potential to simulate complex RL environment state transitions for a diverse set of use cases. However, no standardized benchmark exists to compare, study and improve these models. Through this paper, we attempt to solve this problem by building GlobeBench, a novel benchmark with 1,632 tasks across five separate domains and nineteen simulation skills. The tasks check whether models can predict the environment's response to a specific action given the full agent trajectory by understanding earlier state changes, following environment rules, and predicting the effects of actions on both related and unrelated objects. We compare the predicted environment response with an execution-derived reference, using a hybrid architecture with both deterministic gates and a semantic judge. Among 20 general-purpose LLMs and specialized text world models, GPT-6 Astra achieves the highest task success rate at 49.57%. We find that model accuracy decreases as actions affect a larger portion of the environment. Specifically, when the agent's changes must propagate to dependent objects, task success drops to an average of 5.1% across all 20 models. Although model strengths differ across tasks, 631 tasks remain unsolved by any model, suggesting a significant gap in general LLM capabilities as a world model.
Abstract (translated)
URL
https://arxiv.org/abs/2610.22708
PDF
-
Multimodal Filesystem Memory For Long-Horizon Embodied Agents
2026-10-15 21:56:00Chloe Su, Jiahan Li, Zhenting Qi, Chaoran Cheng, Haoyu Chen, Sejoon Jun, Weirui Ye, Yilun DuAbstract
Agents operating over long horizons need an effective memory that allows them to reuse their experience throughout long interactions. If the entire history of an interaction is retained, it becomes difficult to find the relevant experience, while compressing it into summaries may result in the loss of visual and spatial details that are required later on. We present \textbf{MFM}, a multimodal filesystem memory architecture that keeps complementary types of agent memory. \textbf{Episodic memory} stores the observations, actions, and outcomes, and \textbf{spatial memory} associates experiences with locations. \textbf{Semantic and procedural memory} build up general knowledge and reusable methods of acting, and \textbf{working memory} holds the current plans. MFM makes use of searchable textual records to connect these different representations to the underlying multimodal episodes. The agents are able to find previous experiences by means of contextual cues and to examine the original observations in order to recover details that are missing from their textual descriptions. Through consolidation, the agents convert experience into higher-level knowledge and procedures while still keeping references to the original episodes. In the case of long-horizon Minecraft and computer-use tasks, MFM leads to improved performance with three base models. With GPT-6-Astra, MFM completes \emph{Mine Diamond from Scratch} and reaches 16 out of 24 milestones on the \emph{Ender Dragon} task, as against 12.7 for the best baseline. On CUABench-KiCad it attains a mean reward of 0.440, as compared to 0.400 for the best baseline and 0.267 in the case of no persistent memory.
Abstract (translated)
URL
https://arxiv.org/abs/2610.22686
PDF
-
ProPerAgent: Toward Memory-Driven Proactive Assistance
2026-10-15 21:40:17Shenglai Zeng, Xiao Yang, Tiffany Ma, Xinyuan Zhang, Nanshu Wang, Deren Lei, Kai Sun, Xilun Chen, Tammy Stark, Xianhui Zhu, Yuetai Li, Rulin Shao, Tianxin Wei, Rakesh Wanga, Anuj Kumar, Jiliang Tang, Wen-tau Yih, Xin Luna DongAbstract
As assistants move from reactive answering toward proactive intervention, memory changes role: it no longer only supplies the answer to a query, it determines whether and when to speak at all. Prior work in this setting hands the system the relevant fact in context; we study the regime that makes the problem hard, where the evidence justifying an intervention must be mined from a long-horizon personal memory. We introduce ProPerMem, a benchmark for memory-driven proactive assistance (MPA), pairing egocentric video with year-long, distractor-rich user memories (~618K tokens on average), human-verified trigger windows, and a protocol that scores content and timing jointly. Every baseline we evaluate performs poorly (at most 15%), including frontier models given summarized or retrieval-augmented memory. A controlled study explains why: with oracle evidence delivered inside the trigger window GPT-5.3-Chat reaches 96%; letting it choose the timing drops it to 38%, and replacing oracle evidence with retrieval drops it to 15%. Content and timing are separately necessary, and missing either is catastrophic. Guided by this, we build ProPerAgent (Proactive Personalized Agent) with a plug-and-play harness that surfaces memory from a graph-organized store and gates it down to what is worth interrupting for, plus a lightweight RL stage that curbs early and repeated firing. Without any training, ProPerAgent lifts strict accuracy from below 10% to 37.5% on Qwen3.8-27B and 38% on GPT-5.3-Chat; RL post-training reaches 42.4%.
Abstract (translated)
URL
https://arxiv.org/abs/2610.22671
PDF
-
ToMStories: Interpreting Theory-of-Mind in Synthetic Narratives
2026-10-15 20:38:58Melanie Subbiah, Chia-En Jacklyn Tsai, Kathleen McKeownAbstract
Understanding people often requires inferring what they feel or want from how they behave rather than from what they state outright. We ask whether language models can make these inferences, and how they do so. We introduce Theory-of-Mind Stories (ToMStories), an evaluation dataset of ~6,000 short synthetically generated narrative vignettes, each paired with a multiple-choice or open-ended question about a character's emotion, fear, motivation, or emotional wound. These psychological states are conveyed implicitly (showing), explicitly (telling), or with conflict between dialogue and description. Evaluating thirteen open-weight models (1B-72B parameters), we find that the largest models match humans on multiple choice but not open-ended questions. Our analyses point to simplifications models make in both interpretation and generation of implicit psychological states, indicating that synthetic data can help us understand how models perform interpretation, but likely overestimates how well models interpret humans.
Abstract (translated)
URL
https://arxiv.org/abs/2610.22632
PDF
