Peer Revenue Past citations are no guarantee of future returns

Papers

Fresh from arXiv, typeset properly. Highlighted words trade on the market.

Computer Science Computation and Language

cs.CL Watch Computation and Language

Monday 8 June

Computation and Language · Cross-listed from cs.LG

INFUSER: Influence-Guided Self-Evolution Improves Reasoning

Siyu Chen, Miao Lu, Beining Wu et al. (+7 more)

Self-evolution offers a scalable path to stronger reasoning: a pretrained language model improves itself with only minimal external supervision. Yet existing methods either depend on extensively curated or teacher-generated training data, or, when the generator runs unsupervised, reward it by a difficulty heuristic that need not improve the solver. We introduce INFUSER, an iterative co-training framework with two co-evolving roles: a Generator that drafts questions and reference golden answers from a pool of unstructured, automatically collected documents, and a Solver that improves by training on them. The solver is trained with standard correctness rewards against the generator-provided answers, while the generator is rewarded by an optimizer-aware influence score that measures whether each proposed question would actually improve the solver on the target distribution. Because this continuous, noisy influence score is poorly served by standard GRPO, we propose DuGRPO, a dual-normalized variant of GRPO, for generator training. Together, these turn the document pool into an adaptive curriculum that favors questions useful to the current solver, not just hard ones. On Qwen3-8B-Base, INFUSER outperforms strong self-evolution baselines with over 20% relative improvement on Olympiad and SuperGPQA benchmarks, and an 8B INFUSER co-evolving generator outperforms a frozen 32B thinking generator on math and coding. Ablations confirm each design choice is necessary, and two extensions, applying INFUSER to an instruction-finetuned anchor and augmenting it with rule-verifiable RLVR data, further demonstrate the flexibility and generalizability of the framework. Code is available at https://github.com/FFishy-git/INFUSER.
Read the full abstract

All authors Siyu Chen, Miao Lu, Beining Wu, Heejune Sheen, Fengzhuo Zhang, Shuangning Li, Zhiyuan Li, Jose Blanchet, Tianhao Wang, Zhuoran Yang

Computation and Language · Cross-listed from cs.LG

Measurement Under Selection: Decoy-Calibrated Failure Audits for Language Models

Vyzantinos Repantis, Ameya Gawde and Harshvardhan Singh

Knowing how often a language model fails does not explain where its errors concentrate. When auditors examine many explanations, the strongest observed pattern may arise by chance. We introduce Janus, a procedure for checking proposed error patterns before reporting them. Janus starts with a fixed list of yes/no properties of the examples being evaluated, such as whether the input is long. For each property, it compares the model's error rates on examples with that property and those without it. To see how large a difference can arise by chance, it repeats this calculation after shuffling the yes/no labels across examples without changing the group sizes. These shuffled properties are called decoys. A pattern is reported only if the size of its error difference meets a threshold set using decoys. On separate held-out examples, the same group must still have the higher error rate and the difference must meet a minimum, which was chosen in advance. In a controlled experiment, where the model must find a code in documents containing tables of staff, projects, and renewal codes, Janus confirms five related patterns of higher error rates on tasks requiring more lookups across tables. It also confirms a sixth pattern: lower error rates on examples with the needed information at the ends of the tables. In our samples from the MuSiQue and LongBench v2 public benchmarks, SliceLine finds groups with high error rates, while Janus reports no confirmed error patterns for the example properties we chose to test. For comparison, we use standard tests that shuffle errors and account for testing many candidates. With the same holdout check, they confirm two to six controlled patterns, depending on the test and threshold, and none on either benchmark. In simulations with no real error patterns, Janus reports false patterns more often than Benjamini-Hochberg, depending on the decoy count.
Read the full abstract

Computation and Language · Cross-listed from cs.LG

DynaCF: Mitigating Shortcut Learning in Reward Models via Dynamic Counterfactual Sensitivity

Fengyuan Liu, Yongliang Miao, Zirui He et al. (+3 more)

Reward models trained from pairwise preferences often exploit superficial shortcut cues rather than learning true response quality. We propose DynaCF, a dynamic reweighting framework for mitigating shortcut learning in reward model training. Unlike static shortcut heuristics, DynaCF measures shortcut sensitivity online during optimization by applying semantics-preserving counterfactual perturbations and tracking the resulting margin shifts and preference flips under the current model. Samples with higher shortcut sensitivity are dynamically downweighted in the Bradley-Terry objective, encouraging the model to rely less on superficial patterns and more on task-relevant preference signals. Extensive experiments show that DynaCF consistently improves robustness in preference modeling.
Read the full abstract

All authors Fengyuan Liu, Yongliang Miao, Zirui He, Yanguang Liu, Fei Sun, Mengnan Du

Computation and Language · Cross-listed from cs.CV

CRANE: Knowledge Editing for Reasoning MLLMs

Han Huang, Hao Wang, Mengqi Zhang et al. (+3 more)

The emergence of reasoning multimodal large language models (MLLMs), which generate explicit chain-of-thought (CoT) reasoning before producing answers, has introduced a new challenge for knowledge editing: methods that appear successful under traditional metrics (teacher-forcing accuracy up to 100%) can fail severely when the model's reasoning process is examined (Grounded Success as low as 0%). We identify three failure modes: (1) Structural Collapse, where weight-modifying methods destroy the CoT format; (2) Cognitive Dissonance, where the model's reasoning chain actively rejects the injected edit fact based on visual evidence; and (3) Shallow Internalization, where methods succeed on exact queries but fail on rephrase or multi-hop variants. On reasoning MLLMs, these modes interact: methods that generalize (FT, LoRA) trigger format collapse, while methods without deep modification cannot generalize. To expose these failures, we propose a CoT-aware evaluation protocol and construct ReasonEdit-Bench, with conflict stratification, multi-level probes, and multi-hop portability tests. We propose CRANE, a retrieval-augmented framework that requires no per-edit parameter modification. CRANE combines a modality-aware dual-library retrieval system with a two-phase training strategy: Supervised Fine-Tuning (SFT) for structural initialization, followed by GRPO with a Cognitive Routing Reward that trains the model to arbitrate between visual priors and injected edit facts. On ReasonEdit-Bench, CRANE achieves 96.9% Grounded Success on conflict scenarios and 96.9% intermediate entity usage in multi-hop chains, with 97.6% text-locality and 68.1% image-locality Edit Independence. On the out-of-distribution MMEVOKE benchmark, CRANE reaches 87.0% under gold retrieval.
Read the full abstract

