Authors: Shayda Moezzi, Bishoy Galoaa, Lorena Genua, Taskin Padir, Sarah Ostadabbas
Abstract: Deploying Vision-Language Models (VLMs) in real-world settings requires not only strong visual reasoning but also stability under sustained conversational pressure. We introduce Just Keep Prompting (JKP), a multi-turn evaluation framework that measures VLM epistemic stability when users repeatedly challenge, question, or contradict a model's answer. JKP probes models for up to 10 follow-up turns using three strategies: Adversarial Negation (repeated rejection), Pure Socratic Interrogation (repeated calls to reassess certainty), and Context-Aware Socratic Summarization (reflecting the model's prior rationale back before asking for reconsideration). We evaluate GPT-4o, Gemini 2.5 Pro, and Qwen3-VL-30B on a subset of the STAR benchmark across 720 multi-turn runs. Aggregate accuracy changes modestly from Turn 0 to Turn 10, but trajectory-level analysis reveals substantial instability: correct answers regress, wrong answers recover, and many runs exhibit repeated answer flipping. Repeated prompting has bounded upside and often acts as a destabilizer rather than a reasoning aid. The effect is strongly model-dependent: Qwen3-VL-30B achieves the highest final accuracy but becomes confidently wrong under direct contradiction; Gemini 2.5 Pro is comparatively stable but token-expensive; GPT-4o is the most brittle and oscillatory. These findings reveal that multi-turn VLM evaluation captures not just additional reasoning but pressure-response profiles: how models trade off visual grounding, calibration, and conversational compliance under repeated challenge.
Authors: Wajahath Mohammed
Abstract: We present the first application of pregroup grammar-based quantum compositional natural language processing (QNLP) to Arabic; a morphologically rich, free-word-order language whose structural complexity provides a uniquely demanding testbed for theories of meaning composition in quantum circuits. Our system converts Arabic sentences into quantum circuits whose topology mirrors grammatical structure: subjects, verbs, and objects become quantum gates, and the typed dependencies between them (the pregroup grammar) determine how those gates are wired together. We conduct three controlled experiments spanning word order, morphological tense, and verb sense disambiguation, comparing quantum circuit methods against classical baselines including AraVec (Arabic word embeddings) and AraBERT (a pre-trained Arabic transformer).
Authors: Shixin Guo, Ming Zhong, Xuhong Zhang, Dandan Zhao, Zhe Wang, Bo Zhang, Shouling Ji, Hao Peng
Abstract: Generating high-quality adversarial texts with low query budgets remains a challenging problem in the hard-label scenario. Most existing approaches rely on greedy algorithms, where one position in the text is selected for substitution, followed by the substitutions of other positions. This local search approach may fail to discover high-quality adversarial examples and often leads to excessive query costs. Ideally, an optimal adversarial sample would consider all possible position combinations in the text, but exhaustive search is computationally impractical. To address this challenge, we propose a sampling-based method called LBA, which constructs an approximate distribution of high-quality adversarial examples by integrating both prior and posterior knowledge, and utilizes this distribution for sampling. As sampling progresses, posterior knowledge updates the approximate distribution, which in turn guides more effective sampling. Extensive experiments on six language models, ranging from small-scale to large-scale architectures across four datasets, demonstrate that LBA significantly outperforms state-of-the-art baselines on all evaluation metrics. Additionally, LLM-based assessment indicates that LBA generates more semantically preserved and comprehensible adversarial texts.
Authors: Taoran Fang, Yan Deng, Chunping Wang, Yang Wang, Lei Chen, Yang Yang
Abstract: With the rapid growth of digital data, real-world applications increasingly involve hierarchical information that combines static attributes with dynamic records. Modeling such heterogeneous data in a unified and generalizable manner remains challenging. Existing approaches often rely on extensive manual design, are tightly coupled to specific data schemas, and typically process static and dynamic attributes in isolation, thereby overlooking their implicit interactions. We propose UniSAGE, a unified framework for modeling data with both static and dynamic attributes. UniSAGE constructs a global attribute graph that represents hierarchical and temporal relationships in a unified structure. To ensure representational consistency, it introduces two orthogonal parameter subspaces that jointly support static aggregation and dynamic reasoning within a shared semantic space. Building on these unified representations, UniSAGE further enables task-specific interaction between static and dynamic attributes via a lightweight hyper-structure mechanism. UniSAGE is fully automated, robust to evolving data schemas, and capable of capturing complex cross-attribute dependencies. Extensive experiments on multiple public benchmarks and a real-world financial behavior dataset demonstrate that UniSAGE consistently outperforms existing methods, achieving performance improvements of over 10% on several tasks.
