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

Tuesday 28 July

Computation and Language · Cross-listed from stat.ME

The Confounder Trap: Treatment-Encoding Representations in Causal Inference with Text

Marie Neubrander, Graham Tierney and Alexander Volfovsky

Estimating causal effects of linguistic properties from observational text is difficult because the same document can contain both the treatment of interest and the non-treatment textual attributes needed for adjustment. Existing approaches often learn representations from the full text to capture latent confounding, but when treatment status is itself encoded by words in the text, these representations can directly encode treatment. This creates a confounder trap: richer representations can make treated and control documents separable, inducing overlap violations even when the underlying causal problem satisfies overlap. We study latent text treatments that are encoded through lexicons or other treatment-defining lexical information, and propose masking-based adjustment representations that remove this lexical treatment signal before representation learning. We formalize representation-induced overlap failure, prove that deletion masking preserves overlap for bag-of-words/topic-model representations, and characterize replacement masking as a natural relaxation for large language models that hides treatment-defining tokens while preserving word order and context. Across simulations, masking improves overlap diagnostics, stabilizes treatment effect estimates, and reduces bias relative to adjustment methods that learn from the unmasked text.
Read the full abstract

Computation and Language

AgentGUI: An Interface for Observing and Steering Long-Running AI Agents

Xuan Zhao, Jiwoong Sohn, Qinyue Zheng et al. (+1 more)

AI agents are increasingly adept at tackling complex, long-running tasks. With the rapid surge of autonomous capabilities, human oversight is systematically lagging behind due to limited human-centered interfacing. Aiming to address this, we introduce AgentGUI, a user-friendly, locally hosted GUI for seamlessly observing and steering AI agents amid multiple concurrent, long-running sessions. AgentGUI features 1) rich agent trajectory visualizations, 2) effective manual and automated steering, and 3) integration with and coordination between open-source and frontier agent frameworks. A controlled user study demonstrates statistically significant reduction in the time it takes to identify key elements from agent traces (38% faster, p = 0.023). In a preliminary experiment, AgentGUI's automated drift prevention feature raises the task completion rate of small local agents by as high as 34pp across a 0.8B–9B model ladder (N=50 runs per model). AgentGUI is publicly available through its project website (https://agent-gui-project.github.io) and open-source repository (https://github.com/eth-medical-ai-lab/agent-gui), along with a demo video (https://youtube.com/watch?v=GSDyxN1gTF0).
Read the full abstract

All authors Xuan Zhao, Jiwoong Sohn, Qinyue Zheng, Michael Moor

Computation and Language

Evaluating Prompt Scope and Demonstration Similarity in Local LLM Machine Translation

Mihael Arcan

Large language models (LLMs) are increasingly used as general-purpose translation systems, but their behavior is usually evaluated under a single prompt shape: translate one source sentence into one target language. In practice, users may ask for one target language, for several related languages at once, or for translations conditioned on examples. This paper studies prompt scope and demonstration selection as experimental variables for local LLM machine translation. We evaluate English-to-Romance and English-to-Germanic translation on the full FLORES devtest split for nine official European Union languages. We compare three local instruction-tuned LLMs, llama3.2:3b, mistral:latest, and qwen2.5:14b, against dedicated MT baselines from OPUS-MT and NLLB-200. We test zero-shot prompting and k=5 few-shot prompting with random, lexical-similarity, and embedding-similarity demonstration selection. We also compare single-target prompts with JSON-formatted family-scope prompts that request all languages in a family at once. Results show that dedicated MT systems remain strongest overall, especially for Germanic languages. Few-shot prompting helps mistral:latest and qwen2.5:14b, but hurts llama3.2:3b; embedding retrieval is best on average for the stronger LLMs, but its advantage over random and lexical examples is modest. Family-scope prompting is feasible for stronger local LLMs but exposes structured-output failures in smaller models. These findings motivate evaluating LLM translation not only by language pair and metric, but also by prompt scope, retrieval strategy, and multi-target compliance.
Read the full abstract

Computation and Language

Robostreet Flow: A Lightweight, Ultra-Low-Drag Electric Tractor and Four-Truck Hybrid Convoy Architecture for Minimum-Cost Point-to-Point Freight

Wei Wang, Yiru Veronika Wang, Sumukh Veeramalla et al. (+2 more)

Line-haul trucking costs are dominated by three comparably sized components: energy, driver labor, and equipment. Most efficiency technologies address only one component at a time. This paper presents Robostreet Flow, a freight architecture that jointly optimizes the vehicle, convoy formation, and operating model to minimize cost per ton-mile on high-volume point-to-point corridors. The Flow platform is a battery-electric 6x4 tractor with a teardrop single-seat cab and a drag coefficient of 0.35, approximately 40% below that of conventional Class 8 tractors. A carbon-composite monocoque and structurally integrated batteries reduce net vehicle weight by 50%. A 513 kWh tractor battery and a 340 kWh powered trailer battery provide a 500-mile single-charge range. Four Flow trucks operate as a coordinated convoy with a safety driver only in the lead vehicle, while three followers operate in SAE Level 4 automated mode. Computational fluid dynamics simulations show that close following at an 8 m gap reduces follower drag coefficients by 42-48% and follower peak frontal pressure by approximately a factor of four relative to the exposed lead vehicle. A longitudinal energy model calibrated to these results predicts fleet-average consumption of 1.27 kWh/mi in convoy, compared with 1.60 kWh/mi for an isolated vehicle, for a 20.5% energy saving. Electricity cost is approximately 17% of the equivalent diesel fuel cost. Amortizing one driver across four trucks and accounting for the additional payload enabled by lightweighting reduce operating cost from 9.4 to 4.1 cents per ton-mile, a 56% reduction relative to a diesel baseline. Sensitivity analysis, a hub-to-hub operating concept, and regulatory implications are also presented.
Read the full abstract

All authors Wei Wang, Yiru Veronika Wang, Sumukh Veeramalla, Xiaohui Liang, for the Robostreet Research Team

Computation and Language

A large-scale corpus of religious radio broadcast transcripts from webstream recordings in the United States

Samuel Bestvater, Athena Chapekis, Skyler Seets et al. (+3 more)

