Agent 进展

Recent work on Thinking with Images makes vision a dynamic part of reasoning, but does so through generation: the model invokes external tools, synthesizes code, or imagines new imagery, each at the cost of a tool protocol, brittle code, or an expensive training pipeline. A fourth route makes vision dynamic without generating anything, by retrieving labeled exemplar images and reasoning over them, yet it remains underexplored despite being train-free. We present ReVisIT, a train-free framework that realizes this retrieval-based route by treating each retrieved image-label pair as a unit of visual thought. ReVisIT combines structured class definitions, per-query multimodal retrieval of exemplars, and alternating user/assistant injection of those exemplars before joint multi-attribute decoding, and degrades gracefully to whichever components a task admits. On VL-ICL Bench Fast Open MiniImageNet, Qwen3-VL-30B-A3B with ReVisIT reaches 98.5% at 4-shot, statistically indistinguishable from the 72B LLaVA-OneVision SOTA (98.7%) on this near-saturated task at about 1/2.4 the parameters, while the same backbone without the scaffold sits at chance. The turns layer alone adds 26.1 points to GPT-4.1 on free-form concept induction (Bongard-OpenWorld), and the full stack yields a 4-6 point macro gain across three backbones on MAAC-Bench, a new license-clean 27-class, 5-attribute benchmark, significant by paired bootstrap on the curator-derived attributes. Component analysis shows that retrieval-plus-turns is the universal lever while structured definitions are need-adaptive, and that 83% of the retrieval gain comes from retrieval quality rather than from the presence of exemplars. MAAC-Bench is released with a rubric-grounded LLM verification protocol that replaces author spot-check on subjective attributes.

Mechanistic modeling via ordinary differential equations (ODEs) provides interpretable descriptions of complex dynamics and enables inference of underlying mechanisms, which is particularly valuable in clinical settings. However, in rare diseases, both the structure and parameters of the model are typically unknown, while individual-level data is scarce, noisy, heterogeneous, and subject to privacy constraints. In such settings, population-level summary statistics provide a practical privacy-preserving data representation, while capturing heterogeneity further requires modeling parameters as distributions rather than fixed values. Yet no existing method jointly discovers ODE structure and refines parameter distributions solely from summary statistics. We present AgentODE, an end-to-end framework that addresses this gap. An LLM proposes candidate ODE structures, while a tool-augmented inference agent iteratively refines parameter distributions through a diagnosis--update loop, operating on population-level summary statistics alone. We evaluate AgentODE on three benchmark problems across different fields and two clinical datasets, including the rare disease recessive dystrophic epidermolysis bullosa (RDEB), with only 231 observations across 46 patients. AgentODE recovers functionally consistent ODE structures across all settings, and experiments on RDEB demonstrates that in sparse and noisy data settings reasoning from summary statistics promotes mechanistically principled structure discovery, whereas baselines with individual-level data access recover implausible structures despite better predictive performance. AgentODE opens new possibilities for mechanistic modeling of rare diseases directly from population-level summary statistics, where data scarcity and privacy constraints have traditionally limited such analyses.

Large language model (LLM)-based software engineering agents are increasingly developed to resolve software issues by generating patches from issue reports and code repositories. Bug reproduction tests (BRTs) are an important building block for such agents and have been shown useful for patch validation. However, it remains unclear whether BRTs can also help the more central stage of patch generation. We first conduct a preliminary study and find that directly using advanced BRT generators to guide patch generation is not beneficial: fail-to-fail BRTs can mislead agents, while even fail-to-pass BRTs bring limited or negative gains. Our analysis reveals two reasons: fail-to-pass BRTs may cover only one manifestation of the reported issue, leading to partial patches, whereas fail-to-fail BRTs are unreliable as direct patch-generation targets. Motivated by these insights, we propose SWE-Doctor, a software issue resolution agent that guides patch generation with runtime diagnoses derived from multi-faceted BRT executions. SWE-Doctor first generates multi-faceted BRTs for different behavioral requirements stated in the issue, then executes and debugs these BRTs to construct runtime-grounded diagnosis records, and finally uses the diagnoses together with localization information inferred during BRT generation to guide patch generation and reduce partial patches. We evaluate SWE-Doctor on Python bug-fixing issues from the widely adopted SWE-bench Verified and SWE-bench Pro across five LLM backends. SWE-Doctor consistently outperforms existing agents across all 10 LLM-benchmark combinations, achieving average resolution rates of 75.7% on SWE-bench Verified and 59.4% on SWE-bench Pro. In particular, on the more challenging SWE-bench Pro, SWE-Doctor improves the average resolution rate by 8.0-8.9 percentage points over the baseline agents.