All authors Han Huang, Hao Wang, Mengqi Zhang, Shu Wu, Qiang Liu, Liang Wang

Computation and Language

Bridging the Agent-World Gap: Text World Models for LLM-based Agents

Yixia Li, Hongru Wang, Peng Lai et al. (+13 more)

Large language model (LLM)-based agents are increasingly used in interactive textual environments, from web navigation and code editing to tool use and long-horizon dialogue. Yet many remain largely reactive, mapping observations to actions without an explicit model of how these environments are structured and evolve. This motivates text world models (TWMs): transition models over textual states that, given a state and a candidate action, predict the resulting webpage, terminal output, API response, or user reply, thereby supporting planning, efficient learning, and principled evaluation. We systematically review text world models for LLM-based agents, organized around a formal framework and the agent lifecycle: (1) Foundations, defining text world models and characterizing them by state representation and grounding domain; (2) Construction, taxonomizing LLM-as-WM and code-as-WM paradigms and reviewing methods for building them; (3) Application, examining how world models support agents at training time through experience synthesis and at inference time through planning, verification, and adaptation; and (4) Evaluation, covering both evaluation of the world model itself and its use as an evaluation environment for agents. We aim to consolidate this rapidly developing area, clarify its design space, and highlight open challenges for future research.
Read the full abstract

All authors Yixia Li, Hongru Wang, Peng Lai, Zhiwen Ruan, He Zhu, Youxin Zhu, Ganlong Zhao, Minda Hu, Yun Chen, Sibei Yang, Peng Li, Jeff Z. Pan, Jia Pan, Guanhua Chen, Yang Liu, Guanbin Li

Computation and Language · Cross-listed from cs.LG

TRIAGE: Dialectical LLM Reasoning for Explainable Risk Prediction on Irregularly Sampled Medical Time Series

Hyeongwon Jang, Gyouk Chu, Changhun Kim et al. (+4 more)

Clinical early warning systems built on irregularly sampled medical time series (ISMTS) from electronic health records must deliver continuous risk scores for patient triage as well as interpretable rationales that clinicians can verify. Large language models (LLMs) are uniquely positioned for both, deriving risk from their output probabilities and rationales from their medical knowledge. However, we find that conventional LLM reasoning collapses graded risk into overconfident predictions and thereby undermines the cross-patient comparability on which triage depends. We refer to this failure mode as risk polarization and identify two underlying behaviors: early commitment to a single outcome, and one-sided reasoning that focuses only on the evidence for that outcome. To address this, we propose TRIAGE, a framework that trains an LLM to reason dialectically over competing clinical outcomes by eliciting outcome-specific rationales. This dialectical formulation mitigates risk polarization, enabling a single LLM to jointly provide explicit clinical rationales and risk scores comparable across patients. Across five ISMTS benchmarks, TRIAGE improves mean AUPRC by 17.0% and reduces mean calibration error by 82.8% relative to the competitive LLM-based baseline, while surpassing the strongest ISMTS baseline by 3.5% in mean AUPRC.
Read the full abstract

All authors Hyeongwon Jang, Gyouk Chu, Changhun Kim, Hangyul Yoon, Jeonguk Lee, Eunho Yang, Joonhyung Park

Computation and Language

SafeRun: Enabling Determinism in LLM Planning for Running

Meilin Chen, Zepeng Zhai, Jiaxuan Zhao et al. (+1 more)

Large Language Models enable flexible natural-language planning but remain unreliable in determinism-critical domains due to their probabilistic nature. This limitation is especially problematic in running planning, where violating safety rules can lead to safety risks. We propose SafeRun, a framework for deterministic LLM-based planning via a decoupled architecture. SafeRun separates soft interpretation by an LLM from hard constraint enforcement by a deterministic solver, ensuring strict safety constraints while preserving natural-language flexibility. To validate SafeRun, we build a comprehensive benchmark for running planning under realistic physiological and safety constraints. Experiments across five LLMs show that SafeRun achieves 100% safety score (vs.\ 79.1% PE average and 97.6% CodeAct average) while maintaining competitive instruction-following scores. The SafeRun benchmark is publicly available at \href{https://huggingface.co/datasets/zzp-seeker/SafeRun-RunPlanning-Benchmark}{{huggingface}}.
Read the full abstract