Religious radio is a widespread but understudied form of mass communication in the United States, and content-level analysis of it has been constrained by the absence of large-scale transcript data. This Data Descriptor presents a corpus of transcribed English-language religious radio broadcasts captured from live webstreams over a one-month period in July 2025. Fifteen-minute segments were recorded on a rolling schedule from 785 distinct streams, which together rebroadcast the signals of more than two thousand AM and FM stations, yielding over 700,000 recordings and more than 60 million diarized lines of speech. Each recording was transcribed and speaker-diarized with an automated pipeline, and segmented and labeled by programming format and topic using a large language model. The corpus is organized as linked tables of stream metadata, recording metadata, and transcript lines. It supports descriptive study of religious broadcasting across regions and traditions, analysis of how social and political issues are discussed in religious media, and speech-processing research in an underrepresented domain.
Read the full abstract

All authors Samuel Bestvater, Athena Chapekis, Skyler Seets, Anna Lieb, Sono Shah, Aaron Smith

Computation and Language

Steering Instruction Hierarchies at Inference Time

Siqi Zeng, Sewoong Lee, Han Zhao et al. (+1 more)

Instruction hierarchies are a core safety assumption of language model deployment: higher priority inputs, such as system prompts, should override conflicting lower priority inputs from users or tools. Yet frontier LLMs often violate this hierarchy. We introduce V-Steer, a training-free inference time method that restores privileged influence by editing cached value vectors at prompt positions. Using direct logit attribution on the first next token prediction, V-Steer identifies heads where lower priority spans dominate privileged ones, then boosts privileged spans and suppresses conflicting lower priority spans through in-place multiplicative edits to cached V tensors. Since the method acts only on cached values, it remains compatible with fused attention backends and adds only a one time prefill overhead. Across models from 7B to 70B, this attribution guided intervention raises primary constraint accuracy from under 18% up to 92% on controlled role conflict benchmarks, and on broader instruction hierarchy evaluations substantially outperforms prompt only baselines while matching or exceeding SoTA training based methods on 3 of 4 scales of LLMs, with negligible decoding-speed overhead. The code is available at https://github.com/cindy2000sh/v-steer.
Read the full abstract

All authors Siqi Zeng, Sewoong Lee, Han Zhao, Julia Hockenmaier

Computation and Language

Characterizing Human-Likeness in AI Generated Poetry: A Zero-shot Classification Study

A. N. Biswas, T. Tabassum, A. A. Shohid et al. (+4 more)

With the advancement of AI technologies, Generative AI (GenAI) and human written text have become nearly indistinguishable. Additionally, the global standardization of AI chatbots made academic malpractice more frequent. Furthermore, existing research indicates GenAI poems are the most difficult to distinguish even without any modification thus, GenAI poems are naturally deemed human-like by modern detectors. However, the objectivity of such dissertations needs to be verified against modern detection tools but the subjectivity of poetry and the black-box nature of the modern LLMs (Large Language Models) architectures made verification of such work quite complicated. Hence, the main objective of the research is to deduce the attributes of English poetry that contribute classification and misclassification of both human and AI poems and provide corroborating or contradicting evidence to the poetry distinguishability claim. For such characterizations, we propose a Zero-shot detection pipeline with a dataset consisting of both human and AI poems to verify the distinguishability of human and AI creation and extract the aforementioned crucial attributes for accurate classification. Extraction of such attributes provides benefits in two ways: firstly, it reduces the margin of training needed as only the poems based on misclassifying attributes need to be trained and fine tuned and finally provides a critical insight to the GenAI detection dilemma to strengthen the modern detection pipelines.
Read the full abstract

All authors A. N. Biswas, T. Tabassum, A. A. Shohid, R. M. Mou, A. A. Esha, F. Sadeque, A. Ahmed

Computation and Language

Choosing Where and How to Moderate: End-to-End Trade-offs in Filter Placement and Response Rewriting

Mengya Hu, Susie Park, Suzana Ilic et al. (+5 more)

Content-moderation classifiers are usually evaluated in isolation, but deployment requires choosing where to intervene and what follows a flag. We evaluate these choices using two end-to-end customer-outcome metrics rather than component accuracy: Usefulness, the fraction of turns with a shown, non-harmful, relevant response, and Harmful Exposure, the fraction with a shown harmful response. Latency and error rates are diagnostics. We compare Input only, Response only, and Input + response hard blocking on a human-labelled product benchmark and public ToxicChat evaluation. At the evaluated operating points, Response only achieves the highest filter-only Usefulness in both settings, while Input + response achieves lower Harmful Exposure. Replacing Response only blocking with Response + rewrite recovers most blocked traffic and yields the same observed Harmful Exposure count as Response only blocking for the selected configuration; this equality is not an equivalence result. Probe routing substantially reduces conditional route-and-generation time relative to LLM routing at comparable measured outcomes. A focused output review shows how rewrites balance filter passage with usefulness by generalizing triggering language while retaining benign intent and safe redirection; some sensitive-domain outputs nevertheless omit potentially safety-relevant support information. These results support comparing moderation configurations under deployment-specific safety and latency constraints rather than applying a universal placement rule. Code and public artifacts are available at https://github.com/microsoft/mod-frontier
Read the full abstract

All authors Mengya Hu, Susie Park, Suzana Ilic, Qiong Wei, Sandeep Atluri, Myra Deng, Tucker Fross, Curt Tigges

Computation and Language · Cross-listed from cs.AI

Position: Evaluation Scores Are Perishable Knowledge Claims

Sankalp Gilda and Shlok Gilda

Evaluation methodologies for language models increasingly combine multiple signals, from automated metrics and LLM-as-judge ratings to human assessments and benchmark suite results. When these signals are aggregated via averaging, evaluation confidence can then substantially exceed the reliability of the weakest signal: a phenomenon we call trust inflation in evaluation. We argue that evaluation scores should be treated as epistemic claims with three properties: formality (human evaluation provides stronger evidence than an automated metric), scope (a benchmark result applies to the tested distribution, not universally), and validity windows (benchmark results expire as contamination accumulates and distributions shift). Several converging research traditions (chain-of-thought analysis, possibilistic logic, and algebraic theory) establish weakest-link aggregation as the conservative endpoint of a parameterized operator family controlled by a single pessimism parameter. Drawing on those traditions, and on concrete lessons from building an evaluation harness for agentic AI, we propose that evaluation results carry explicit metadata (formality tier, scope declaration, and expiration date) to make their epistemic status transparent. We illustrate the cost of mean aggregation on the public HELM leaderboard: across 54 frontier models on ten scenarios, the top-five models ranked by mean score and by weakest-link are completely disjoint.
Read the full abstract