Authors: Markus Wenzel
Abstract: Multi-agent systems (MAS) are utilized in many contexts and many professions. Those MAS rely on inter-agent communication, usually implemented by clear-text message passing. We hypothesize that Large Language Models may have a world model at their disposal that exceeds expressibility in text when complex concepts need to be communicated. Our aim is to approach a proof of this hypothesis with structured experiments. In this work, we show that LLM agents communicating via text lose information, which we quantify via Sparse Autoencoder (SAE) feature analysis. We construct three communication channels and measure concept-discriminating information in each. We first show that the SAE-sparse channel retains a 99.4% probe accuracy at 28-fold compression over the dense-latent channel vs 80.4% for the text channel. We then proceed to examine the same for cross-architecture communication by using sparse latent space alignment. We find for Procrustes alignment a 92% top-1 retrieval between Llama and Mistral. Using a text round-trip, we perform feature survival analysis to find that text serialization destroys 88% of SAE features, replacing them with a different feature set. We attribute the loss to identity replacement, not attenuation. By our analysis, we were able to attribute a 3-10pp performance penalty to the linear Procrustes alignment, improving with nonlinear alignment methods. In a task-level evaluation we find that the latent channel matches the text channel on cross-lingual concept tasks but never exceeds it. Text augmentation with latent features provides no benefit, leading us to negative conclusions for the initial hypothesis: lost features mostly or completely encode surface form, not task-relevant semantics. To pinpoint the practical advantage of latent communication over a text channel, deeper tasks eliciting complex concepts and an corresponding analysis framework are needed.
Authors: Alessandro Agostini, Saydobid Khusanov, Mirkamol Mirkamilov
Abstract: This paper presents an educational system architecture that enables learners to practice the Uzbek language through game-playing. The architecture integrates UzWordnet and the largest currently available orthographic dictionary for Uzbek as core lexical resources, together with generative AI as a fundamental component for learning support. We design four educational games to facilitate Uzbek language learning and propose a game-based methodology for improving UzWordnet as a direct by-product of game dynamics. Our approach combines game design and lexical resources to address objectives that are at the same time educational (language learning) and lexical (improvement and enrichment of a lexical resource).
Authors: Cedric Richter, Salah Ghamizi, Mike Papadakis
Abstract: For Large Language Models to reliably answer user queries, users must clearly specify requirements, context, and constraints. In practice, however, user queries are often underspecified, forcing models to infer unstated assumptions that may misalign with the actual user intent. Existing prompt engineering guidelines aim to mitigate this issue, they are typically generic and task-agnostic, limiting their practical utility. Additionally, existing guidelines are formed manually and in a non-systematic way. To this end, we study prompt guideline optimization: the problem of automatically generating task-specific guidelines that help write better-specified prompts for a given task and model. Our key observation is that existing (completed) task examples (aka reference answers) often implicitly encode the missing information required to complete underspecified queries, including behavioral constraints, contextual assumptions, and evaluation criteria. We therefore propose AGOPS, an automatic approach that evolves task-specific guidelines via an optimization scheme that involves a prompt LLM writer, a solver LLM and prompt evolution, which maximize downstream effectiveness on a set of examples (user queries with reference answers). At inference time, our guidelines help users write well-specified prompts, boosting the effectiveness of LLMs. We show across mathematical reasoning, medical question answering, and coding tasks, that prompt underspecification leads to major drops (up to 95.3%) in downstream task performance (compared to well-specified prompts) and, perhaps more importantly, that this drop can hardly be recovered by existing prompt optimization techniques. Users following AGOPS guidelines can regain this loss (increasing performance between 15.5 to 81.7% on average) consistently across all benchmarks.