Computer-assisted synthesis planning breaks target molecules into accessible precursors using large libraries of reaction rules that assign each transformation a deterministic, interpretable label. But chemistry is long-tailed, making manual encoding intractable, and existing tools rely on fixed rulesets that cannot adapt to new chemistries. Here we present a fully automated pipeline in which a multi-agent framework of large language models (LLMs) classifies reactions and writes the rules themselves across 665,901 US patent reactions, generating each rule under a verification loop that tests it against the corpus. It expands a standard taxonomy from 68 to 14,073 classes without human curation. With a lightweight fingerprint classifier, it classifies 97.7% of unseen reactions, matching a leading proprietary classifier while resolving chemistry more finely and extending on demand to chemistry outside its training distribution. The result is a living reactivity database and a general route to turning generative models into reliable, self-expanding symbolic systems.

Real-world e-commerce image editing often requires multiple, localized, and auditable operations rather than global restyling. This compositional nature poses a dual challenge: models must precisely apply all requested edits to the correct regions while preserving unmodified content, even under ambiguous instructions. Existing one-shot editors conflate intent resolution, spatial grounding, and synthesis into a single step, frequently resulting in partial execution failures, which is unacceptable for commercial scenarios. To address this, we introduce GMO-E$^2$DIT, an agentic editing framework that couples a Vision-Language Model (VLM) with a mask-conditioned image editor to tackle structured multi-turn task completion. Given an underspecified instruction, the VLM agent constructs a region-grounded edit agenda, effectively decoupling cognitive reasoning from generative rendering. The framework then executes sub-programs via operation-aware masks and references, utilizing a reflection-driven loop to inspect intermediate results and determine the subsequent state. This iterative mechanism reliably preserves safe partial progress, retries unfinished operations, and recovers from errors. Furthermore, we develop a unified data pipeline providing aligned supervision for planning, execution, and reflection, alongside EComEditBench, a comprehensive benchmark for instruction-driven evaluation. Extensive experiments demonstrate that GMO-E$^2$DIT achieves competitive performance compared to strong closed-source models, yielding superior instruction accuracy and edit fidelity over existing baselines.

Quantum computing is increasingly explored for software engineering (SE) optimization, but translating natural-language (NL) task-level requirements into executable quantum applications still demands substantial quantum and programming expertise. We present QPipe, a large language model (LLM)-based multi-agent architecture that autonomously turns NL requirements into traceable quantum-application workflows through specialized agents for requirement parsing, formulation, code generation, review, execution, and verification. We evaluate QPipe on 20 NL requirements, each associated with a real-world benchmark and a test-optimization problem. QPipe successfully completes the key stages of quantum-application generation across requirements, achieving average rates of 100% for code compilation and 96.7% for application execution and final-result combination, with average generation costs of 260.1 seconds and 1.89M tokens per requirement. Among the generated quantum applications that execute successfully, the returned solutions outperform the offline genetic algorithm baseline in most cases. Ablation results further show that QPipe's advantage depends on retaining code-generation skills, task knowledge, review feedback, and multi-agent decomposition. These results indicate that agentic coordination can support generation of executable quantum applications for tackling test optimization problems from real-world benchmarks.

Large language model (LLM)-based conversational agents (CAs) are now ubiquitous, creating new opportunities for AI-mediated behavior change. Their capacity to project nuanced personalities and adopt diverse metaphorical roles raises a design question: how should an agent's persona and personality be calibrated to the moment? Recent evidence suggests that (i) moderate personality expression outperforms low or high extremes on trust, enjoyment, and intention to adopt in goal-oriented tasks, and (ii) context-appropriate metaphors outperform static one-note assistants on user experience and uptake. Yet most CAs still fix both persona and style, risking misalignment when dynamics, urgency, and formality vary, for example in medical information seeking, fitness coaching, and reflective learning. We propose a Fluid Personality Framework that jointly adapts (1) the agent's metaphorical persona, such as coach, tutor, librarian, or tool, and (2) its personality expression intensity, low, medium, or high, as a function of task context, user goals and traits, and situational urgency. We sketch the framework and its core design dimensions.

