Peizhi Niu, Ge Shi, Vaibhav Gorde, Sarthak Ghosh, Christos Faloutsos, Min Cheng, Kevin Chen
LLM agents struggle to automate complex enterprise procedures because static, pre-built harness cannot capture proprietary policies, undocumented behaviors, or state-dependent recovery paths that surface only during execution. Procedural memory accrued from agent experience offers a training-free path to closing this gap -- yet unlike episodic records or semantic facts, it must consolidate repeated successes and diagnosed failures into reusable actions and recovery strategies. We introduce WASP-Mem (Wiki-based Agent-Synthesized Procedural Memory), a nimble but effective method that distills execution traces and feedback into persistent, interlinked wiki pages, enabling continual agent learning. On TheAgentCompany benchmark, WASP-Mem raises the checkpoint pass rate by 4.6-6.1% over the strongest baseline we compared across three models while requiring 57-61% less storage. We further present a systematic controlled study spanning consolidation method, feedback type, trace ingestion scope, and refinement rounds. Our results reveal that prescriptive knowledge organization such as human defined schemas and explicit deduplication rules hurts performance rather than closing the procedural-knowledge gap. In contrast, human diagnosis converts failed traces into corrective knowledge and drives sustained improvement: a single round of human diagnosis outperforms four rounds of self-judgment by 7.0-10.9%. These findings point to a simple division of labor: let agents organize procedural memory, and invest human effort where it matters most -- diagnosis.
Rongzhe Wei, Ge Shi, Min Cheng, Sarthak Ghosh, Vaibhav Gorde, Leman Akoglu, Pan Li, Na Zhang
Enterprises increasingly want LLM agents to automate standard operating procedures over their software stacks -- mapping each step of a procedure to the right call among thousands of internal APIs, without hand-authored step-to-API mappings. The core difficulty is API selection at scale: choosing and composing the correct tools over a long horizon, judged only by whether the overall task succeeds. Current tool-use benchmarks do not test this regime -- they expose only a handful of tools, evaluate single-step invocation, or grade multi-step behavior with subjective LLM-as-a-judge scores that penalize valid alternative trajectories. We introduce SLATE (Synthetic Large-scale API Toolkit for E-commerce), a large-scale benchmark that tests large-toolset, long-horizon API selection in a realistic enterprise domain. SLATE couples (i) queries decomposed into hierarchical, tool-grounded procedures over a ~1,000-tool e-commerce library, (ii) a deterministic, context-aware simulator that returns recorded outputs for matching calls and default responses otherwise, and (iii) an end-to-end task-success metric -- a task succeeds when it reaches every reference outcome -- that credits diverse-but-valid trajectories while remaining fully automatic and reproducible. Experiments with six open decision models and the language models they are built on, from 3B to 12B parameters, show that even the best model completes only 5.0% of the tasks, establishing SLATE as a challenging testbed for reliable enterprise tool use. We release SLATE, its simulator, and the evaluation harness at https://geshijoker.github.io/slate/
Do multilingual translation models organize meaning around a shared conceptual space, or do they primarily encode language identity and surface form? We probe the encoder geometry of Meta's 600M-parameter NLLB-200 model across 135 languages and 101 Swadesh concepts in contextualized carrier-frame and decontextualized bare-word regimes against null and confound controls. First, two-way variance partitioning reveals an additive language-shift plus conceptual-manifold geometry: language shifts account for 45.7% of variance on bare words vs. 13.7% in context, and mean-centering improves the between-to-within-concept distance ratio (vs. a 0.99 permutation null) by 2.14x on bare words (1.15 to 2.46; 2.27x across disjoint scripts) and 1.19x in context (2.25 to 2.69), with conceptual separation peaking at Layer 10 (mutual k-NN overlap jumping 4.35x). Second, semantic category explains 25.4% of Swadesh convergence variance ($F(7,93)=4.53$, $p=2.21\times 10^{-4}$; partial $F=7.74$ controlling for fertility and length)---over 20x orthography or phonology ($R^2\le 0.012$)---while the advantage of Swadesh over non-loanword concrete nouns ($d=0.54$--$0.69$) is mediated by Zipfian subword compactness ($p=7.43\times 10^{-7}$). Third, language distances track ASJP NLD and LDND phonetic distances within subfamilies under block-permutation and cross-script controls ($ρ_s=0.65$--$0.67$, block $p=10^{-4}$; global bare-word Mantel $ρ_s=0.24$, $p=0.001$), and CLICS$^3$ colexified concepts lie closer across string-distinct, within-category ($ρ_s=0.32$), and bare-word ($d=1.79$) controls ($d=1.43$, $p=1.55\times 10^{-33}$). Finally, basic color terms recover the Berlin and Kay hierarchy ($ρ_s=0.91$--$0.93$, $p\le 2.18\times 10^{-4}$), and semantic offsets exceed distance-matched controls ($p=3.55\times 10^{-5}$, $d=1.06$). We release InterpretCognates, an interactive toolkit and reproducible suite.