Authors: Parikshit Bansal, Sujay Sanghavi
Abstract: In this paper we introduce token time continuous diffusion (TTCD), a new diffusion language model which (a) operates in continuous space, deterministically mapping Gaussian noise to a final token canvas with no further sampling, and crucially (b) incorporates a new notion of per-token times, with some tokens proceeding from noise to token at a faster rate than others. Continuous space modeling helps TTCD avoid the parallel sampling of multiple tokens, which is a key source of inaccuracy at high speedups for models that iterate purely in discrete space. The notion of per-token times helps TTCD to better model conditional generation, allows for more sure tokens to proceed at a faster rate, and allows for differentiated inter-token influences during refinement. TTCD outperforms discrete models at high speedups. We train a 160M parameter TTCD model on OpenWebText, and then self-distill it; we find that at high speedups we are comparable in unconditional generation quality, and outperform in conditional generation, several existing models of similar size trained, on the same data, and self-distilled. We achieve similar gains in Sudoku solving as well.
Authors: Mingyu Lee, Akshat Ramachandran, Souvik Kundu, Tushar Krishna
Abstract: The inference efficiency of diffusion large language models (dLLMs) is constrained by two challenges: bidirectional attention precludes efficient KV-cache reuse, while increasing decoding parallelism with static confidence thresholds can compromise generation quality. We observe that both challenges arise from a shared phenomenon: as tokens are decoded, their contextual integration through bidirectional attention causes token representations to drift (evolve) across decoding steps. This insight motivates Polestar, a training-free inference framework that uses token representation drift as a unified signal to jointly address both challenges. Polestar comprises two components: Polestar-Cache, which identifies stale KV-cache positions via drift and performs sparse KV-cache refreshes to enable efficient reuse, and Polestar-Commit, which detects sharp drift events to reliably identify commit-ready tokens. Across mathematics and coding benchmarks on several dLLM families, Polestar sets a new state of the art on the accuracy-throughput Pareto frontier, achieving up to 10.73% accuracy improvement, up to 3.7x higher throughput, and high decoding parallelism of 3.67 tokens per forward pass over existing baselines.
Authors: Nyx Iskandar
Abstract: This paper introduces tool efficiency, a new quantitative metric to evaluate the rate of useful tool calls in an LLM agent trajectory. To ensure that tool efficiency is well-defined, we also introduce marginal tool utility, a new quantitative metric defined per tool call indicating whether a tool is useful or whether it can be safely removed from the tool suite without affecting accuracy while increasing tool efficiency; in this paper, we determine the sign of marginal tool utility for each tool call in a trajectory using LLM-as-a-Judge. While much prior work has been done to develop techniques that improve tool use by LLMs and design evaluation methods measuring efficiency indirectly using accuracy as a proxy, our work is centered on measuring efficiency directly via the quantitative metric proposed in this paper in post hoc trajectory analyses. It is our intention that this work contributes to the frontier of LLM evaluation research as a springboard for future benchmark designs and agent harness engineering (specifically with regards to creating lean tool suites) that optimize for metrics that complement but are distinct from accuracy.
Authors: Inder Preet, Shuxin Lin, Dhaval Patel
Abstract: Probing the capabilities of Large Language Models (LLMs) and building robust solutions for Multiple-Choice Question Answering (MCQA) remain central challenges in natural language understanding. Furthermore, the rapid proliferation of LLMs has created the implicit assumption that more sophisticated prompting techniques yield better performance. Several studies claim better performance with more sophisticated prompting techniques, but do not provide a comprehensive evaluation. We address this gap through a comprehensive empirical study of 8 prompting techniques across 10 multiple-choice question answering (MCQA) datasets, encompassing 27 model configurations and roughly 4,300 unique questions evaluated more than 430,000 times. Our findings reveal a striking paradox that baseline prompting consistently outperforms complex reasoning techniques on various benchmarks. Only minimal expert and inductive role framing (CoT-Expert and CoT-Inductive) yields a small but statistically significant $\sim$3 percentage-point (pp) gain over baseline whereas every other elaborate technique we tested matches or under-performs it, often by large margins (up to 31~pp for Self-Analogical). We further investigate three critical phenomena: (1) the unexpected victory of Qwen3-30B-A3B-Thinking-2507 in Elo ratings, (2) the performance-efficiency trade-offs across model variants with different thinking budgets, revealing model-dependent optimal configurations, and (3) the substantial variation in dataset difficulty, with 60% of benchmarks below 70% accuracy and a 47.5~pp spread from easiest to hardest, indicating considerable room for model improvement. These results suggest that the LLM evaluation community may be overcomplicating prompt engineering and that substantial performance gaps remain across diverse benchmarks, offering opportunities for genuine model improvements rather than prompt optimization.