Assistive autonomous robots operating in multi-agent environments require efficient strategies to locate specific individuals among multiple residents. Current social navigation methods focus on reactive collision avoidance and trajectory adaptation, but lack mechanisms to proactively gather information through human-robot communication. We introduce Communication-enabled Social Navigation (CommNav). In this novel task, robotic agents actively seek assistance from residents to locate target individuals by requesting information about recent sightings, locations, and movements. To evaluate CommNav, we extend Habitat 3.0 to create Habitat 3.0c, a communication-enabled variant supporting multi-human environments with information exchange protocols. Adding our communication module (COMM) to a state-of-the-art social navigation model yields a 10 percentage-point improvement in Episode Success. We further investigate the transition from structured data to natural language by evaluating models trained on LLM-generated instructions and on colloquial instructions collected from a human study. Our experiments reveal that: (i) explicit human-robot communication substantially enhances multi-person navigation performance; (ii) pre-training COMM on a communication pretext task effectively addresses the challenge of occasional interaction signals; and (iii) the navigation policy is highly robust to natural, colloquial human language, achieving an episode success statistically similar to the model using perfect structured data.

LLM agents are rapidly being deployed in production, including coding assistants, customer-support chatbots, and scientific research assistants, yet they remain fundamentally static in enterprise deployment. The LLM weights, system prompts, tool repertoires, and in-context harnesses are frozen at deployment time, and any improvement requires a manual loop of human-curated data collection, offline fine-tuning, modification of the agentic paradigm, and re-deployment. Recent work on self-evolving agents, such as OpenClaw for individual users, indicates that the next leap in agent capability will come from agents that continually learn from their own experience. In this paper, we argue that this vision for self-evolving agent deployment is being held back for enterprise-level large-scale agentic service not by reinforcement learning (RL) algorithms but by agentic online RL systems. Specifically, current agentic RL systems and the surrounding observability software stack are inadequate along three essential aspects: (i) there is no standardized agent trajectory data protocol capable of carrying RL learning signals at step granularity across heterogeneous agent paradigms; (ii) there is no enterprise-grade comprehensive data proxy that converts real workloads into governed learning substrates; and (iii) there is no unified agent evolution control plane that automatically decides, based on trajectory statistics, when to update policy weights or evolve the in-context harness. The next generation of agentic RL systems must be co-designed around these three pillars, and we sketch concrete architectures, case studies, and counter-arguments. We instantiate one branch through AReaL2.0, reorganizing existing RL infrastructure into an agent-oriented online RL loop for policy weight updates from deployed workloads.

Open-source libraries and tools are widely reused, but compatibility maintenance is expensive. Once maintainers leave, useful repositories can stop working as runtimes and dependencies evolve. We study whether LLM agents can adapt old repositories to modern environments, a task we call compatibility rescue. Unlike bug repair, compatibility rescue starts from a repository that worked in its original environment but fails after ecosystem drift. RepoRescue gives agents only the repository and its failing modern environment; the agent must diagnose the failure, locate affected code, and produce a source-code rescue that restores the historical test suite. We build RepoRescue from 193 Python and 122 Java repositories, each verified to pass historically and fail after modernization. We evaluate five deployed agent systems on Python and three on Java. Beyond full-patch pass rate, we rerun patches after removing test-file edits to measure source-only repair, add a runtime-enforced regime that blocks test edits, and validate practical use for repositories whose suites pass after rescue. We find that Claude Code systems sometimes edit failing tests even when prompted not to; with runtime blocking, Kimi still rescues 41.5% of repositories. Systems are complementary: their union reaches 62.7%, exceeding the best single system by 10.9 points. Difficulty concentrates in cross-file coordination: on 14 repositories requiring coordinated whole-codebase changes, GPT-5.2 through Codex passes all 14, while every Claude Code system passes at most two. Finally, a passing suite is only an initial signal: among 34 unmaintained Python candidates whose suites pass after rescue, 22 work in realistic scenarios and 12 pass bug-hunt with patches that address the compatibility failure. RepoRescue benchmarks compatibility rescue with source-only auditing, runtime enforcement, practical validation, and reasoning labels.

Reasoning segmentation requires localizing targets based on complex, implicit queries. Current end-to-end models typically entangle perception and deduction into an opaque black box, severely limiting interpretability and scalability. To address this, we propose GEAR-Seg (Grounded Explainable Agent for Reasoning Segmentation), an explicitly decoupled agent that shifts the paradigm by translating visual pixels into dense, attribute-rich text. By decoupling class-agnostic segmentation, semantic description, and Large Language Model (LLM) deduction, GEAR-Seg transforms implicit reasoning into an explicit, trackable logic chain. As a zero-shot inference framework, it achieves highly competitive performance across diverse reasoning and fine-grained referring segmentation benchmarks. Furthermore, GEAR-Seg inherently functions as a highly scalable data engine. Utilizing this engine, we construct GEAR-131K, a massive benchmark (over 38k images, 656k QA-mask pairs) introducing a multifaceted taxonomy tailored for complex real-world manipulation-oriented reasoning. Finally, distillation experiments demonstrate that lightweight models supervised exclusively by our automated pipeline closely match the upper-bound performance of costly human-annotated baselines.