Ge Shi, Jun Yin, Donglin Xie, Fangyi Liu, Yucan Li, Menglin Liu
Jailbreak attacks expose persistent safety weaknesses in large language models (LLMs), but existing stateless single-turn methods face a trade-off: hand-crafted prompts are expressive but static, while iterative prompt optimization can adapt but often relies on low-level mutations that require many target-model responses. We propose JailbreakOPT, a method that maintains a library of reusable prompt transformations (actions) and composes them into macro-actions for within-episode search, generating standalone attack prompts. To reuse experience across episodes, JailbreakOPT learns an action-selection policy using contextual Thompson sampling and outcomes from prior search paths. Experiments across multiple LLMs and benchmarks show that JailbreakOPT improves attack success rate (ASR) over single-action and iterative baselines. More broadly, exploration and exploitation over human-designed attack strategies can help probe residual vulnerabilities and identify failure patterns that may inform defenses aimed at broader classes of attacks.
We present E-GenRM (E-commerce Generative Reward Model), a generative reward model tailored to e-commerce shopping agents. User expectations are shifting from simple product queries to open-ended shopping tasks. However, existing trajectories and reward designs remain largely bounded by current agent usage patterns and known failures, providing limited guidance for further capability improvement. These limitations call for an integrated reward modeling framework for open-ended shopping agent tasks. On the data side, E-GenRM leverages naturally occurring user behavior across the shopping app to construct diverse, long-horizon conversational trajectories beyond established usage patterns. On the modeling side, E-GenRM first learns from behavior-grounded evaluation chains through supervised fine-tuning. It is then optimized through criterion-guided reinforcement learning, which rewards correct preference judgments and evidence-supported factual claims under each criterion. Personalized hard-negative training further improves its sensitivity to user preferences reflected in shopping behavior, beyond explicit requirements. We additionally develop an active evidence retrieval extension that acquires missing e-commerce information for evaluation. On E-GenRM-Bench, E-GenRM-27B achieves 87.40\% preference accuracy, outperforming the strongest baseline by 6.97 percentage points. In a 14-day online A/B test on Taobao, a shopping agent post-trained with E-GenRM improves the effective shopping-assistance rate by 1.39 percentage points and user satisfaction by 2.33 percentage points.
Anirudh Potlapally, Brijesh Nanda Brunda Yogananda, Jianyang Gu, Maanav Jajodia, Luke Meyers, Wei-Lun Chao, Hari Subramoni
Ecological monitoring increasingly relies on camera traps and aerial surveys that generate large image collections. High-recall animal detectors can localize candidate animals at scale. However, they also produce many detections that still require expert review, including false positives, ambiguous cases, and animals difficult to identify at the species level. Since exhaustive review requires substantial effort, we propose training-free active prototype triage, where reviewed detections become reusable evidence to (1) Resolve supported foreground classes, (2) Clear likely background false positives, or (3) Defer insufficiently supported cases for expert review. Across camera-trap and aerial datasets, foreground recognition and false-positive rejection benefit from different pretrained representations: BioCLIP2 provides foreground evidence while DINOv3 provides background evidence, reducing erroneous foreground clearing while improving automatic foreground recovery. Deferring unsupported cases avoids the higher error rates caused by forced decisions. Compared with confidence-thresholded multiclass k-NN and uncertainty-sampling logistic regression, prototype triage improves foreground recovery while maintaining competitive or lower review burden. Acquisition goes beyond class discovery: even after classes are represented, prototype choice determines how reliably evidence propagates through the remaining detections. Our results suggest that once animal localization reaches high recall, the challenge shifts from finding animals to deciding which examples to review, how to reuse them, and when evidence does not support an automatic decision.