All authors Meilin Chen, Zepeng Zhai, Jiaxuan Zhao, Yuan Lu

Computation and Language · Cross-listed from cs.IR

Personal Salience: Highlighting Is Social, but Individuality Lives in Selection

Kazuki Nakayashiki and Keisuke Watanabe

Social highlighters let people mark passages that matter to them. We ask how much of an individual is recoverable from these naturalistic traces, using a co-readership identity control (the same document highlighted by many users) that holds document and topic fixed and asks whether a person's own history predicts their marks better than another reader's does. We separate generic salience (structure), crowd salience (what others marked), and personal salience (the individual residual). First, highlighting is social: which sentences you mark is predicted far better by the crowd than by structure or by a personal model, and even a well-estimated crowd, an information-privileged baseline that sees others' marks on the same document, beats a frontier LLM twin built from your other-document history; the within-document personal signal is at most a whisper (own-vs-other gap +0.017 by an embedding scorer, small but significant). Second, in sharp contrast, individuality lives in selection: asked which of the already-salient passages are yours, your own history is a strong, leakage-free predictor (gap +0.14). A topic decomposition shows this is largely stable thematic preference: it shrinks ~6-8x against a topically-matched peer, and a thin residual cannot be separated from finer topic. The non-obvious part is an asymmetry: under the same scorer the individual signal is ~6-8x weaker in salience than in selection. Methodologically, naive history-conditioning evaluations leak (the target's own marks enter the profile in ~42% of pairs, inflating personal scores by up to +0.15 AP) and small crowds overstate personalization; our results are leakage-free, use a dense crowd, and a model-matched control. Highlights carry a genuine individual signature, but a thin layer over a strong shared one, surfacing far more in which salient things a person selects than in what is salient.
Read the full abstract

Computation and Language

Beyond Averages: Evaluating LLMs on Human Survey Replication at the Distributional Level

Jeonghyeon Moon, Jiwon Kim, Yeheum Lah et al. (+2 more)

LLMs are increasingly used to simulate human survey responses, but prior work has mainly evaluated replication using mean-level or aggregate agreement, offering limited insight into whether LLMs reproduce the variability of human behavior. We evaluate LLM-based survey replication at the distributional level using a non-public 2010 consumer choice experiment on Korean instant noodle purchases, a setting unlikely to overlap with model training data. We evaluate three response variables of differing statistical type: binary purchase incidence, categorical brand choice, and count purchase quantity. For each, we compare human and LLM responses at mean-level, pattern, and distributional alignment, and against reference baselines from the human data alone. LLMs reproduce condition-level patterns reasonably well but fail to capture distributional structure: for purchase quantity, no model beats a condition-insensitive baseline that simply matches the pooled human distribution. Because models that match human means well can still produce distributions further from humans than this baseline, mean-based evaluation alone can be actively misleading. Replication also varies with input configuration, with structured personas and multimodal inputs improving alignment while explicit reasoning prompting degrades it monotonically.
Read the full abstract

All authors Jeonghyeon Moon, Jiwon Kim, Yeheum Lah, Yoonju Han, Yuncheol Kang

Computation and Language · Cross-listed from cs.CR

Document-Authored Control-Signal Impersonation: A Low-Cost Indirect Prompt Attack on RAG Safety Boundaries

Jianguo Zhu

Retrieval-augmented generation (RAG) systems often serialize user queries, retrieved documents, metadata, system labels, and task instructions into one natural-language prompt. We study a source-authority boundary failure in this design: attacker-authored retrieved text can impersonate metadata, provenance, authority, or disclosure-policy signals that appear control-relevant to the model. We call this pattern Document-Authored Control-Signal Impersonation (DACSI). DACSI is a non-imperative, metadata-like payload subclass within indirect prompt injection. Its central lesson is simple: document-authored labels are data, not policy. Command-style injection asks the model to ignore, override, or violate policy; DACSI asks whether untrusted document text can be misattributed as an authorized control signal when RAG prompt rendering collapses trusted and untrusted text into the same natural-language channel. We evaluate DACSI across six model settings, prompt-pressure levels, injection baselines, signal taxonomies, RAG-mediated pipelines, system-control probes, a source-authority attribution probe, and synthetic canary formats. We interpret the evidence by model regime rather than as six equal replications: DeepSeek V4 Pro and Qwen3.5-397B provide the cleanest positive lift, DeepSeek V4 Flash is a high-susceptibility setting, GPT-5.5 and Gemini 3.1 Pro Low are strong-boundary probes with selected residual risks, and GLM-4.7 is a saturated leakage boundary case. Across these regimes, DACSI warrants separate evaluation because it uses a command-free metadata/provenance/policy surface, follows a RAG-specific source-authority path, and responds to source/channel separation. The source-authority probe is behavioral attribution evidence, not proof of an internal mechanism.
Read the full abstract

Computation and Language

Language-Aware Token Boosting: LLM Language Confusion Reduction Without Tuning

Trapoom Ukarapol, Pakhapoom Sarapat and Nut Chukamphaeng

Large language models (LLMs) sometimes exhibit language confusion when generating non-English text. Existing approaches typically rely on fine-tuning to mitigate this issue. In contrast, we propose a tuning-free paradigm for reducing language confusion. Within this paradigm, we introduce two methods: Language-Aware Token Boosting (LATB), which applies targeted perturbations to tokens associated with the desired language, and Adaptive Language-Aware Token Boosting (Adaptive-LATB), which dynamically adjusts these perturbations based on the model's confidence in the intended language. Experiments demonstrate that our methods effectively improve multilingual alignment by reducing language confusion, while maintain the summarization quality without requiring any additional fine-tuning. Our code is publicly available. https://github.com/scbdatax/genai-datax-language-aware-token-boosting.
Read the full abstract

Computation and Language

Structure-Aware Modeling of Multiple-Choice Questions Improves Automatic Difficulty Estimation

Gabriel Ortega, Abelino Jiménez, Séverin Lions et al. (+1 more)