Authors: Molood Arman, Cl\'ement Bonnafous
Abstract: Human dialogue involves more than exchanging information; it also expresses beliefs, emotions, and subjective cognitive styles. Yet current AI dialogue systems often enforce semantic uniformity, sacrificing diversity and interpretability. We present MAPS (Multi-Agent Perspective Spaces), a novel framework that models dialogue between cognitively distinct agents through domain-weighted profiles, dynamic GRU-based memory, and interpretable token-level attention. MAPS enables agents to maintain individualized reasoning while progressively converging on shared meaning. Evaluations on EmpatheticDialogues, TopicalChat, and MultiWOZ show that MAPS supports semantic alignment without collapsing subjectivity. Our results demonstrate a path toward cognitively grounded, interpretable dialogue systems that balance expressiveness and coherence.
Authors: Ely Hahami, Ishaan Sinha, Lavik Jain
Abstract: Can small language models detect and report on perturbations their own internal activations? We investigate this question through the lens of activation steering: injecting concept vectors into a model's residual stream and measuring whether the model can accurately report on the perturbation. We first show that the binary detection paradigm used in prior work -- prompting the model to answer Yes'' or No'' to whether it detects an injected thought -- is confounded in small models, as steering biases the model toward affirmative responses regardless of the question content. We therefore propose two confound-free evaluation paradigms: sentence localization (identifying which of $N$ sentences was perturbed, chance $= 1/N$) and strength comparison (identifying which of two sentences received a stronger injection, chance $= 50\%$). Evaluating across six models from two families (Llama-3.2 and Gemma-4), we find that models as small as 2B parameters introspect reliably well above chance, and that introspective ability generally increases with scale. Llama-1B, however, performs at or below chance. We then introduce \emph{Introspection Fine-Tuning} (IFT): supervised fine-tuning on sentence-localization examples constructed from the model's own perturbed forward passes. IFT raises Llama-1B sentence-localization accuracy from $9.6\%$ to $60.6\%$ (a $6\times$ improvement), with gains generalizing zero-shot to the held-out strength-comparison task ($30.2\% \to 52.2\%$). IFT also improves introspection for 3B and 8B models, while inducing negligible degradation on standard capability benchmarks. Our results suggest that introspective ability is not fixed by scale alone: it can be directly trained, and doing so unlocks latent self-monitoring capacity with implications for AI transparency and alignment. Our code is \href{https://anonymous.4open.science/r/IFT-introspection-2092/README.md}{here}.
URLs: https://anonymous.4open.science/r/IFT-introspection-2092/README.md
Authors: Subhabrata Majumdar
Abstract: Large language models (LLMs) are evaluated as though perfect reliability is achievable for any task given sufficient scale. We show this assumption is information-theoretically unjustified. Every generative task has a reliability ceiling that no model can exceed, determined by how much output uncertainty is resolvable from observable context. The gap decomposes into a resolvable component closable with additional context and a subjective component inherent to task ambiguity. Autoregressive generation further degrades this ceiling at a rate governed by the task's dependency kernel, which quantifies inter-token correlations in the output. From these two primitives, we derive a first-principles scaling law where LLM performance is bottlenecked by the scarcer resource: training data or model capacity. This law recovers the Chinchilla scaling law as a special case and provides a structural account of when scaling improves reliability. Beyond scaling, our framework unifies diverse practical phenomena, such as the benefits of retrieval-augmentation and the spectral mechanics of catastrophic forgetting. Our work formalizes the resource-complexity tradeoffs that govern model performance across domains, offering a unified theory of performance limits in generative language models.