Frankly Speaking: The Company Election Integrity Keeps in Congressional Communication
Michael A. Robinette, Christina Barsky, Amanda D. Clark, Helen Flannery, Juniper Lovato
AI 速览：通过分析2010至2025年国会电子通讯，发现两党使用相同的选举词汇但搭配完全不同的词语，导致选举诚信的含义分化。
Election integrity is a shared commitment in American politics, but Democrats and Republicans in Congress increasingly mean different things by it. Analyzing 10,420 congressional e-newsletters sent from 2010 to 2025, we examine how the two parties invoke election terms such as ``integrity,'' ``security,'' and ``fraud.'' Using diachronic word embeddings, which track semantic change through how words are used in context, we measure how differently each party uses each term, how much each party shifts, and whether they share nearest semantic neighbors. We show that members of Congress do not need different words to tell different stories about elections. They need only place those same words in different company. Across nearly every term and year, the two parties surround election-integrity vocabulary with entirely different words. That divergence is easy to miss, because most election vocabulary is no more divergent than other words used at roughly the same frequency, and neither party's usage shifted more than the other's. Only a subset of terms grew more divergent, most notably ``certification,'' while ``rigged'' and ``stolen'' entered election discourse after 2020 and polarized there. By 2024, Democrats surround ``voter,'' ``election,'' and ``integrity'' with the language of access and administration. Republicans surround the same terms with ``citizenship verification,'' ``noncitizen voting,'' and ``border security.'' For one party, integrity means protecting eligible voters' ability to cast a ballot. For the other, it means protecting the electoral process from illegitimate ballots. Election officials, who must administer a single system under both definitions, inherit this tension.
In reinforcement learning with verifiable rewards (RLVR), whether entropy rises or falls can depend on choices that matter little under exact sampling, such as whether over-length responses are masked or counted as failures. We trace this to sampler mismatch: rollouts are drawn with top-k/top-p truncation or at a temperature the learner does not apply. We prove that a constant advantage shift then changes the entropy of a softmax policy, at first order and in expectation, with a sign fixed by the sampler. Positive shifts sharpen the policy under any sampler whose likelihood ratio to the policy increases with token probability, including top-k, top-p, min-p, and temperatures below one, and flatten it under flatter samplers. In group-based RLVR with token-level losses, the baseline-dependent part of this drift is set by the token-weighted net advantage, which masking and failing can push to opposite signs. The rest comes from a covariance between advantages and the sampler's sharpness gap and feeds back on itself: in a two-level model, a push away from a sharper sampler takes off in finite time, while a pull toward it saturates. On a 2B model, the entropy slope is linear in shifts of the net advantage under truncated sampling and indistinguishable from flat under exact sampling. Token-level importance ratios leave 98% of the drift, and four clipped minibatch updates per batch amplify it sevenfold. In 100-step runs, a net-zero baseline only delays an entropy explosion, and entropy raised at the start keeps rising only under truncation; cancelling the first-order drift at every step keeps entropy bounded without changing the learner. On a 9B reasoning model forked from an uncorrected run, zeroing the net advantage or renormalizing the learner raises held-out accuracy after ten steps by 8.2 and 8.8 points over the uncorrected rule, with no difference under exact sampling.
Systems such as The AI Scientist have shown that AI agents can carry out and write up research, but in those demonstrations humans chose the research area and selected the outputs to show. We study letting the AI formulate the research question itself, and we report every run. Our system gives an off-the-shelf coding agent an instruction document and two skills: one for formulating the question, in which the agent diverges from recent technologies into problems of other fields, screens its candidates for known failure modes and obtains a critique from a different model; and one for peer review, in which a reviewer from another model family reruns the study from its records before the agent revises the paper. Over 13 development runs, changes to the one or two sentences that set the research area coincided with large shifts in what the candidate questions and the chosen papers were about, which led us to the area of a new technology applied to a societal problem. We froze the system after a pilot run and report five consecutive runs: the pilot, counted by a disclosed amendment to the pre-registration, and four pre-registered runs. All five runs went from question formulation to a revised, re-computable paper, with local computation on a laptop and no research input from humans, and external AI reviewers reading only the papers judged them at or near workshop level. We release the system, the five papers and the records needed to reproduce them.