Memory has emerged as a cornerstone of modern LLM-based agents, supporting their evolution from single-turn assistants to long-term collaborators. However, memory is not always beneficial: retrieved memories often induce a critical issue of sycophancy, causing agents to over-align with the user at the cost of factual accuracy or objective reasoning. Despite this emerging risk, existing memory benchmarks primarily evaluate whether memories are correctly stored, retrieved, or updated, while overlooking how retrieved memories influence downstream reasoning and decision-making. To bridge this gap, we propose MemSyco-Bench, a comprehensive benchmark for evaluating memory-induced sycophancy in agent systems. MemSyco-Bench measures when memory should influence a decision and how valid memory should be used. Specifically, it covers five tasks that assess whether agents can reject memory as factual evidence, respect its applicable scope, resolve conflicts between memory and objective evidence, track memory updates, and use valid memory for personalization. All related resources are collected for the community at https://github.com/XMUDeepLIT/MemSyco-Bench.

Existing egocentric benchmarks have primarily constructed the egocentric setting from first-person-view data, which makes it difficult to evaluate egocentric perspective itself in isolation. However, understanding first-person-view input and taking an egocentric perspective are separable abilities, especially when first-person body cues are absent or when other agents are present. To isolate egocentric perspective understanding, we introduce EgoGapBench, a diagnostic benchmark for measuring action selection in multi-agent egocentric scenes. We define the ability measured by this benchmark as Egocentric Action Selection (EAS): selecting an appropriate action from the agent's perspective in the presence of other agents. On EgoGapBench, humans answer reliably, whereas both open-source and proprietary MLLMs perform substantially worse and systematically select actions performed by other visible agents. Fine-tuning on existing egocentric data fails to close this gap and can even be detrimental. In contrast, fine-tuning on EgoGapBench training data improves accuracy but does not reach human performance. These results show that EAS is difficult to acquire from first-person-view data alone, and that MLLMs should be evaluated and trained not only for scene understanding but also for egocentric action selection.

Autonomous scientific discovery systems offer the potential to accelerate research by automating the process of hypothesis generation and validation. However, current systems operate within constrained search spaces or require predefined research questions, limiting their capacity for true open-ended inquiry. Furthermore, while they generate hypotheses iteratively, they largely lack the ability to explicitly synthesize their own accumulated findings to uncover complex, interconnected phenomena. We introduce DiscoPER, an autonomous large language model-powered framework that conducts open-ended research by dynamically generating and executing code to explore datasets without pre-specified research objectives. To ensure rigorous scientific validity, every proposed discovery must pass statistical testing. To overcome the limitations of isolated search, our framework introduces a second-order reasoning mechanism that periodically analyzes its own accumulated discoveries. By treating prior discoveries as empirical data, DiscoPER identifies structural patterns, confounds, and epistemic gaps, actively redirecting hypothesis exploration toward uncharted regions of the search space. The search space is further expanded by incorporating tool use, enabling the system to explore hypotheses beyond structured metadata by seamlessly processing and extracting useful information from multimodal sources like images. Evaluated on iNatDisco, a new multimodal ecological knowledge benchmark with pattern-level ground truth obtained from peer-reviewed literature, DiscoPER recovers 8 of 9 known patterns with a 72.7% hypothesis support rate, outperforming both classical causal discovery and LLM-guided baselines. Ablations show that DiscoPER scales with more data, and confirms the benefits of second-order meta-reflection.

Although large language models (LLMs) have demonstrated impressive creative fiction generation, they struggle to maintain narrative consistency and coherent plot lines in long-form stories. In this work, we introduce a unified framework for long-form narrative generation and verification. MAGNET, a multi-agent goal-driven narrative engine for storytelling, generates stories with persona-grounded character agents that propose actions based on a shared world state and evolving story goals, while ATLAS is a graph-based pipeline that compares scene-level world representations across a generated story to detect hallucinations. By evaluating MAGNET using an LLM editor, pairwise rubric scoring, and ATLAS, we show that our framework produces coherent narratives compared to single-model prompting and IBSEN. At 100 pages, MAGNET reduced annotations and hallucinations by 41 and 50%, respectively, compared to the single model baseline and by 34 and 45%, respectively, compared to IBSEN, with pairwise rubric evaluation showing similar results. These results suggest that long-form narratives can emerge from explicit world-state tracking and goal-driven multi-agent generation, providing a foundation for controllable and structurally coherent long-form narrative generation.