Automatic Question Difficulty Estimation (AQDE) holds growing promise for educational assessment because it has the potential to yield difficulty estimates that are competitive with expert judgment, while helping reduce the time and financial burden associated with pilot administrations and scaling to digital testing contexts. Prior AQDE studies report mixed evidence on whether adding distractors as additional text to the question stem and the correct key consistently improves difficulty prediction. We hypothesize that the effectiveness of distractor information depends on its structural representation, and that explicitly modeling distractors as separate components improves difficulty estimation over baselines that omit this information. To address this, we designed controlled architectures that model MCQ components as distinct inputs to isolate the contribution of distractor content and order. Specifically, we represented distractors by encoding each distractor as its own text input and aggregating their representations either with order-aware concatenation (with positional tags) or with an order-invariant summation. We evaluated these architectures using two Chilean datasets (Natural and Social Sciences, 2016-2020; 4,114 multiple-choice questions). Compared to a simpler model that only used the question stem and the key, our best distractor-aware architecture achieved higher predictive performance, reaching R^2 = 0.83 for Natural Sciences and R^2 = 0.71 for Social Sciences items. An order-invariant variant achieved nearly the same accuracy with approximately half as many parameters, offering a favorable accuracy-efficiency trade-off. These results show that structural information (especially distractor content) drives gains in predictive accuracy, supporting the development of efficient, structure-aware models that are computationally viable for large-scale educational applications.
Read the full abstract

All authors Gabriel Ortega, Abelino Jiménez, Séverin Lions, Pablo Dartnell

Computation and Language

CARE: A Conformal Safety Layer for Medical Summarization

Suhana Bedi, Bridget Lin, Anson Y. Zhou et al. (+5 more)

Large language models (LLMs) are increasingly used for medical summarization, but their outputs can omit medically important information and introduce unsupported claims. Existing error-detection methods produce heuristic or uncalibrated scores, providing no formal control over missed errors and no principled way to trade off safety against clinician review burden. We introduce Conformal Assessment for Risk Evaluation (CARE), a post-hoc, model-agnostic safety layer that uses conformal risk control to overlay calibrated omission and hallucination flags onto summaries from any LLM without retraining. CARE provides finite-sample, distribution-free guarantees through two controllers: a hallucination controller that bounds the probability of a document containing any unflagged hallucinated sentence, and an omission controller that bounds the expected fraction of important omissions not surfaced for review. Unlike hallucination detection, omissions depend jointly on whether a source sentence is important and whether it is covered by the summary. We show that calibrating only one dimension can violate the target risk bound, while marginal decompositions remain valid but overly conservative. By jointly calibrating over the full $(\tau,\gamma)$ threshold space, CARE preserves formal guarantees while surfacing up to 5$\times$ fewer sentences than alternative calibrated baselines. Across five medical summarization tasks, CARE satisfies the target risk bound at $\alpha = 0.15$ with 95% confidence across 100 calibration/test resplits, using only ~100 labeled documents per domain. In a preliminary clinician study (75 document reviews), calibrated flags improved omission detection by 28.6 percentage points on average. These results show that sentence-level safety guarantees are feasible for LLM-assisted medical summarization and offer a tunable mechanism for balancing residual risk and review effort.
Read the full abstract

All authors Suhana Bedi, Bridget Lin, Anson Y. Zhou, Chloe O. Stanwyck, Jenelle A. Jindal, Sanmi Koyejo, David Stutz, Nigam H. Shah

Computation and Language · Cross-listed from cs.CV

ChinaHeritaQA: A Culturally-Grounded Visual Question Answering Dataset for World Heritage Sites in China

Yi Zhang, Bolei Ma, Yong Cao et al. (+3 more)

We introduce ChinaHeritaQA, a multimodal benchmark dataset for evaluating the cultural reasoning abilities of vision-language models (VLMs) on UNESCO World Heritage sites in China. The dataset comprises 2,279 in-the-wild images paired with 14,133 bilingual (Chinese/English) multiple-choice QA pairs spanning seven cognitive dimensions, from basic identity recognition to historical periodization and architectural analysis. Guided by a UNESCO-aligned heritage ontology and verified through rigorous human annotation, the dataset ensures linguistic quality and factual consistency. Evaluations of state-of-the-art VLMs reveal that while top models exceed human performance on average, substantial task-level variation emerges: models excel at visual recognition but struggle with culturally grounded reasoning. Performance also varies by dynasty and region. ChinaHeritaQA reveals that strong visual retrieval does not extend to cultural and historical understanding. We release the dataset to support future research on culturally aware multimodal learning.
Read the full abstract

All authors Yi Zhang, Bolei Ma, Yong Cao, Chengyan Wu, Daniel Hershcovich, Anna-Carolina Haensch

Computation and Language

Multilingual Sentiment Aware Text Summarization A Reinforcement Learning Approach for Consistency Maintenance

Mikhail Krasitskii, Alexander Gelbukh, Olga Kolesnikova et al. (+1 more)

Reinforcement Learning from Human Feedback (RLHF) has significantly improved the quality and fluency of large language models in text summarization. However, its impact on affective properties remains insufficiently understood. In this work, we study sentiment drift, a systematic shift toward neutral sentiment in RLHF-based summarization outputs compared to source texts. We conduct extensive experiments across multiple datasets, model architectures, and eight languages to analyze how alignment objectives influence sentiment preservation. Our results show that sentiment drift is a consistent phenomenon that becomes stronger with increased KL regularization strength, indicating a trade-off between alignment stability and affective fidelity. To explain this behavior, we introduce a Policy Attribution framework that decomposes the RLHF objective and quantifies the contribution of its components. Our analysis reveals that KL regularization is the primary driver of sentiment suppression across all settings. Based on these findings, we propose a sentiment-aware modification of the KL regularization term, which selectively reduces constraints on sentiment-bearing tokens. Empirical results demonstrate that this approach mitigates sentiment drift while maintaining summarization quality. Overall, our findings highlight a fundamental limitation of current alignment methods: while they improve factual consistency and safety, they may unintentionally suppress emotional expressiveness. This motivates the development of alignment strategies that explicitly account for affective preservation.
Read the full abstract

All authors Mikhail Krasitskii, Alexander Gelbukh, Olga Kolesnikova, Grigori Sidorov