Computation and Language · Cross-listed from q-bio.NC

Cognitive Convergence: Deep Similarities Between Large Language Models and Human Cognition

Chandra Sripada and Richard Lewis

LLMs are widely regarded as alien intelligences, systems whose cognitive operations are fundamentally unlike our own. Apparent similarities to human cognition are therefore often seen as the result of anthropomorphic projection. We argue that this framing is mistaken. LLMs clearly differ from humans in important respects, including their physical substrate, learning history, and the environments with which they interact. These differences make it all the more striking that contemporary LLM-based systems converge with human cognition on a number of principles of cognitive organization with longstanding support in cognitive science. We identify structural correspondences across five dimensions: inferential organization, computational architecture, representational structure, prediction-driven learning, and reinforcement-learning-like mechanisms supporting goal-directed action. These correspondences support a broader model of intelligent cognition in which core principles long used to explain human intelligence also characterize contemporary LLM-based systems.
Read the full abstract

Computation and Language

DuplexGen: Adaptive Synthesis of Human-AI Turn-Taking Dialogues

Takyoung Kim, Kang-wook Kim, Sang Hoon Woo et al. (+3 more)

Turn-taking is a central component of full-duplex interaction. Which turn-taking behaviors are appropriate varies with the scenario, yet current models apply a single norm regardless of context. This limitation originates in their training data: human-human speech corpora capture natural timing phenomena but provide little role grounding or scenario-specific norms, while heuristic or prompted synthesis methods inject turn-taking behaviors without basing them on human preferences. We introduce DuplexGen, a framework for generating dialogues with scenario-adaptive turn-taking by calibrating LLM predictions against a small set of slot-level human preference annotations. In six cooperative and competitive tasks, human turn-taking preferences differ systematically, and DuplexGen aligns substantially more closely with those preferences than uncalibrated prompting or training solely on generic human-human data; a full-duplex model trained on DuplexGen-generated data exhibits distinctive, human-preferred turn-taking behaviors. These results show that human calibration, not corpus scale or prompt design alone, is what allows turn-taking synthesis to be scenario-specific.
Read the full abstract

All authors Takyoung Kim, Kang-wook Kim, Sang Hoon Woo, Julia Hirschberg, Gunhee Kim, Dilek Hakkani-Tür

Computation and Language · Cross-listed from cs.AI

Probing the Origins of Reasoning Performance: Representational Quality for Mathematical Problem-Solving in RL vs. SFT Fine-Tuned Models

Antyabha Rahman, Akshaj Gurugubelli, Omar Ankit et al. (+2 more)

Large reasoning models trained via reinforcement learning (RL) have been increasingly shown to outperform their supervised fine-tuned (SFT) counterparts on mathematical reasoning tasks; Yet the mechanistic basis for this advantage remains unclear. We therefore ask, what internal representational differences enable RL models' superior performance? Our work presents two converging lines of evidence: First, linear probes trained on layer-wise hidden states reveal that RL models tend to achieve higher accuracy in predicting answer correctness compared to SFT models, indicating more linearly separable and structured representations. Second, mean ablation studies show that RL models develop a hierarchical architecture where deeper layers become progressively more critical, whereas SFT models distribute importance uniformly across layers. Together, these findings demonstrate that RL training fundamentally restructures how models represent and process reasoning problems. Finally, we analyze token-count variability under repeated sampling across problems to assess adaptive compute allocation. While we observe higher variability in some RL-tuned models than in their SFT counterparts, we see strong consistency in others, suggesting that token allocation may depend more on the overall training pipeline than on RL versus SFT alone. We believe this token-allocation variability reveals the spread of plausible on-policy reasoning, highlighting which models exhibit stable policies versus those that are under-determined, potentially non-identifiable solution behaviour.
Read the full abstract

All authors Antyabha Rahman, Akshaj Gurugubelli, Omar Ankit, Kevin Zhu, Aishwarya Balwani

Computation and Language · Cross-listed from cs.CR

GPT-Red: Automated Red Teaming via Self-Play at Scale

Eric Wallace, Christopher A. Choquette-Choo, Nikhil Kandpal et al. (+15 more)

We introduce GPT-Red, an automated red-teaming agent that is trained to discover novel prompt injection attacks against frontier LLMs. The goal of this model is to evaluate and improve the robustness of our production systems. To this end, we use it to adversarially train GPT-5.6, our most robust model to prompt injections to date. To create GPT-Red, we design a scalable self-play algorithm where the model is tasked with attacking a diverse population of simultaneously-trained defender agents. We train the model on realistic red-teaming environments using compute on the same scale as some of our largest RL post-training runs, making it the single-largest LLM safety training run ever documented. GPT-Red excels at red-teaming: it reliably breaks our past models up to GPT-5.5, it finds more successful attacks than human red-teamers, and it generalizes to held-out environments, defender models, and harnesses. In the future, we expect that as we improve the robustness of each new GPT model, it will in turn will provide better learning signal for even stronger red-teamer agents, thus unlocking a self-improvement flywheel.
Read the full abstract

All authors Eric Wallace, Christopher A. Choquette-Choo, Nikhil Kandpal, Sam Toyer, Dylan Hunn, Stephanie Lin, Yuxin Wen, Xiangyu Qi, Christopher Wolff, Zizhao Wang, Milad Nasr, Sicheng Zhu, Chuan Guo, Juan Felipe Cerón Uribe, Kaiwen Wang, Aiden Low, Kai Xiao, Kai Chen

Computation and Language · Cross-listed from cs.LG

Meta-Learned Reward Shaping for Reinforcement Learning from Human Feedback

Yunpeng Chu

Reinforcement Learning from Human Feedback (RLHF) is the standard approach for aligning large language models with human preferences, but its quality is limited by static, task-agnostic reward models. This mismatch leads to sparse learning signals and suboptimal alignment. We introduce MeRLa (Meta-Learned Reward Shaping), a principled framework that meta-learns a task-aware shaping function $\Phi(x,y;\phi)$ across auxiliary tasks before RLHF training. The learned shaping produces a composite reward that preserves policy optimality while providing task-specific learning signals. Our meta-objective combines task discrimination, entropy regularization, and potential-based conservation for stable convergence. We provide theoretical guarantees for policy invariance, analyze representation drift sensitivity, and formally address incentive misalignment from entropy maximization. Experiments on LLaMA-3-8B across four benchmarks show consistent improvements over PPO, DPO, GRPO, and DAPO, achieving a 90.8% length-controlled win rate on AlpacaEval 2.0 and a score of 9.14 on MT-Bench, with 41% less training instability. MeRLa retains its benefits when combined with process-based and rubric-based enhanced rewards.
Read the full abstract