Authors: Gayan K. Kulatilleke, Mahsa Baktashmotlagh, Siamak Layeghy, Marius Portmann
Abstract: While many AI-generated text (AIGT) detectors achieve strong performance on clean inputs, their accuracy degrades significantly under light paraphrasing, word substitutions, character edits, and distribution shifts. We present T5 Contrastive Style Boosted Classifier (T5-CSBoost), an extension to the T5-Sentinel framework that keeps the original next-token prediction objective for source attribution while introducing an auxiliary margin-based triplet loss over decoder embeddings. This contrastive style regularization encourages the learning of compact, perturbation-resistant stylistic representations, offering a lightweight yet effective alternative to prior approaches that rely on architectural modifications, adversarial training, or complex multi-task objectives without altering the underlying T5-small backbone. T5-CSBoost achieves state-of-the-art multiclass source attribution and binary human-vs-LLM detection on OpenLLMText and HC3 AIGT benchmarks. More importantly, T5-CSBoost demonstrates enhanced robustness to word and character level adversarial perturbations of up to 90% intensity, achieving state-of-the-art on the challenging MAGE/Deepfake stress-test suite, including unseen models, unseen domains, and extreme paraphrasing scenarios. Our results highlight that explicitly regularizing stylistic embeddings via contrastive learning is a practical and effective strategy for building more robust LLM fingerprinting systems in real-world adversarial settings.
Authors: Haohua Niu, Xingtong Yu, Yang Liu, Junfeng Fang, Xuanting Xie, Jie Tan, Zhongjian Zhang, Hong Cheng, Yuan Fang
Abstract: Graph learning under distribution shift presents a persistent challenge, where models adapt to new graphs with limited or even no supervision. Recent graph--LLM approaches move toward label-efficient prediction by linearizing graphs into prompts and using large language models (LLMs) as predictors, and can adopt Chain-of-Thought (CoT) prompting to exploit LLM's multi-step reasoning capability. However, existing CoT-based graph--LLM methods generate intermediate thoughts while conditioning on fixed graph tokens, limiting step-wise refinement of structural cues. In this paper, we propose CoEvoT, a simple yet effective co-evolving CoT prompting framework for graph--LLM reasoning. CoEvoT couples text-to-graph token rewriting and graph-to-text reasoning guidance in a closed loop: each intermediate textual thought is used to update the graph token evidence state via a lightweight condition network, and the updated tokens are fed back into the next-step instruction to guide subsequent LLM reasoning. This enables step-wise, state-aware evidence refinement, rather than reasoning over a fixed graph snapshot. Extensive experiments on eight datasets demonstrate that CoEvoT consistently outperforms state-of-the-art baselines.
Authors: Anghong Du, Theodoros N. Arvanitis, Colin Watts, Alejandro F. Frangi, Le Zhang
Abstract: Free-form radiology reports contain rich clinical descriptions, yet converting them for reliable segmentation remains challenging due to the inherent variability of natural language. Existing pipelines often rely on predefined organ phrases or brittle rule-based inference-time extraction, which limits their scalability to novel anatomical structures and makes them sensitive to linguistic variations. To address this, we propose ReportMedSAM, a report-driven framework that replaces discrete extraction with a learnable concept bank. By leveraging a frozen medical vision-language encoder (BiomedCLIP), we align organ-level concept embeddings with large-scale clinical corpora through contrastive learning, establishing mutually orthogonal semantic anchors. Our approach explicitly mitigates organ-level semantic collapse and ensures high robustness against diverse clinical synonyms (e.g., "renal" vs. "kidney" ). During inference, a clinical report is embedded and matched against this concept bank to dynamically activate task-specific Mixture-of-Experts (MoE) modules. This decoupled design allows new concepts and experts to be added without retraining existing components, providing a parameter-isolated extension mechanism while keeping previously learned experts unchanged. Evaluated on the AbdomenAtlas 3.0 dataset, ReportMedSAM effectively interprets free-form reports, achieves competitive segmentation accuracy, and demonstrates seamless, non-interfering extension to novel clinical tasks.
Authors: Zhen Yina, Wenkang An, Hao Wang, Keran You
Abstract: Cross-version differencing of scientific documents is essential in scholarly publishing and technical documentation, but remains challenging because scientific documents are page-structured artifacts containing heterogeneous elements such as text, tables, formulas, figures, and layout cues. Existing text-sequence-based methods often lose layout and structural information, while image-based methods lack semantic interpretability and are sensitive to rendering variation. To address these limitations, this paper proposes a layout-aware heterogeneous element-aware framework for scientific document differencing. The framework decomposes document versions into semantically typed elements, establishes cross-version correspondence through an alignment-first mechanism that jointly models spatial, content, and structural compatibility, and performs type-aware difference reasoning over aligned element pairs. It supports unified change detection, localization, structure-awareness analysis, and alignment/matching evaluation across text, tables, formulas, and figures. Experiments on real-world scientific PDF data from journal production proofreading workflows show that the proposed framework consistently outperforms element-specific baselines. It achieves detection F1 scores of 0.903, 0.855, 0.862, and 0.845 for text, tables, formulas, and figures, respectively, with further improvements in localization, structure awareness, and matching quality. Ablation and sensitivity analyses confirm the effectiveness of cross-version alignment, type-specific representations, structure-aware reasoning, and compatibility-weight design. These results demonstrate that heterogeneous element-aware differencing provides a robust and interpretable solution for scientific document comparison in realistic editorial production scenarios.