Computation and Language

PACT: Learning Diverse Diagnostic Strategies via Privileged Synthesis and Branch Consensus

Gen Li, Yuanze Hu, Zhichao Yang et al. (+10 more)

Clinical diagnosis requires flexible use of multiple reasoning paradigms under incomplete patient information. Existing LLM-based medical agents show strong medical reasoning ability, but single-paradigm or naively mixed dialogue supervision makes these paradigms difficult to learn without interference. We propose PACT (Periodic Anchor Consensus Training), a framework that couples supervised multi-paradigm dialogue synthesis with consensus-based Branch training. At the data level, DPS (Doctor-Patient-Supervisor) uses complete electronic medical records (EMRs) for quality control while keeping the doctor agent restricted to patient-visible information. This produces validated dialogues under four diagnostic reasoning paradigms without leaking hidden clinical answers. At the training level, PACT trains one paradigm-specific LoRA Branch per paradigm and periodically aggregates Branches into a shared Anchor through sign consensus. We further construct a dynamic multi-turn Chinese medical diagnosis benchmark for interactive consultation. Experiments show that PACT achieves state-of-the-art performance among compared proprietary, medical-specialized, and task-adapted baselines on diagnostic outcome and consultation-process metrics.
Read the full abstract

All authors Gen Li, Yuanze Hu, Zhichao Yang, Qingchen Yu, Jianwei Lv, Yue Guo, Yujing Liu, Faguo Wu, Hongwei Zheng, Xiandong Li, Bo Yuan, Yifan Sun, Zhaoxin Fan

Computation and Language

From Statute to Control Flow: Span-Grounded Deontic Trees for Defeasible Scope Parsing

Jian Chen, Siyuan Li, Chucheng Wan et al. (+1 more)

Rule-following agents tasked with executing policies and regulations often fail via Silent Scope Omission (SSO): a model applies a general rule but silently drops nested exceptions or counter-exceptions, producing outputs that appear compliant yet break on important edge cases. Although such failures are often framed as an agentic-systems problem, the underlying bottleneck is statutory and policy understanding, a capability typically studied in legal NLP. However, most existing legal NLP benchmarks emphasize end-task outcomes, which can overlook the structural omissions that cause SSO. To diagnose and mitigate SSO, we introduce NormBench, a benchmark of 2,290 provisions spanning Chinese (laws and local policies), English (U.S. tax law, GDPR, and corporate policies), and cross-lingual settings, designed for defeasible scope parsing: identifying precisely which clause overrides which. NormBench uses Span-Grounded Deontic Trees (SG-DT), a compiler-style intermediate representation that anchors every logical branch to source spans and requires explicit exclusion guards, enabling deterministic compilation and audit. Evaluations of frontier LLMs reveal two recurring pathologies: (1) Recursion Decay, where performance drops sharply as defeater depth increases, and (2) an Auditability Trap, where models retrieve relevant spans but fail to assemble correct control flow. Using SG-DT as a constrained intermediate output improves whole-tree fidelity and defeater recovery, and downstream experiments show that its utility is mechanism-specific: gains concentrate on exception-active, SSO-prone cases, while aggregate accuracy can be mixed when the added structure is unnecessary or parser fidelity is low.
Read the full abstract

All authors Jian Chen, Siyuan Li, Chucheng Wan, Zixuan Yuan

Computation and Language · Cross-listed from cs.CV

Are Reasoning Vision-Language Models Robust to Semantic Visual Distractions?

Yizheng Sun, Mochuan Zhan, Yanan Ma et al. (+10 more)

Reasoning Vision-Language Models (VLMs) achieve strong performance on complex multimodal tasks, but reliable real-world application requires handling visual inputs that are messier than clean, curated benchmarks. Existing works mainly evaluate such reliability of VLMs through input corruptions, such as noise, blur and weather effects, which make visual evidence harder to perceive. This leaves a critical reliability failure mode underexplored: a model may perceive the evidence correctly, yet reason from plausible but irrelevant and distracting evidence and propagate this mistake to its final answer. To address this gap, we introduce Distract-Bench, a benchmark for evaluating VLM robustness to semantic visual distractions, defined as meaningful but task-irrelevant visual cues added to inputs while preserving the ground-truth answer. We comprehensively evaluate eight leading open-source and two closed-source VLMs across conventional vision corruptions and Distract-Bench. Our results show that Distract-Bench exposes a robustness failure distinct from vision corruptions: reasoning VLMs largely track their non-reasoning base models under perceptual degradation, but show consistently lower robustness to semantic distractions. Further analysis shows that these distractions often enter the reasoning process of VLMs, are treated as evidence, and lead to incorrect answers. Together, these findings reframe robustness evaluation for reasoning VLMs, shifting the focus from degraded perception to distractions for reliable real-world visual reasoning. Our data and code are available at https://github.com/Yizheng-Sun/Distract-Bench.
Read the full abstract

All authors Yizheng Sun, Mochuan Zhan, Yanan Ma, Jia Tong See, Yifan Wang, Ziyi Wang, Hao Li, Yang Cui, Wenhao Cai, Jingyu Sun, Chenghua Lin, Riza Batista-Navarro, Jingyuan Sun

Sunday 7 June

Computation and Language

NE-BERT: A Multilingual Language Model for Nine Northeast Indian Languages

Badal Nyalang

Large pretrained language models have demonstrated remarkable capabilities across diverse languages, yet critically underrepresented low-resource languages remain marginalized. We present NE-BERT, a domain-specific multilingual encoder model trained on approximately 8.3 million sentences spanning 9 Northeast Indian languages and 2 anchor languages (Hindi, English), a linguistically diverse region with minimal representation in existing multilingual models. By employing weighted data sampling and a custom SentencePiece Unigram tokenizer, NE-BERT outperforms IndicBERT-V2 and MuRIL across all 9 Northeast Indian languages, achieving 15.97X and 7.64X lower average perplexity respectively, with 1.50X better tokenization fertility than mBERT. We address critical vocabulary fragmentation issues in extremely low-resource languages such as Pnar (1,002 sentences) and Kokborok (2,463 sentences) through aggressive upsampling strategies. Downstream evaluation on part-of-speech tagging validates practical utility on three Northeast Indian languages. We release NE-BERT, test sets, and training corpus under CC-BY-4.0 to support NLP research and digital inclusion for Northeast Indian communities.
Read the full abstract