Computation and Language

Pass the Baton: Trajectory-Relayed On-Policy Distillation

Haolei Xu, Xiaowen Xu, Haiwen Hong et al. (+5 more)

On-policy distillation (OPD) grounds token-level supervision in the student's own trajectory, yet suffers from prefix failure: once the student commits to a wrong reasoning direction, all subsequent generation builds on this deviation, producing misdirected continuations that elicit unreliable supervision and waste compute. We identify a teacher-student continuation asymmetry on failed prefixes, where the teacher tends to redirect while the student continues along the original direction, and convert it into a label-free handoff trigger in Relay On-Policy Distillation (Relay-OPD). During training, Relay-OPD constructs relay trajectories by letting the teacher briefly take over at detected trigger points to produce a teacher leg, after which the student resumes and is optimized on the resulting trajectory. A limited relay budget concentrates intervention on critical early positions while limiting departure from the student policy. With a Qwen3-4B-Instruct-2507 teacher and Qwen3-0.6B/1.7B-Non-Thinking students on eight mathematical reasoning benchmarks, Relay-OPD achieves the best or second-best results on every benchmark, outperforming standard OPD by +5.73% and the strongest baseline FastOPD by +1.49% on average for 1.7B, with consistent gains at 0.6B. Training trajectory length is reduced by over 50%.
Read the full abstract

All authors Haolei Xu, Xiaowen Xu, Haiwen Hong, Zixuan Ni, Hongxing Li, Yiwen Qiu, Weiming Lu, Yongliang Shen

Computation and Language

UniMem: Complementary Episodic-to-Parametric Memory for Boundary-Agnostic Task Streams

Siyu Xia, Chenheng Zhang, Yanting Wu et al. (+8 more)

Memory is essential for LLM agents to accumulate task experience and reuse task-specific execution strategies. However, real-world deployment over boundary-agnostic and evolving task streams exposes a fundamental stability-plasticity dilemma. External retrieval-based memory can rapidly absorb new evidence, but it often fails to internalize recurring execution patterns and incurs inference-time retrieval overhead. Parametric memory enables stable and efficient execution once learned, but typically relies on explicit task boundaries and fixed parameter budgets. Inspired by the human brain, which balances plasticity and stability through complementary episodic storage and gradual consolidation, we propose UniMem, a self-routing framework for autonomous memory management. UniMem uses learnable routing tokens as memory controllers, enabling adaptive coordination between complementary memory pathways: novel or sparse tasks are retained in an episodic buffer for retrieval-augmented execution, while recurring and reliable patterns are consolidated into expandable parametric memory. By decoupling task identification from task execution with routing tokens and parametric memory blocks, UniMem expands memory on demand without task labels during deployment or uncontrolled parameter growth. Experiments on long-horizon streaming task sequences show that UniMem consistently outperforms baselines while maintaining execution fidelity, achieving an average gain of 4.0 EM points across three backbone models.
Read the full abstract

All authors Siyu Xia, Chenheng Zhang, Yanting Wu, Haoxuan Li, Jiajun Chai, Xiaohan Wang, Guojun Yin, Wei Lin, Zhouchen Lin, Haifeng Zhang, Jun Wang

Computation and Language

Instruction-Tuned Models Locally Reuse Human Syntax More Than Humans Do

Zandi Eberstadt

Syntactic convergence (the tendency of speakers to adapt in language towards the grammatical profiles of their interlocutors) is a well-documented feature of human dialogue widely considered to operate below conscious awareness. Whether large language models exhibit analogous syntactic convergence toward human users relative to human baselines and across a broad range of syntactic constructions remains an open question. Using substitution-paradigm data in which model generations replace one speaker's turns in pre-existing human dialogues, this study measures turn-adjacent reuse of context-free grammar (CFG) rules across sixteen open-weight Llama and Gemma models (1B-70B, pretrained and instruction-tuned) at 1,901 matched positions per model. Every model showed greater CFG-rule overlap with the preceding human turn than with a sampled unrelated human prime, and in every model this actual-versus-random difference was larger for lower-frequency rules. Each instruction-tuned model also showed greater natural-output overlap with the actual prime than the human response it replaced, and all eight matched architecture pairs exhibited greater actual-prime overlap after instruction tuning. However, relative to pretrained variants, instruction-tuned outputs overlapped more with unrelated primes, showed a smaller actual-versus-random increment, and had lower conditional rule-reuse odds once target rule-set size was held constant. In exploratory analyses, each model exhibited greater mean lexical and semantic similarity to the preceding turn than the matched human responses did. Instruction-tuned models additionally produced responses with greater mean semantic similarity than their pretrained counterparts in all eight architecture pairs, whereas the lexical similarity results were more heterogeneous.
Read the full abstract

Computation and Language

Detecting Knowledge Inconsistencies Across Text, Tables, and Knowledge Graphs

Fanfu Wei, Thibault Ehrhart and Raphaël Troncy

Wikipedia and Wikidata are widely used for information access, LLM pre-training, and retrieval-augmented generation. Their knowledge is deeply connected but scattered across text, tables, and knowledge graphs. This raises a practical question: when these modalities disagree, how can we detect and explain the conflict? We study this problem as modality-level inconsistency detection. We first introduce a taxonomy of cross-modal knowledge inconsistencies, covering information granularity differences, direct conflicts, temporal changes, and KG incompleteness. We then present Kontrast, an automatic framework that uses Text-to-SPARQL and LLM reasoning to compare table-based answers with KG evidence and categorize the resulting inconsistencies. Experiments on various Table-QA datasets show that cross-modal inconsistencies are common and informative. They reveal not only true knowledge conflicts, but also missing KG structure and temporal mismatches while being limited by Text-to-SPARQL errors and noise. Our analysis shows that text, tables, and KGs can complement and correct one another through systematic comparison. Kontrast provides a practical tool for large-scale knowledge auditing and establishes a benchmark for future work on cross-modal knowledge consistency. Code and data are available at https://github.com/ECLADATTA/KONTRAST.
Read the full abstract