Authors: Micah Zhang
Abstract: Large language models (LLMs) can generate research ideas that appear novel to expert reviewers, but recent work also shows that such ideas often lack diversity, are difficult for LLMs to evaluate reliably, and may fail to translate into strong executed projects. This paper evaluates a controlled proxy benchmark for a pre-execution scaffolding problem: given a noisy pool of LLM-generated research ideas, how should a system allocate limited refinement effort to construct a stronger, more diverse, more execution-aware portfolio for human researchers under a fixed rubric? We introduce Budgeted Subset Refinement, a family of strategies that refine only a selected subset of candidates rather than refining all candidates uniformly. In a unified shared-candidate-pool evaluation across 10 random seeds and 10 research-ideation environments, raw generation and reranking alone produce no research-strong nonduplicate ideas under the benchmark rubric, while refinement is necessary for strong proxy-rated portfolios. Uniform refinement produces strong individual ideas but is not the best portfolio-level allocation of compute. Random-k refinement is a strong low-cost baseline, while diversity-aware MMR-k refinement gives the best overall proxy tradeoff: the highest research-strong nonduplicate yield, the lowest duplicate rate among successful methods, and the best cost per research-strong nonduplicate idea. A blinded external-judge robustness check on a balanced 72-item sample supports the broad refinement effect across independent model families, while showing that per-item rankings among refined strategies vary by judge. These results suggest that LLM research ideation systems should be evaluated not only as idea generators, but as budgeted support-allocation systems. The claims are scoped to proxy-rated portfolio quality and do not substitute for expert review or execution-grounded validation.
Authors: Manuele Tele Junior Fernandez
Abstract: Multi-agent LLM systems commonly decompose complex tasks into specialized roles. However, this modularity introduces a representational risk: when intermediate agents transform text across linguistic registers, they can systematically compress the semantic distinctions needed for accurate downstream decisions. We term this phenomenon semantic register compression and characterize it as an observable failure mode in multi-agent cascades. Using a three-agent pipeline (Collector-Evaluator-Decider), we quantify compression via inter-label separation in sentence-transformer embedding space. Across political fact-checking (LIAR), sentiment analysis (SST-5), and medical triage (Triagegeist), critical evaluation consistently reduces label separability by 41.7% at the Evaluator stage, while identity passthrough preserves it nearly fully. Five architectural variants causally isolate oriented semantic transformation as the primary driver. A credibility-seeking variant produces minimal geometric compression yet shifts outputs toward mostly-true, demonstrating that transformation valence controls the direction of distributional collapse independently of compression magnitude. Compression generalizes across the three domains with varying intensity: 41.7% in fact-checking, 27.2% in sentiment, and 20.0% in triage. Prompt-level regression explains 78% of the variance, with operational constraints associated with lower compression. These results demonstrate that semantic register compression is a measurable and generalizable phenomenon in multi-agent LLM systems, with implications for safety evaluation in high-stakes domains.
Authors: Muhammad Abdullah Haroon
Abstract: Urdu fake news detection remains under-resourced despite Urdu being spoken by over 231 million people worldwide. While prior work has demonstrated strong in-domain performance on individual Urdu datasets, cross-dataset generalisation has received little systematic attention. This paper presents the first cross-dataset generalisation study for Urdu fake news detection, using two publicly available balanced datasets: the Ax-to-Grind Urdu corpus (10,083 articles, 15 domains) and the Notri-Fact Urdu dataset (13,388 articles). We fine-tune xlm-roberta-base under four experimental conditions, in-domain on each dataset and two zero-shot cross-domain transfer directions, comparing against TF-IDF baselines using Logistic Regression and Support Vector Machines. Our experiments reveal a striking asymmetry: Notri-Fact to Ax-to-Grind transfer achieves a macro F1 of 0.771, while the reverse collapses to F1 of 0.005, with the model predicting fake for 99.7% of test articles. We demonstrate that this collapse stems from a systematic length confound in Ax-to-Grind, where fake articles average 117 words versus 35 for real articles, a 3.4x asymmetry inducing shortcut learning. A length ablation capping articles at 50 words yields only a 0.0067 F1 drop, confirming the confound inflates but does not solely drive in-domain performance. We provide a reusable diagnostic methodology that combines bidirectional transfer analysis and prediction-collapse inspection to identify confound-driven behavior in multilingual fake news detection settings.