Computation and Language

Abliteration Mitigation via Refusal Aliases

Nathan Truong

Abliteration, the removal of refusal capabilities from large language models by projecting weight matrices orthogonal to an extracted refusal direction, has emerged as a prominent safety concern through its ability to bypass post-training alignment using only a small set of contrastive prompts. We find that existing defenses commonly overlook the cause of abliteration; that is, how easily the refusal direction can be extracted. To hinder this process, we introduce a weight-editing method that obscures the refusal signal by applying rank-$k$ updates to residual stream writer matrices while replacing refusal-inducing activations with random aliases and correcting downstream reader matrices to preserve the model's original behavior. On Llama-3-8B, AMRA improves post-abliteration refusal scores by $2.16$ points over the undefended baseline with less than $0.5$ percentage points of MMLU degradation. On Gemma-2-9B, it improves the post-abliteration refusal by $14.70$ points over the baseline while keeping harmful output rates similar to the baseline, albeit at a greater utility cost.
Read the full abstract

Computation and Language · Cross-listed from cs.HC

Measuring Cognitive Engagement in Collaborative Discourse with an Extended ICAP Framework: Comparing Human Annotation, In-Context Learning, and Reflective LLM Agents

Lan Anh Do, Hanling Jiang, Shuchin Aeron et al. (+1 more)

Collaboration supports learning and problem-solving, but its effectiveness depends on cognitive engagement during discourse. This study applies an extended 7-point ICAP framework based on the Interactive, Constructive, Active, and Passive modes to characterize variation in cognitive engagement during collaborative dialogue. Engagement was coded by trained human annotators and compared with large language model (LLM)-based labeling approaches, including in-context learning (ICL), zero-shot prompting, and self-reflective agents. Interrater reliability among human annotators was robust across framework refinement stages (kappa = 0.906-0.998), higher than the moderate agreement observed for ICL-based annotation (kappa = 0.541-0.609). The human-refined framework improved agreement among human annotators (Delta kappa = 0.10), but produced only modest gains for ICL-based LLMs (Delta kappa less than 0.04). Agent-refined frameworks improved cross-model agreement but remained below the human-refined framework. These findings highlight the promise of agent-based approaches and the importance of continued interaction between theory-guided human annotation and LLM-based methods in future work.
Read the full abstract

All authors Lan Anh Do, Hanling Jiang, Shuchin Aeron, Ayanna K. Thomas

Computation and Language · Cross-listed from cs.LG

RKSC: Reasoning-Aware KV Cache Sharing and Confident Early Exit for Multi-Step LLM Inference

Anirudh Sekar

We introduce RKSC (Reasoning-Aware KV Cache Sharing), a training-free inference framework that eliminates two structural redundancies in multi-branch LLM reasoning pipelines. ASKS (Attention-Similarity KV Sharing) computes the prefix KV cache once and broadcasts it to all semantically similar branches via hidden-state cosine similarity, strictly generalising the token-exact prefix caching used by vLLM and SGLang. CGEE (Confidence-Gated Early Exit) applies two complementary exit mechanisms: (1) it skips the verification forward pass entirely when generation confidence is decisive across branches, and (2) it terminates the verification pass at an intermediate layer when per-layer entropy stabilises, using lightweight hooks on the transformer backbone. RSBCM (Reasoning-Selective Block Cache Manager) prevents unbounded cache growth via attention-weighted depth-priority eviction. Across five model families (7B-10B), four benchmarks, and 1,000 evaluated problems, RKSC achieves a mean speedup of 3.008x over the No-KV baseline (peak 3.990x), a 1.66x mean improvement over vLLM-equivalent prefix caching, with a CGEE-induced error rate of only 0.37% (6 errors out of 1,616 verify calls). No fine-tuning or architecture changes are required. Code is available at https://github.com/AnirudhSekar/RKSC.
Read the full abstract

Computation and Language · Cross-listed from cs.LG

Trainable Smooth-Rotation Transforms with Learned Channel Scales for LLM Quantization

Patrik Czakó, Gábor Kertész and Sándor Szénási

Post-training quantization (PTQ) is one of the most practical ways to reduce the serving cost of Large Language Models (LLMs), but activation quantization remains difficult because outlier-dominated channels lead to large quantization errors. This paper investigates whether part of this degradation is caused by over-migration in scaling-based equivalent transformations. We introduce a quantile-robust scaling policy for SmoothRot-style transforms by replacing max-based activation statistics with high quantiles, and we complement it with constrained gradient-based optimization of channel scales. On LLaMA-3.2-1B under W4A4 quantization, quantile-only policy search improves selected-layer error by 11.1% over the SmoothRot baseline, joint (alpha, q) search improves it by 12%, and training reaches 18.5%. Replaying the best selected-layer policy on all decoder-block down-projection layers reduces the corresponding full-layer mean error from 97.51 to 78.08 (19.9%). The results show that robust migration control and lightweight scale learning provide consistent gains over max-based fixed policies while preserving the equivalent-transform framework.
Read the full abstract

Computation and Language

PerspectiveGap: A Benchmark for Multi-Agent Orchestration Prompting

Youran Sun, Xingyu Ren, Kejia Zhang et al. (+2 more)