Computation and Language

Polistemics: Evaluating LLMs as Information Mediators in Politics & Elections

Baran Peters, Gabor Hollbeck, Robert Jakob et al. (+1 more)

As LLMs increasingly shape the political information citizens rely on, no standard exists to assess whether they do so responsibly. We introduce Polistemics, a theory-grounded diagnostic benchmark for evaluating LLMs as mediators of political information in elections. Prior work has treated this task as reproduction rather than mediation, leaving its epistemic dimensions and interaction with imperfect information unaddressed. We ground the evaluation in Epistemic Modesty, a normative standard derived from citizens' epistemic agency, and test it across controlled settings that vary the clarity, noise, and consistency of the available evidence. Applying the benchmark to three state-of-the-art LLMs across the 2025 German and Dutch elections, we find that high aggregate scores mask systematic failures. Models mediate reliably under clear evidence but break down when it is absent, vague, or contradictory, while flattening the intensity of political language throughout. These failures point to party priors, shifting with party labels and output language. Reliable mediation appears achievable, but no model delivers it consistently.
Read the full abstract

All authors Baran Peters, Gabor Hollbeck, Robert Jakob, Kevin O'Sullivan

Computation and Language · Cross-listed from cs.AI

A Cost-Effective Multimodal LLM Reasoning Framework for Question Answering over Irregular Clinical Time Series

Frank Nie, Ethan B Liu, Yuan Zhu et al. (+2 more)

Question answering (QA) over irregular clinical time series (ICTS) plays a pivotal role in a wide range of healthcare applications. Although recent multimodal time-series large language models (LLMs) have shown considerable promise in general-purpose time-series QA, they remain poorly equipped to model the sparsity, asynchrony, and irregular sampling patterns of clinical observations. To fill this gap, we propose ClinPRISM, a cost-effective multimodal LLM reasoning framework for question answering over ICTS data. First, we devise an irregularity-aware multi-scale encoder to capture sparse clinical evidence at diverse temporal scales. Then, we propose a temporal evidence distiller to integrate representations across these scales and compress them into a small number of LLM-compatible tokens. Moreover, we introduce a progressive alignment strategy that sequentially aligns the irregular trajectories with the LLM's textual embedding space. To facilitate training, we construct 30,000 clinical time series paired with multi-scale descriptions, together with 41,000 instruction-tuning instances spanning 11 tasks. Using a 4-billion-parameter LLM backbone, ClinPRISM achieves state-of-the-art performance on the held-out evaluation benchmark while using only 16 time-series tokens and achieving an average inference latency of 0.15 seconds per question.
Read the full abstract

All authors Frank Nie, Ethan B Liu, Yuan Zhu, Wei Fan, Jindong Han

Computation and Language

Evaluating Multi-Turn Multimodal Diagnostic Reasoning on Challenging Real-World Clinical Cases

Rui Yang, Weihao Xuan, Yi Lin et al. (+23 more)

Clinical diagnostic evaluation should not only assess whether models can provide correct diagnoses, but also reflect the realities of clinical practice, including progressive disclosure of multimodal information, dynamic updating of diagnostic hypotheses, and continuous refinement of clinical reasoning. However, existing evaluations of multimodal large language models (MLLMs) typically rely on single-turn or isolated tasks, making it difficult to fully capture the complexity of real-world clinical diagnosis. To bridge this gap, we developed ClinMM-Bench, the largest multi-turn multimodal clinical diagnostic evaluation benchmark to date. ClinMM-Bench contains 1,089 challenging real-world clinical cases and 3,760 medical images across eight specialties. We systematically evaluated 15 representative MLLMs using a two-level evaluation framework that assessed both diagnostic accuracy and diagnostic reasoning quality. Results showed that proprietary models achieved the highest overall diagnostic accuracy, but the proportion of completely correct diagnoses remained limited across all models. In terms of diagnostic reasoning quality, current models can identify plausible diagnostic directions but still have considerable limitations in generating reliable diagnostic reasoning. Error analysis further identified five representative failure modes: information synthesis failure, knowledge mapping error, perception error, premature closure, and visual hallucination.
Read the full abstract

All authors Rui Yang, Weihao Xuan, Yi Lin, Zhuhan Bao, Jonathan Chong Kai Liew, Matthew Yu Heng Wong, Nicolás Lescano, Nikita R. Paripati, Emily Ling-Lin Pai, Jiarui Liu, Heli Qi, Heng-Jui Chang, Benny Kai Guo Loo, Huitao Li, Kunyu Yu, Yufan Wang, Chuan Hong, Shijian Lu, Douglas Teodoro, Naoto Yokoya, Ross Koppel, Mona Diab, Hua Xu, David W. Bates, Nan Liu, Yifan Peng

Computation and Language · Cross-listed from cs.LG

Minimizing Targeted Activations: Input-Only Suppression of Evaluation-Awareness Latents in Large Language Models

Deepanshu Mody, Samarth Agarwal, Utkarsh Mittal et al. (+1 more)

Activation steering controls model behavior by editing internal activations at inference time. We study its input-side dual: optimizing a fluent prompt so that a chosen internal latent is driven toward zero, with no inference-time model access. Our target is an "evaluation-awareness" latent-linearly readable and steerable in recent work-whose control would threaten the validity of safety evaluations if models behave differently when they detect being tested. Adapting Fluent Dreaming / EPO with a negated feature term (GCG-style token optimization plus a self-cross-entropy fluency regularizer, swept over a fluency weight), we suppress the latent under five target constructions-a CAA direction, a subspace norm, an SAE feature, a single MLP neuron, and a behavioral logit-on Llama-3.2-3B and Llama-3.1-8B. The latent is robustly suppressible ($z\approx-7$), and a causally-validated Llama Scope SAE feature can be fully and selectively turned off. But our controls tell a cautionary story about the CAA direction: a placebo random direction is suppressed just as hard and shifts behavior just as far, and when we hold a real eval passage in context and optimize only a prefix, suppressing the eval-direction fails to reduce-and slightly increases-the model's behavioral eval judgment. Activation-readability, in short, is not behavioral controllability. We further find that a single MLP neuron is eval-correlated but not causal at both scales, and that scanning the real Pile yields a natural-text baseline competitive with the optimizer for the internal direction. A positive control validates our erasure detector, bounding an erasure-vs-rotation question earlier left open.
Read the full abstract