Large language models (LLMs) are powerful pattern-completion systems, but their default operating mode - predicting the next token from a static context - does not reliably produce persistent, manipulable representations of an external world. Many tasks that look like "reasoning" in text become substantially harder once the environment is partially observable, stateful, and requires memory and structured hypotheses about hidden state. AGI Maze is a lightweight framework for building such environments without requiring high-dimensional sensory inputs. It provides a family of grid-based maze tasks with a clean API and multiple difficulty regimes. The goal is to create benchmarks where agents must learn and use world state representations, not just infer a local rule over readily provided observations. We provide an initial evaluation of several vanilla LLMs on simple mazes showing that they fail to represent mazes internally at LLM inference time. We also introduce a baseline agent, which is allowed to use its message history as a working memory to construct descriptions of observations at agentic runtime. Although this can improve performance, it is still insufficient for an LLM agent to reliably solve even small mazes within a step budget that is more than enough for humans.

While Large Language Model (LLM) agents demonstrate proficiency in static benchmarks, their deployment in real-world scenarios is hindered by the dynamic nature of user queries, tool sets, and interaction dynamics. To address this generalization gap, we formalize OpenAgent (Tool-Use Agent in Open-World), a problem setting characterized by distributional shifts across query, action, observation, and domain dimensions. To systematically diagnose its impact, we construct a controlled sandbox environment where we define fine-grained environmental shifts across a four-tier hierarchy, Perception, Interaction, Reasoning, and Internalization, and conduct a comprehensive series of experiments. Our analysis yields a series of key insights, demonstrating that agents trained via both Supervised Fine-Tuning(SFT) and Reinforcement Learning suffer from varying degrees of performance degradation when confronting open environmental shifts. Building on these insights, we propose Perturbation-Augmented Fine-Tuning, a disturbance-based intervention strategy for SFT that lays the foundation for enhancing agent robustness and utility in realistic environments. Our code will be released at: https://github. com/LAMDA-NeSy/OpenAgent.

Long-horizon LLM agents accumulate tool results, files, plans, and user constraints that are too structured to be treated as a disposable text suffix. Current systems mostly rely on in-run heuristics such as chronological pruning and tool-output masking, or on final self-summary near a context limit. Heuristics are cheap but blind to future dependencies; summaries preserve narrative state but often hide exact evidence, locators, and editable artifacts. We present Self-GC, where GC denotes self-governing context while deliberately echoing garbage collection: the system does not merely reclaim unused tokens, but governs the lifecycle of agent context objects. Self-GC turns user turns, tool spans, and skill state into indexed objects; asks a side-channel planner to propose fold, mask, and prune actions; and lets the harness enforce recoverable sidecars, safe commit boundaries, and cache-aware commit. On a 33-session Hard Set, Self-GC prunes 43.95% of prefix tokens while leaving 84.85% of future continuations unaffected, compared with no-impact rates of 54.55% to 69.70% for heuristic baselines. On a 332-session production-derived suite, three planner backbones reach no-impact rates of 91.27% to 94.58%, while baselines remain at 77.71% to 87.46%. In production, an online account-level split reduces daytime average input tokens by 10% to 15%, with peak reductions near 20%. These results point to context management as runtime lifecycle control over indexed, recoverable objects rather than post hoc text cleanup.

Long-horizon egocentric question answering involves answering about events that have occurred hours or days in the past. This requires memory representations that remain both retrieval-effective and scalable over days or weeks of recording. Existing long-horizon egocentric QA methods construct memory as hierarchical textual summaries of observations. While effective for reducing memory size, summarization optimizes for descriptive compression rather than retrieval: repeated interactions are absorbed into coarse textual descriptions instead of being preserved as explicit, recurring memory units, making long-horizon evidence aggregation difficult. We propose Imprint, an interaction-centric memory framework that formulates long-horizon egocentric memory as an online memory compression problem rather than summarization. Incoming observations are first represented as structured Interaction Records and continuously organized into recurring interaction patterns. Using human memory consolidation signals of recurrence, recency, and distinctiveness, Imprint selectively retains and compresses interactions into a compact retrieval-oriented memory. We evaluate Imprint on EgoLifeQA, a seven-day egocentric benchmark containing questions that require reasoning over interactions occurring hours to days before the query. With the same LLM, Imprint improves QA accuracy from 31.0% to 35.8%, increases evidence-grounded answers by $6\times$ compared with EgoRAG, reduces memory footprint by $2.3\times$, and decreases retrieval latency by $11.8\times$. These results demonstrate that memory compression provides a scalable and retrieval-effective foundation for long-horizon egocentric question answering.