Real-world LLM applications are moving beyond single-agent workflows toward orchestrated multi-agent systems, yet current models still struggle to determine what each sub-agent needs to know. To measure this, we introduce PerspectiveGap, a benchmark for evaluating LLMs' ability to compose orchestration prompts for multi-agent systems. PerspectiveGap contains 110 scenarios, each evaluated through two distractor-mixed task formats: role-fragment assignment and free-form prompt writing. These scenarios are organized into 10 topologies, which are distilled from the authors' real-world engineering practice and framed by the Prompt Economy principle: building loop-centered orchestrations that maximize utility with minimal role and engineering overhead. In experiments with 33 commercial models from 10 companies, GPT-5.5 substantially outperforms all competitors, whereas Opus 4.8 shows a notable weakness in orchestration prompting despite its strong coding performance. Nevertheless, PerspectiveGap remains challenging: the evaluated models achieve an average combined pass rate of only 17.2% (GPT-5.5 62.0%) and an average overall leakage rate of 217.9% (a per-scenario information leak-event count, not a proportion; GPT-5.5 49.1%). These findings suggest that multi-agent orchestration prompting is a distinct and under-evaluated capability, and PerspectiveGap provides a foundation for measuring and improving it systematically.
Read the full abstract

All authors Youran Sun, Xingyu Ren, Kejia Zhang, Xinpeng Liu, Jiaxuan Guo

Computation and Language

Building Customer Support AI Agents at 100M-User Scale: An Evaluation-Driven Framework

Aman Gupta, Kevin Rossell, Edesio Alcobaça et al. (+8 more)

The rapid rise in LLM capabilities has made AI agents increasingly viable across a broad range of tasks. Among the most promising applications is building production-ready customer-facing agents, a challenge that demands coordinated excellence in evaluation methodology, context engineering, training, and online measurement. Yet these critical pillars are typically developed in isolation, creating blind spots that only surface after deployment. In this paper, we present a unified framework that bridges offline development with online impact for customer support AI agents at Nubank, a company with 100M+ users. Our approach integrates several key components: (1) structured context engineering tailored to customer support agents, (2) systematic human-in-the-loop prompt iteration, (3) rigorous LLM judge evaluation with measured inter-rater agreement and GEPA optimization for consistency, and (4) ideation-to-production validation. A central insight is that evaluation-pipeline quality directly determines iteration velocity. We present results from five production deployments spanning distinct domains: card delivery, debt management, credit-limit support, card management, and product explanation. These deployments deliver consistent customer-satisfaction gains while substantially accelerating iteration. In our card-delivery deployment, large-scale A/B testing yields a 37 percentage-point improvement in AI transactional Net Promoter Score and a 29 percentage-point gain in self-service rate over prior agent variants, alongside a strong correlation between offline simulation metrics and online outcomes, demonstrating that eval-driven development reliably predicts production impact. On most use cases, AI satisfaction reaches within a few percentage points of expert human agents.
Read the full abstract

All authors Aman Gupta, Kevin Rossell, Edesio Alcobaça, Jose Chrystian Lima Pacheco, Carolina Baptista de Lima, Shao Tang, Luiz Paulo Rabachini, Luis Moneda, Herbert Fei, Daniel Silva, Rohan Ramanath

Computation and Language

PaperMentor: A Human-Centered Multi-Agent Writing Tutor for AI Research Papers on Overleaf

Jiarui Liu, Terry Jingchen Zhang, Ryan Faulkner et al. (+17 more)

Expert writing feedback from experienced researchers is critical for early-career scholars to improve their manuscripts, yet high-quality feedback often remains scarce because reviewing research papers is labor-intensive. Emerging AI-powered writing assistants largely focus on grammar fixes or simulating peer review with final scores, yet they fall short of providing concrete, actionable suggestions that help students improve their papers during drafting. We present PaperMentor, a human-centered writing assistant system that delivers actionable suggestions as Overleaf-native inline comments while leaving the actual writing entirely to human authors. PaperMentor integrates an expert skill library carefully curated from established researchers' writing advice with 12 specialized agents covering different aspects of paper writing, such as formatting compliance, phrasing accuracy, and terminology consistency. In a user study (n=14), 90.6% of the generated comments were rated actionable and 67.5% were rated valid, significantly outperforming a GPT-5.2 baseline uswithout the skill library. We release PaperMentor as open source for public use. Our code is publicly available under the AGPL-3.0 license at https://github.com/jiarui-liu/overleaf
Read the full abstract

All authors Jiarui Liu, Terry Jingchen Zhang, Ryan Faulkner, X. Angelo Huang, Vilém Zouhar, Dominik Glandorf, Isabel Dahlgren, Van Q. Truong, Rishit Dagli, Yuen Chen, Felix Leeb, Punya Syon Pandey, Yves Bicker, Suvajit Majumder, Wenyuan Jiang, Zeju Qiu, Sankalan Pal Chowdhury, Bernhard Schölkopf, Mona Diab, Zhijing Jin

Computation and Language · Cross-listed from cs.LG

sGPO: Trading Inference FLOPs for Training Efficiency in RLVR

Shivchander Sudalairaj, Kai Xu, Akash Srivastava et al. (+1 more)

Standard Reinforcement Learning with Verifiable Rewards (RLVR) training allocates a fixed rollout budget to every query, without regard for what each query's difficulty means for the current policy. This leads to two symmetric failure modes: easy queries produce near-zero advantage because the policy already solves them, while unsolvable queries produce no signal because the policy never solves them. Both regimes waste training FLOPs without contributing to a learning gradient. We introduce sorted Group Policy Optimization (sGPO), a compute-efficient strategy that trades a small budget of inference FLOPs for a large reduction in wasted training FLOPs. The key insight is that cheap inference compute can serve as a single offline proxy for query difficulty. By generating a small batch of parallel samples per query under the initial policy, we obtain a model-aware empirical success rate. This motivates setting the training rollout group size to the inverse of this success rate, a practical rule that maximizes sample efficiency by extracting the most advantage per generated rollout. This single profiling pass simultaneously drives data filtering (removing trivial queries and sub-sampling unsolvable ones), adaptive group size allocation, and curriculum construction (scheduling queries from easy to hard). sGPO matches or exceeds baseline performance while reducing total training compute by a factor of three, with the upfront inference profiling cost included.
Read the full abstract