All authors Deepanshu Mody, Samarth Agarwal, Utkarsh Mittal, Dipesh Mahato

Computation and Language · Cross-listed from cs.SE

RSIBench-Data: Benchmarking Data-Centric Research for Recursive Self-Improvement

Fanqing Meng, Lingxiao Du, Qiguang Chen et al. (+4 more)

Recursive self-improvement requires turning evidence of model failures into better models. Data-centric post-training research entails diagnosing capability gaps, designing and validating training-data strategies, and learning from checkpoint feedback. Can LLM agents automate this loop? Existing benchmarks entangle research decisions with optimization, serving, evaluation, and systems implementation, obscuring agents' research capability. We introduce RSIBench-Data, a controlled benchmark of LLM agents as data-centric researchers with a fixed post-training stack. Agents iteratively revise training-data strategies for a fixed target model; training and serving use Tinker-backed services, official evaluation runs through Harbor and E2B sandboxes, and budgets are fixed across agents. We evaluate four frontier agents on six benchmarks across software engineering, terminal use, scientific question answering, and mathematics. Agents demonstrate core data-centric research capabilities: in 58.33% of settings, they improve upon the first valid attempt by refining strategies from feedback. However, improvement is inconsistent. Among searches continuing after the best observed score, 78.26% end with a lower-scoring final attempt, while the rest only recover the same peak. A strong candidate may therefore appear early or midway through a run even as later revisions fail. Trajectory analysis identifies four patterns in stronger runs: accurate hypotheses, validation-grounded supervision, behavior-aligned data, and preservation of strong checkpoints. These findings suggest that current agents can make useful data-centric discoveries but cannot yet translate feedback into consistent improvements. RSIBench-Data provides a measurable, auditable testbed for the research capabilities required for recursive self-improvement. We open-source our code at https://github.com/evolvent-ai/RSIBench-Data.
Read the full abstract

All authors Fanqing Meng, Lingxiao Du, Qiguang Chen, Ziqi Zhao, Haocheng Lu, Mengkang Hu, Michael Qizhe Shieh

Computation and Language

AI's Capability in Assisting Scientific Research in Physics, Astrophysics, and Cosmology II: Project Planning and Proposal Evaluation

Jia Liu, Veena Krishnaraj, Kateryna Vovk et al. (+18 more)

We investigate how well large language models (LLMs) can assist scientific project planning and proposal evaluation. One-page project plans were independently generated for eight expert-conceived research projects in physics, astrophysics, and cosmology by human researchers and three contemporary LLMs (ChatGPT, Claude, and DeepSeek; mid-2025 models, used with their default tool access). The resulting 32 proposals were blindly evaluated by four human reviewers and two newer frontier LLMs (Claude Opus 4.8 and ChatGPT Pro 5.5) using a four-aspect evaluation rubric. Reviewers were also asked to identify whether each proposal was written by a human or an AI. Human reviewers rated human- and AI-written proposals similarly overall, whereas both AI reviewers scored AI-written proposals about one point higher (on a five-point scale) than human-written proposals. Human reviewers correctly identified human- and AI-written proposals 72% and 79% of the time, respectively, while both AI reviewers correctly classified all 32 proposals (100%). These results suggest that current LLMs can produce project plans comparable to human-written ones in the eyes of human reviewers, but that AI reviewers show a systematic preference for AI-generated proposals. Our results suggest caution when deploying LLMs widely in proposal preparation and evaluation.
Read the full abstract

All authors Jia Liu, Veena Krishnaraj, Kateryna Vovk, Kosuke Aizawa, Adrian E. Bayer, Linda Blot, Jessica Cowell, Suyog Garg, Jonathan Grée, Anamaria Hell, Ben Horowitz, Masaya Ichikawa, Kanyuni Iemoto, Keigo Kondo, Zacharie Lorsin, Kevin McCarthy, Jamie Robinson, Miguel Ruiz-Granda, Leander Thiele, Ievgen Vovk, Mingshen Zhou

Computation and Language · Cross-listed from cs.CR

Stemma: Induced Decision Regions Reveal LLM Provenance

Keyu Zhang, Vadim Safronov and Andrew Martin

LLM provenance testing asks whether a suspect LLM belongs to the same lineage as a source. Existing black-box methods largely infer this relationship from response-level characteristics, but these characteristics may shift under adaptation or deployment even when the underlying meaning remains unchanged, weakening the reliability of provenance evidence. To address this limitation, we introduce induced decision regions by mapping open-ended outputs into a finite decision space, thereby abstracting away surface-form variation and reframing provenance testing as measuring the inheritance of decision regions. Empirical analysis shows that the source's induced regions are preserved more strongly in related models than in unrelated models. Building on this signal, we propose Stemma, a practical black-box LLM fingerprinting method that operationalises stability, robustness, and specificity as complementary probe-selection principles for reliably estimating induced decision region inheritance. Across 770 source-suspect pairs drawn from 56 public checkpoints and spanning diverse model-weight transformations, Stemma achieves 0.967 AUC and 87.8% TPR at 1% FPR, substantially outperforming four representative baselines. It further achieves 0.995 AUC and 93.5% TPR at 1% FPR on 1,260 pairs covering 91 deployment instances, demonstrating robustness to diverse inference-time deployment settings.
Read the full abstract

Computation and Language

Shieldstral

Antonia Calvi, Avinash Sooriyarachchi, Giada Pistilli et al. (+273 more)

We introduce Shieldstral, a 3B-parameter policy-adaptive multimodal safety classifier that matches or outperforms models nearly 7$\times$ its size on text safety benchmarks and sets a new state of the art on multimodal safety classification. Shieldstral formulates content moderation as a binary question-answering task. This simple formulation unifies diverse moderation tasks into a single yes/no problem, enabling heterogeneous safety datasets with divergent taxonomies to be consolidated under one training framework. We present the data construction recipe, covering curation and generation of approximately 54.1M samples and a fine-grained evaluation set to evaluate policy adaptability. Together, these enable a small adaptive model to match or outperform much larger models.
Read the full abstract