Visual-token compression is usually evaluated by clean accuracy and efficiency, but these averages do not reveal failures that compression introduces under an image shift where the uncompressed model stays correct. We introduce VTC-AUDIT, which estimates this incremental risk only on pairs where the full model is correct on both images and the compressed model is correct on the clean image. The four arms then cross two binary choices: which image selects the retained token positions, and which image supplies the visual features. We audit three open 8-9B vision-language models on four tasks and eight pre-specified shifts. We evaluate two training-free compressors---hard-pruning DivPrune and prune-and-merge ApET---at a matched 25% retained-position budget, producing 708,000 inference records. Reselecting positions on the shifted image, rather than reusing the clean-image positions, raises the failure rate in all six model-compressor settings. The increase is 3.04-9.60 percentage points for DivPrune and 2.05-4.20 percentage points for ApET, with every 95% interval above zero. Repeating the audit on a separately resampled second manifest reproduces all six increases. In a pooled monitor audit across the four benchmarks, support Jaccard instability fails the practical AUROC criterion of 0.65 in all six settings, whereas clean compressed-model uncertainty clears it. Thus, support turnover is a weak ranking signal under the audited benchmark mixture, while clean compressed-model uncertainty is substantially more informative. These results suggest that future visual-token compression evaluations should report conditional reliability alongside clean accuracy and efficiency.
Mengxiao Zhu North China University of Technology, Haixu Chen, Jiu Sha Minzu University of China, Jie Liu North China University of Technology, Ge Shi
Low-resource multilingual OCR faces a dual challenge: complex script structures and severe data scarcity. In such settings, existing OCR models often struggle, as coarse visual representations combined with weak linguistic priors lead to frequent errors among visually similar characters.To address this, we present BASA (Beyond Atomic Sub-character Alignment), a OCR framework built upon high-resolution visual and language backbones with a novel glyph-aware interface. The core technical contribution is the Glyph-Aware Fine-grained Adapter (GAFA). Unlike standard linear projectors, GAFA employs learnable glyph prototypes to actively align sub-character structural primitives (e.g., strokes and radicals) with visual features, explicitly resolving topological ambiguities during vision–language alignment. To complement this, we introduce a two-stage curriculum learning strategy supported by a Glyph-Aware Reverse Synthesis pipeline, which generates large-scale multilingual training corpora with automatic, zero-cost component labels. Furthermore, we construct BASA-Bench, a representative benchmark spanning 11 languages with diverse script structures and 23 authentic scenarios. Experiments demonstrate that BASA achieves consistent improvements over strong OCR baselines, particularly on scripts with complex compositions. Our model and benchmark will be available at https://github.com/NcutLLM/BASA.
尚无 AI 解读，展示原文摘要
cs.CL · cs.CV
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Weisi Liu University of Memphis, Guangzeng Han University of Memphis, Xiaolei Huang University of Memphis
Time introduces fundamental challenges in model development and deployment: models are usually trained on historical data while deployed on future data where semantic distributions and domain knowledge may evolve. Unfortunately, existing studies either overlook temporal shifts or hardly capture rich shifting patterns of both semantic and knowledge. We develop Knowledge-driven Augmentation and Retrieval for Integrative Temporal Adaptation (KARITA) to capture diverse temporal shifts (e.g., uncertainty and feature shift), construct and integrate rich knowledge sources (e.g., medical ontology like MeSH), and leverage shifting insights for selecting-retrieval augmented learning. We evaluate KARITA on classification tasks across multiple domains, clinical, legal, and scientific corpora, demonstrating consistent improvements across all domain with temporal adaptation. Our results show that knowledge integration can be more critical and effective in temporal augmentation and learning.