Zeroth-order (ZO) optimization enables fine-tuning large language models when backpropagation is unavailable or memory-prohibitive, but existing methods often perturb full model weights or randomly constructed low-dimensional subspaces, yielding high-variance estimates and limited performance. We propose ZO-Act, an activation-informed ZO fine-tuning method that restricts perturbations to a fixed low-rank subspace derived from input activations. For each linear layer, ZO-Act computes a small activation basis once at initialization and optimizes only lightweight coefficient matrices using forward-only loss evaluations. This reduces the effective perturbation dimension, exposes explicit trainable variables compatible with momentum-based optimizers such as Adam, and naturally supports quantized LLM fine-tuning by keeping low-bit weights frozen. We analyze ZO-Act as zeroth-order optimization over a restricted coefficient space and show that perturbing the low-dimensional coefficients reduces both the variance-dependent convergence term and the finite-difference error of the ZO estimator, at the cost of a controlled subspace approximation bias that is mitigated by the low-rank structure of LLM activations and gradients. Experiments on Llama-3-8B, OPT-13B, and INT4 Llama-3-8B show consistent gains over strong ZO fine-tuning baselines across language understanding, question answering, and commonsense reasoning.

Large language models increasingly generate C++, a memory-unsafe language where a single overlooked violation can become an exploitable bug. Yet most security evaluations of AI-generated code rely on static analysis alone, which flags warnings without confirming runtime violations or reasoning about untested paths. We ask whether AI-generated C++ is measurably less safe than human-written code, and whether common verification tools agree on the risk. We introduce VULBENCH-CPP, a benchmark of 8,918 C++ programs from three open-weight LLMs (Gemma 3 27B IT, LLaMA 3.3 70B Instruct, Qwen 2.5 Coder 32B Instruct) and human authors across 851 competitive-programming tasks. Each program is annotated by four verification tiers: functional testing, static analysis (cppcheck, clang-tidy), dynamic analysis (ASan/UBSan), and bounded model checking (ESBMC). Accounting for the correlation among solutions to a shared task, we find that AI-generated code is roughly twice as likely as human code to trigger a confirmed runtime violation, even after controlling for code length and test pass-rate. Under static analysis the two look equally safe, but this is misleading: the apparent similarity reflects code length rather than real safety, and the tiers detect largely different classes of violation, so no single tier is sufficient. The gap is consistent across independent generations.

This paper explores multi-turn visual reasoning and observes that MLLMs repeatedly fail to localize the target, leading to long, redundant trajectories. We attribute this failure to the entanglement of reasoning and perception within a single model, the MLLM reasons and localizes simultaneously, and inaccurate localization triggers additional reasoning turns that bloat the trajectory. To solve this problem, we propose PixelEyes, a multi-turn visual reasoning agent that explicitly decouples reasoning from perception, i.e., the reasoner decides what to look for, while a specialized perception tool answers where it is. Specifically, PixelEyes introduces 1) Mask-guided Visual Search. A referring segmentation model is invoked to provide mask-precise localization, freeing the reasoner from the need to compensate for imprecise grounding. 2) Semantic-region Breadth-first Search (BFS). To eliminate redundant loops caused by repeatedly cropping incorrect sub-regions, we organize exploration as a breadth-first search over semantic regions. To internalize these capabilities, we construct the PixelEyes-6K dataset by resynthesizing expert trajectories from existing data. This explicitly embeds our mask-guided search and BFS logic into the model. We further introduce Pinpoint-Bench, a zero-hint visual search benchmark, i.e., no location cues are provided in the question, with instance-level masks and bounding boxes that separate localization failures from reasoning failures, enabling fine-grained analysis of failure modes such as inattentional blindness. Recent state-of-the-art MLLMs and visual reasoning agents leave large headroom on Pinpoint-Bench, demonstrating its quality and difficulty. Code and models are open-sourced.

How do two agents invent a shared language from scratch? In a Lewis signaling game, a sender and receiver must coordinate on a code using only their interaction history. We study five memory architectures across varying channel configurations with LLM agents and find that memory architecture matters more than channel capacity. Agents with a persistent private notebook benefit from surplus channel capacity and avoid the high-capacity collapse seen in stateless agents, achieving the most reliable coordination ($0.867 \pm 0.023$ at capacity = 25). Stateless agents peak at moderate capacity and then degrade as the vocabulary grows beyond what a rolling context window can track The notebook externalizes learned conventions, freeing agents from having to re-derive codes each round. An information bottleneck-inspired argument predicts an optimal capacity equal to the number of objects. Instead, the bottleneck (capacity = 8) proves to be a fragility point, and surplus capacity is generally better. We show that channel capacity alone cannot predict coordination; memory architecture determines whether agents turn interaction history into stable conventions, and both dimensions are needed to understand how signals become language.