All authors Antonia Calvi, Avinash Sooriyarachchi, Giada Pistilli, Guillaume Lample, Maarten Buyl, Maximilian Augustin, Maximilian Müller, Pierre Stock, Tom Bewley, Wassim Bouaziz, Yimu Pan, Abdelaziz Bounhar, Abhijeet Somani, Aditi Kabra, Adrian Valente, Adrien Petralia, Adrien Sadé, Alan Jeffares, Albert Jiang, Aleksandr Timashov, Alexandre Cahill, Alexandre Gavaudan, Alexandre Laval, Alexandre Sablayrolles, Amélie Héliou, Amos You, André Jonasson, Andrew Bai, Andrew Ehrenberg, Andrew Zhao, Angele Lenglemetz, Anmol Agarwal, Arata Suzuki, Arjun Majumdar, Arthur Fournier, Artjom Joosen, Aylin Guliz Akkus, Aysenur Karaduman, Baptiste Bout, Baptiste Rozière, Baudouin De Monicault, Benjamin Holzschuh, Benjamin Lefaudeux, Benjamin Tibi, Bernhard Stadlbauer, Błażej Osiński, Camille Le Scao, Chaoran Yu, Charlotte Cronjäger, Chen-Yo Sun, Chris Bamford, Christian Wallenwein, Christophe Renaudin, Clémence Lanfranchi, Corentin Barreau, Corentin Sautier, Cristiana-Diana Diaconu, Cyprien Courtot, Daniel Marczak, Darius Dabert, Diego de Las Casas, Dominik Nuss, Dylan Rubini, Dzmitry Soupel, Elizaveta Demyanenko, Elliot Chane-Sane, Emilien Fugier, Emmanuel Gottlob, Erik Aas, Etienne Goffinet, Étienne Millon, Eujeong Choi, Fabian Paischer, Fabian Schlager, Faruk Ahmed, Federico Baldassarre, Filip Szatkowski, Florian Wiesner, Gabrielle Berrada, Gaëtan Ecrepont, Gaétan Lepage, Gaspard Blanchet, Gaspard Donada-Vidal, Gauthier Delerce, Gauthier Guinet, Genevieve Hayes, Georgii Novikov, Gianluca Galletti, Guillaume Breton, Guillaume Kunsch, Guillaume Martin, Guillaume Raille, Gunjan Dhanuka, Gunshi Gupta, Han Zhou, Harshil Shah, Hasan Furkan Vural, Hédi Hadiji, Hope McGovern, Hugo Cisneros, Hugo Thimonier, Indraneel Mukherjee, Ivan Cuevas Salazar, Jacques Sun, Jan Ludziejewski, Jason Rute, Jean Quentin, Jean-Hadrien Chabran, Jean-Malo Delignon, Jie Zhang, Joachim Studnia, Joep Barmentlo, Johannes Brandstetter, John Harvill, Jonas Amar, Jonas Schweizer, Joséphine Delas, Josselin Somerville, Julien Denize, Julien Tauran, Kartik Khandelwal, Khyathi Raghavi Chandu, Kilian Tep, Kush Jain, Larissa Laich, Laura Calem, Laurence Aitchison, Laurent Callot, Laurent Fainsin, Léo Cotteleer, Léonard Blier, Lingxiao Zhao, Louis Martin, Louis Serrano, Lucile Saulnier, Ludovic Ho Fuh, Luis Montero, Manon Chossegros, Marcin Możejko, Margaret Jennings, Markus Hennerbichler, Martin Alexandre, Mathieu Poirée, Mathieu Schmitt, Mathilde Guillaumin, Matthieu André, Matthieu Dinot, Matthieu Futeral, Maurits Bleeker, Mauro Comi, Max Mynter, Maxim Berman, Maxime Darrin, Maxime Louis, Melina Jingting Laimon, Mert Unsal, Mia Chiquier, Michael Pilcer, Michał Pietruszka, Michał Zając, Mikhail Biriuchinskii, Minh-Quang Pham, Minwoo Kang, Morgane Rivière, Namit Katariya, Nathan Grinsztajn, Nathan Simpson, Neeraj Aggarwal, Neha Gupta, Ola Mysiak, Oliver Leicht, Olivier Bousquet, Olivier Duchenne, Parag Jain, Patricia Wang, Patrick Blies, Patrick von Platen, Paul Jacob, Paul Wambergue, Paula Kurylowicz, Pavan Kumar Reddy, Pavel Kuksa, Philippe Pinel, Philomène Chagniot, Pierre-André Savalle, Piotr Milos, Prateek Gupta, Pravesh Agrawal, Quentin Desreumaux, Quentin Torroba, Quercus Hernandez, Ram Ramrakhya, Randall Isenhour, Ranjit Parva, Raul Perez Pelaez, Reinhard Sonnleitner, Rémi Delacourt, Richard Kurle, Rishi Shah, Rob Romijnders, Rohin Arora, Romain Sauvestre, Roman Soletskyi, Rosalie Millner, Rupert Menneer, Sagar Vaze, Samuel Barry, Samuel Belkadi, Samuel Humeau, Sanchit Gandhi, Sandeep Subramanian, Sarthak Mittal, Saskia Adaime, Sean Cha, Sebastian Kaltenbach, Shashwat Dalal, Shashwat Verma, Sherif Waly, Shrimai Prabhumoye, Siddhant Waghjale, Siddharth Gandhi, Simon Lepage, Simon Sorg, Soham Ghosh, Sophie Marbach, Srijan Mishra, Stanislas Lange, Steve Hong, Sumukh Aithal, Szymon Antoniak, Tarun Kumar Vangani, Teven Le Scao, Théo Cachet, Thibaut Lavril, Thomas Chabal, Thomas Coste, Thomas Defard, Thomas Foubert, Thomas Robert, Thomas Wang, Tianyu Zhang, Tim Lawson, Timothée Lacroix, Tobias Kronlachner, Tom Edwards, Tomas Hodan, Tuhin Das, Tyler Wang, Ulrick BLE, Umar Jamil, Umberto Tomasini, Valentin Macé, Van Phung, Vedant Nanda, Victor Jouault, Victor Letzelter, Victor Paltz, Victor Poucheret, Vincent Maladière, Vincent Pfister, Virgile Richard, Vladislav Bataev, Wen Ding Li, William Havard, William Marshall, Xinghui Li, Xingran Guo, Xinyu Yang, Yann Dreze, Yassine El Ouahidi, Yassir Bendou, Yihan Wang, Yves Martin des Taillades, Zaccharie Ramzi, Zhenlin Xu, Zsofia Csakany