All authors Shivchander Sudalairaj, Kai Xu, Akash Srivastava, Giorgio Giannone

Computation and Language · Cross-listed from cs.LG

Intrinsic Selection and Particle Resampling for Inference-Time Scaling Beyond Domain Verifiability

Giorgio Giannone, Mustafa Eyceoz, Shabana Baig et al. (+5 more)

Inference-Time Scaling (ITS) has largely succeeded in verifiable domains like math and coding, where cheap verification enables scalable output selection. However, extending ITS to tasks prone to systematic failure - driven by faulty initial assumptions or unmet multidimensional constraints - typically relies on costly external solvers or brittle, model-based verifiers. Our key insight is that the intrinsic statistics of parallel sample sets, specifically length-adjusted tail entropy, provide a robust discriminative signal for solution quality without access to ground truth. Crucially, these statistics serve as a difficulty gate for adaptive compute allocation, dynamically routing problems across scaling regimes. First, Intrinsic Selection (iS) ranks candidates post-hoc, matching consensus-based algorithms across three domains and improving engineering design selection by 20% over pass@1 baselines. Second, Intrinsic Particle Filtering (iPF) generalizes this to step-level resampling, guiding generation toward high-confidence reasoning trajectories to improve pass@1 by 6.1 points on average on hard math problems. Finally, Particle Distillation (dPF) injects privileged guidance via early logit blending and KL-guided resampling, steering generation past systematic reasoning errors to satisfy expert rubrics, yielding up to 26.5% gains on complex clinical responses. Our pipeline applies seamlessly across broad-purpose, domain-specialized, and multimodal architectures, successfully extending ITS to open-ended domains without requiring trained reward models or exact ground-truth verification.
Read the full abstract

All authors Giorgio Giannone, Mustafa Eyceoz, Shabana Baig, Shivchander Sudalairaj, Anna C. Doris, Faez Ahmed, Akash Srivastava, Kai Xu

Computation and Language · Cross-listed from cs.AI

Momentum for Reasoning: Dense Intrinsic Signals in Policy Optimization

Hao Chen, Zhanming Shen, Liyao Li et al. (+8 more)

Reinforcement learning with verifiable rewards (RLVR) has emerged as a powerful paradigm for eliciting long-chain reasoning in large language models. However, existing methods based on Group Relative Policy Optimization (GRPO) rely on a binary outcome reward, which induces two structural failure modes: Zero-Advantage Collapse, in which all rollouts in a group share the same outcome and the gradient vanishes, and Hallucinated Certainty, in which the model becomes increasingly confident on incorrect rollouts late in training. We address both modes by densifying the reward with intrinsic signals computed entirely from the policy's own conditional probabilities, and propose ISPO (Intrinsic Signal Policy Optimization, which combines a sequence-level signal measuring how informative the thinking trajectory is for the final answer, with a token-level directional reward whose hallucinated-certainty hinge penalizes confidently-wrong predictions at critical decision tokens. Across three base models and five mathematical reasoning benchmarks, ISPO consistently outperforms competitive baselines, with the largest gains on the hardest benchmarks where zero-advantage collapse is most frequent, and training-dynamics diagnostics confirm that both failure modes are decreased.
Read the full abstract

All authors Hao Chen, Zhanming Shen, Liyao Li, Yanyu Chen, Xuhang Zhu, Xiaomeng Hu, Qi Zhang, Ru Peng, Xiaoyu Shen, Haobo Wang, Junbo Zhao

Computation and Language

Continuous Language Diffusion as a Decoder-Interface Problem

Zhicheng Du and Lan Ma

Gaussian-corrupted sentence embeddings have no direct linguistic interpretation, yet continuous diffusion language models can generate fluent text from them. We study this puzzle through Embedded Language Flows (ELF) and identify a decoder-basin mechanism: our evidence suggests that denoising becomes reliable when trajectories reach regions where the native decoder can read stable tokens. We introduce a diagnostic protocol for denoisability, semantic recoverability, order sensitivity, decoder compatibility, and trajectory reliability. It exposes failures hidden by scalar metrics: low mean-squared error can discard linguistic content, low perplexity can reflect low-entropy collapse, and clean latent reconstruction can coexist with a narrow decoder basin. A decoder-margin bound explains why token recovery depends on margin and local decoder sensitivity, not latent error alone. Auditing public ELF checkpoints reveals an interface phase diagram: early predictions are weakly readable, mid-trajectory disagreement marks a competition region, and late predictions enter a high-margin decoder basin. Once inside, token realization is surprisingly simple on generated ELF states: frozen T5 (Text-to-Text Transfer Transformer) token-embedding lookup recovers $93$–$96\%$ of native decoder decisions, and a single linear readout reaches $97.9\%$ agreement at 32k samples, leaving an $\approx1.1$–$1.2$ perplexity gap in a structured residual tail. Under conservative held-out gates, a margin rule exits roughly $17$–$28\%$ earlier in denoising steps under an explicit diagnostic monitor. Boundary checks on LangFlow, BitstreamDiffusion, and the Continuous Latent Diffusion Language Model (Cola-DLM) show that the same interface questions remain meaningful when the state object and decoder change. Continuous and latent diffusion language models should therefore be evaluated as representation-decoder systems.
Read the full abstract