Multimodal large language models (MLLMs) show strong promise for clinical VQA and radiology report generation, yet inference-time hallucinations still undermine trustworthy use: models can produce fluent conclusions that conflict with imaging evidence. Existing mitigation strategies typically rely on additional training, external retrieval/knowledge bases, or multi-stage post-hoc verification, which increases cost and pipeline complexity and often generalizes poorly across models and tasks.To address this, we propose a holistic, training-free evidence-injection framework that systematically mitigates hallucinations through dual-side evidence injection. By leveraging ROI priors acquired using MedSAM in our implementation, we recalibrate the visual perception trajectory via ROI-guided activation modulation while anchoring the textual reasoning trajectory by mapping anatomical coordinates into discrete semantic tokens as verifiable external memory. Then we introduce a task-aware dynamic router to select modality-specific interventions based on task semantics, balancing perceptual grounding and linguistic fluency. We conduct systematic evaluations on 2 tasks and 5 datasets using LLaVA-1.5-7B, LLaVA-Med-1.5-7B, Qwen3-VL-8B/32B, and InternVL-3.5-8B/38B. Controlled ablations and visualizations further validate the framework, which consistently outperforms baselines across medical benchmarks, improving close-ended accuracy by up to $\sim\mathbf{6}%\uparrow$ and reducing open-ended hallucinations by $\sim\mathbf{35}%\downarrow$. The code has been made available on GitHub: \href{https://github.com/Henry991115/SPRG}{\textcolor{blue}{https://github.com/Henry991115/SPRG}}.

LLM-based agents execute multi-turn workflows with continuously growing contexts, where LLM calls are interleaved with tool invocations and environment feedback. To maintain model quality, modern agent frameworks rely on context engineering strategies such as offloading, reduction, and isolation to control the context length. However, these strategies introduce significant context transformation overhead: each transformation invalidates existing KV caches and triggers re-prefill, leading to increased time-to-first-token (TTFT). In this paper, we identify that context transformations are segment-decomposable, where the transformation of a prefix is independent of future tokens. This property enables transformations to be executed ahead of time. Based on this insight, we propose a lookahead programming model that allows agent frameworks to express context transformations as asynchronous operations without modifying their execution logic. The runtime proactively executes these transformations and prepares transformed KV caches in advance, enabling direct context replacement without blocking. We further design a lookahead-aware scheduler in LLM serving systems to support these asynchronous requests alongside latency-critical workloads with controlled interference. We implement our approach to support representative context engineering strategies and integrate it into existing agent frameworks and LLM serving systems. Experiments show that our approach effectively eliminates transformation overhead and reduces TTFT by up to 11.9x.

Specialist epilepsy expertise is scarce in resource-constrained settings, making LLM-based decision support attractive for frontline clinicians managing longitudinal treatment. Such systems must adapt to local prescribing practice and know when to defer. We study this problem in Ugandan pediatric epilepsy care, predicting anti-seizure medication regimens from longitudinal unstructured clinic notes. Standard prompting achieves non-trivial agreement with physician prescriptions, but neurologist review shows that many errors reflect distribution-miscalibrated prescribing defaults rather than failures to parse the local record. We introduce MANANA, a non-parametric prompt-learning framework that learns local prescribing guidance from a small patient-level training set. MANANA converts observed prescription errors into auditable prompt memories, instantiated in single-agent and multi-agent variants, and improves over classical ML models, direct LLM prompting, and prompt-optimization baselines across two independently collected Ugandan cohorts. We further propose Bayesian prompt averaging, which converts the learned prompt trajectory into prescription likelihoods and an uncertainty-based deferral signal. On the independently collected held-out cohort, this improves visit-level top-3 prescription accuracy by 4-8 percentage points over prompt-optimization baselines and enables selective prediction: the system can auto-handle the most confident half of cases at 95% precision, or the most confident quarter at 99% precision, while deferring lower-confidence cases for specialist review.