Computation and Language

AngelSpec: Towards Real-World High Performance Inference with Speculative Decoding

Hong Liu, Rui Cen, Junhan Shi et al. (+10 more)

Speculative decoding accelerates large language model inference without changing the target distribution, but no single drafting structure performs best across real-world workloads. Autoregressive multi-token prediction (MTP) is a lightweight, stable proposal mechanism, whereas block-parallel diffusion amortizes drafting latency over much longer candidate sequences; the better choice depends strongly on the output distribution. We present AngelSpec, a unified training framework for MTP and block-parallel speculative decoding that addresses this heterogeneity at three levels. At the training level, rather than fitting one universal drafter to a uniform data mixture, we co-specialize structure and data: the MTP drafter is trained on diverse conversational data for high-entropy open-ended chat, and the block-diffusion drafter on code and mathematics data for longer predictable continuations. At the architecture level, we propose DFly, a block-diffusion framework combining a hybrid target-conditioning backbone with a predecessor-conditioned autoregressive head, improving target-feature utilization and intra-block dependency modeling while keeping generation parallel. At the inference level, both acceptance length and verification cost vary with domain, request, online load, and hardware, so DFly treats verification as a shared batch-level resource: it reallocates compute toward high-confidence prefixes across requests and combines expected utility with a profiled cost model to adapt verification depth online. Across the Hy3 series, DFly raises the average accepted length on Hy3-A21B by roughly 30% and attains the highest average throughput at every tested concurrency from 4 to 64, a 1.98-2.40x speedup over autoregressive decoding and 10.5-11.8% higher throughput than DFlash. We release AngelSpec to support training and extending these methods.
Read the full abstract

All authors Hong Liu, Rui Cen, Junhan Shi, Guangshuo Qin, Jiebin Zhang, Tianyu Liu, Runzhi Fan, Guoliang Zhao, Ruobing Xie, Kai Zhang, Song Liu, Guanghua Yu, Jianchen Zhu

Computation and Language

Evaluation of Adversarial Robustness in Arabic Language Models

Anwar Alajmi, Ayed Salman and Imtiaz Ahmad

The emergence of the recent outstanding capabilities of Arabic Language Models has opened doors for exposing their vulnerabilities. One of the major security risks associated with such Natural Language Processing models is adversarial attacks. These attacks can deceive the model into the wrong prediction, raising critical model security and safety concerns. This study aims to assess the robustness of five state-of-the-art Arabic Language Models under a distinct set of Arabic adversarial attacks applied at various levels of granularity and using different example generation strategies. We also explore a defense technique based on adversarial training to enhance model robustness. The results show that insertion of diacritics can reduce the accuracy of some models by 92% while maintaining a low perturbation distance. For word-level attacks, manipulating Arabic conjunctions preserves high semantic similarity scores, low perturbation distance, and leads to an accuracy degradation of up to 58%. For sentence-level attacks, paraphrasing proves its effectiveness by an average reduction of 76% in the victim models' performance. While adversarial training improves overall resilience, with MARBERT being the most robust and AraBERT showing the greatest relative gains, challenges persist, particularly against character-level noise. These findings highlight both the potential and limitations of current defense strategies in morphologically rich languages like Arabic.
Read the full abstract

Computation and Language

WorkSurface-Bench: Benchmarking Enterprise Agents on Multi-Surface Knowledge Routing

Hao Liang, Meiyi Qiang, Sizhe Qiu et al. (+2 more)

Enterprise agents often need to integrate heterogeneous knowledge sources: documents for narrative facts, tables for computation, and dependency graphs for file relationships. Existing benchmarks typically evaluate retrieval or tool use without distinguishing whether an agent first selects the appropriate knowledge sources. We introduce WorkSurface-Bench, a benchmark for evaluating this capability as surface routing. It contains 1,151 atomic tasks derived from persona-scoped Workspace-Bench-Lite workspaces, spanning document, table, graph, and cross-surface questions. Its reference answers are auditable: table answers are reproduced through executed DuckDB queries, document answers are grounded in verified text spans, and graph answers are traced to source dependency annotations. We evaluate four model backbones across six controlled agent settings, yielding 27,624 protocol-error-free trajectories. Under gold-constrained tool access, agents achieve 98.7-99.8 Route F1, while Answer remains only 56.1-75.3 percent, showing that correct surface selection is necessary but insufficient for task completion. Matched interventions further show that surface hints improve Answer for three of four models, whereas removing irrelevant tools primarily improves routing and efficiency. In an independent three-annotator audit, all 200 sampled tasks pass all six quality criteria by majority vote, with 192 receiving unanimous judgments on every criterion. We release the dataset, construction pipeline, scoring code, and agent harness at https://github.com/haolpku/WorkSurface-Bench.
Read the full abstract

All authors Hao Liang, Meiyi Qiang, Sizhe Qiu, Linzhuang Sun, Wentao Zhang

Computation and Language

SpeechLLM Meets Federated Learning for End-to-End ASR: English and Italian Case Studies

Mohamed Nabih Ali, Daniele Falavigna and Alessio Brutti

Federated learning (FL) enables privacy-preserving training of automatic speech recognition (ASR) systems across distributed data sources, yet its application to large-scale speech language models (SpeechLLMs) remains unexplored. This paper presents the first systematic study of federated training for SpeechLLM-based end-to-end ASR systems. We design a communication-efficient federated optimization strategy tailored to the unique challenges of SpeechLLM architectures, addressing high-dimensional parameter spaces, gradient communication overhead, and computational constraints in distributed settings. Through extensive empirical evaluation on monolingual ASR tasks in English and Italian, we demonstrate the effectiveness and stability of our federated approach compared to centralized training baselines across diverse acoustic conditions and speaking styles. Additionally, we conduct a comprehensive ablation study analyzing the impact of different speech encoder architectures on monolingual English ASR performance within the federated framework, providing insights into optimal model configurations for decentralized training. Our results achieve competitive word error rates while reducing communication costs, establishing practical foundations for federated SpeechLLM deployment in real-world multilingual scenarios.
Read the full abstract