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
Junpeng Liu, Jiuyi Li Dalian University of Technology, Kaiyu Huang Beijing Jiaotong University, Bo Jin Dalian University of Technology, Degen Huang Fuyao University of Science and Technology, Hui Xiong Hong Kong University of Science and Technology (Guangzhou)
Current large language models (LLMs) often exhibit performance imbalances between dominant languages (e.g., English) and non-dominant ones due to the skewed distribution of pretraining data. A common strategy to address this issue is to enhance cross-lingual alignment, thereby facilitating non-dominant language processing. However, existing methods typically rely on additional training objectives or language-specific parameters, which increase training complexity and cost. In this work, we propose a selective bidirectional language projection framework that enables efficient multilingual alignment and language shift using the intrinsic parameters. Specifically, we first identify the layers most sensitive to language projection between non-dominant and dominant languages through neuron activation analysis. We then perform sequential language projection within the selected layers by mapping non-dominant representations into the dominant language space and reverting them before generation. The bidirectional projection benefits the subsequent instruction tuning in non-dominant languages. Experiments on seven benchmarks demonstrate that our method remarkably enhances the performance of non-dominant languages. Further analyses indicate that our method learns better internal representations and exhibits strong generalization 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
Parisa Rabbani, Priyam Sahoo, Ruben Mathew, Aishee Mondal University of Illinois at Urbana-Champaign, Harshita Ketharaman, Nimet Beyza Bozdag, Dilek Hakkani-Tür
LLMs are increasingly used as third-party judges, yet their reliability when evaluating speakers in dialogue remains poorly understood. We show that LLMs judge identical claims differently depending on framing: the same content receives different verdicts when presented as a statement to verify (“Is this statement correct?”) versus attributed to a speaker (“Is this speaker correct?”). We call this dialogic deference and introduce DialDefer, a framework for detecting and mitigating these framing-induced judgment shifts. Our Dialogic Deference Score (DDS) captures directional shifts that aggregate accuracy obscures. Across ten domains, 3k+ instances, and five models, conversational framing induces large shifts (mean |DDS| = 15.9 percentage points (pp) across models, p < .0001) while accuracy remains stable (<2 pp), with effects amplifying 2–5× on naturalistic Reddit conversations. This effect is domain-dependent: a single model can shift toward disagreement (skepticism) on graduate-level science and toward agreement (deference) on social judgment. Ablations reveal that human-vs-LLM attribution drives the largest shifts (17.7 pp swing), suggesting models treat disagreement with humans as more costly than with AI. Mitigation attempts can reduce deference but over-correct into skepticism, revealing a calibration problem beyond accuracy optimization.
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
Embacher, Felix, Uhrig, Jonas, Cordts, Marius, Enzweiler, Markus
Retrieving rare and safety-critical driving scenarios from large-scale datasets is essential for building robust autonomous driving (AD) systems. As dataset sizes continue to grow, the key challenge shifts from collecting more data to efficiently identifying the most relevant samples. We introduce SearchAD, a large-scale rare image retrieval dataset for AD containing over 423k frames drawn from 11 established datasets. SearchAD provides high-quality manual annotations of more than 513k bounding boxes covering 90 rare categories. It specifically targets the "needle-in-a-haystack" problem of locating extremely rare classes, with some appearing fewer than 50 times across the entire dataset. Unlike existing benchmarks, which focused on instance-level retrieval, SearchAD emphasizes semantic image retrieval with a well-defined data split, enabling text-to-image and image-to-image retrieval, few-shot learning, and fine-tuning of multi-modal retrieval models. Comprehensive evaluations show that text-based methods outperform image-based ones due to stronger inherent semantic grounding. While models directly aligning spatial visual features with language achieve the best zero-shot results, and our fine-tuning baseline significantly improves performance, absolute retrieval capabilities remain unsatisfactory. With a held-out test set on a public benchmark server, SearchAD establishes the first large-scale dataset for retrieval-driven data curation and long-tail perception research in AD: https://iis-esslingen.github.io/searchad/
尚无 AI 解读，展示原文摘要
cs.CV
Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2026 · 正式发表