While Large Language Models (LLMs) have demonstrated exceptional capabilities in mathematical reasoning, they frequently produce subtle errors that evade human detection. Formal mathematical languages like Lean 4 offer mechanical proof checking, strongly motivating the need for autoformalization: the automatic translation of natural language mathematics into verifiable code. Recent trends indicate that general-purpose LLMs, heavily optimized for standard programming, now outperform smaller models explicitly fine-tuned for Lean. Leveraging this shift, we introduce an agentic autoformalization framework powered by general coding LLMs. At the core of our system is an orchestrator that manages a multi-agent pipeline tailored for research-level mathematics. Because cutting-edge research frequently relies on concepts outside the scope of existing libraries like Mathlib, our system dynamically extends necessary type definitions and validates them via a novel Auxiliary Lemma technique before formalizing the primary theorems. We applied our approach to PutnamBench, producing machine-checked Lean proofs for a random sample of 32 problems. Furthermore, we evaluate our system on five papers from the ACM Symposium on Theory of Computing (STOC) spanning combinatorics, communication complexity, mechanism design, and learning theory, successfully formalizing their main theorems and validating the generated formalizations with human experts; for all five we also formalize the proofs alongside the statements, and notably two of them are proved with no axioms beyond Lean's kernel. All of our formalizations are available at https://beyondthelibrary.github.io/formal_arxiv .

AI-assisted research has entered a stage in which the central question is not only whether systems can generate hypotheses, run experiments, or produce manuscripts, but whether their scientific claims are calibrated to the evidence that supports them. This Perspective-style paper develops a conceptual and methodological framework for evidence-licensed claims in AI-assisted research. Motivated by representative routes including specialized scientific foundation models, LLM research assistants, multi-agent co-scientists, AI Scientist pipelines, mathematical discovery agents, and self-driving laboratories, it represents AI-assisted research as five operators: hypothesis generation, model-mediated consequence derivation, external validation, belief update, and claim calibration. The central claim is that calibration is not merely cautious wording but a mechanism for managing scientific assertion rights: evidence licenses some forms of speech and withholds others. The paper distinguishes linguistic, consequence-based, interventional, and evidence-licensed semantics; defines the claim-evidence gap and epistemic debt; and treats minimal structural reconstruction across heterogeneous outputs as an upward form of claim calibration. AISim-Cal is included as an illustrative synthetic dynamics exercise, not as an empirical forecast or benchmark. The resulting principles are: no claim without license, validation does not determine claim level, and automation amplifies the need for calibration. Reliable AI-assisted research is therefore evaluated as a loop that generates hypotheses, derives testable consequences, accepts independent adjudication, updates beliefs, and outputs only evidence-licensed claims.

Tool-augmented vision-language models (VLMs) can solve multimodal, multi-step tasks by calling external tools, yet they remain fragile in practice. Existing works have two common gaps. Supervised fine-tuning (SFT) is built mostly on successful trajectories and offers little signal for recovery after tool failures, while sparse trajectory-level RL rewards provide limited guidance on which step failed and how to repair it. We introduce ReGRPO (Reflection-augmented Group Relative Policy Optimization), a framework that learns reflection-guided correction in tool-using agents. ReGRPO starts with a structured reflective data engine: we execute near-miss actions to collect grounded failure observations, then build Reflection-of-Thought triplets (ErrorType, Evidence, FixPlan) paired with corrected actions for warm-start SFT. We then optimize reflection tokens and corrective actions jointly within local trajectories using group-relative advantages, and include a reflection-cost term to reduce unnecessary reflection. Experiments on GTA and GAIA show that, under the same backbone and tool suite, ReGRPO consistently outperforms strong open-source baselines and achieves the best results among the compared open-source controllers. Code and RoT data are available at https://github.com/showlab/ReGRPO.

We present SimpleSearch-VL, an efficient, reliable, and practical framework for multimodal agentic search. Its core idea is to improve the agent's own search-and-verification process rather than scaling data, tools, or auxiliary model components. For efficiency, Factorized Adaptive Rollout (FAR) improves sampling efficiency by forming more informative training groups while using redundant samples to mitigate long-tail latency and expose hard samples. For reliability, SimpleSearch-VL performs evidence-verified reasoning, explicitly using chain-of-thought verification to assess the relevance of retrieved visual and textual cues to the original context. For practicality, SimpleSearch-VL keeps a lightweight tool interface and performs webpage self-summary within the agent, requiring no additional external dependencies. With only 5K supervised tool-interleaved trajectories and 2K RL data, SimpleSearch-VL improves Qwen3-VL agentic baselines by 15.8 and 16.0 average points for the 8B and 30B-A3B variants, respectively. The SimpleSearch-VL-30B-A3B model further achieves performance competitive with agentic Gemini-3-Pro.