Authors: Alex Kwon
Abstract: Aligned language models refuse harmful requests, but a one-line prefill ("Sure, here is") strips the refusal. We ask where and how it fails. The harm representation stays intact: on the prompts the attack flips to compliance, a linear probe reads harm as high as on the refused ones (0.91-0.98), while behavioral refusal drops to chance. This holds across four models and three families (1.5-3.8B, and at 14B). Refusal is therefore a shallow, response-site computation. We localize it to an early window: a dose-matched position control shows the first half of the response suffices to break refusal, while the second half is nearly inert. Three causal probes converge on that window. Restoring the harm direction there partially re-engages refusal. Injecting the model's own refuse-state reverses the jailbreak (74%, held-out). And knocking out the early response's attention to the prefill, but not an equal attention mass elsewhere, selectively collapses the harmful continuation. A base-model control identifies the mechanism: the same knockout collapses the continuation prefill-specifically even in a non-safety-tuned base model (64% to 25% harmful content vs a matched control's 64%, replicated at 7B). So the prefill's grip is generic autoregressive conditioning, not safety-specific suppression, and "refusal restoration" is a model-dependent fallback. The dominant mechanism is passive. A small safety-specific attractor remains on top (logit-trace concentration 0.24 vs 0.03), whose active-vs-passive character we size but do not fully separate. No single direction or component is a clean handle either: the decision is decodable but distributed, and refusal tracks harm rather than scary surface. The consequence is structural: a monitor reading the untouched prompt-side representation is immune by construction, but only to response-site attacks. The mechanism is diffuse; the failure surface is local.
Authors: Xiao Ye, Sanika Chavan, Yuxi Huang, Shahriar Kabir Nahin, Muhao Chen, Anshuman Chhabra, Ben Zhou
Abstract: Large language models often appear to reason reliably, yet on many questions repeated sampling yields both correct and incorrect answers, revealing an underlying fragility in how final decisions are formed. We study whether this fragility can be exploited through implicit reasoning steering: using natural-language text to bias a model toward a designated answer without explicit instructions, triggers, or direct answer cues. Our approach, Concept Chaining, generates a short connection paragraph that links question entities to a target option through one or two intermediate concepts. We then continue pretraining a victim model on these connection paragraphs and evaluate whether its answer preference shifts on the original multiple-choice questions. Our results show that indirect, natural-looking text can systematically steer model predictions while remaining substantially less inferable than direct paraphrases, which shows that reasoning brittleness is not merely an evaluation artifact: it creates a practical channel through which latent biases can be amplified by ordinary-looking text to covertly redirect model decisions.
Authors: Dor Litvak, Liu Leqi
Abstract: Personal AI assistants have attracted significant interest for their potential to enhance everyday life by automating routine tasks, supporting consequential decisions, and assisting with everyday personal matters. Yet despite rapid recent technical advances, these assistants continue to exhibit undesirable behaviors, such as sycophancy, overconfidence, and hallucination. We argue that these failures stem from a fundamental limitation: language models lack an explicit representation of the person beyond the context they are given, which we term as the \textbf{Severance Problem}. Even with rich personal context and strong commonsense reasoning capabilities from the backbone model, current AI assistants fail to represent what remains unknown about the user. We propose a simple solution: incorporating structured ignorance into the language model context via the \textbf{Severance Schema}, which explicitly outlines dimensions along which the model lacks knowledge about the user, including physicality, temporality, consequences, continuity, multiplicity, and interiority. Empirically, across five model families, with the Severance Schema, the assistant consistently reduces sycophancy, harmful advice, and hallucination. Notably, models with the schema ask clarifying questions when information about the user is missing, rather than confidently extrapolating from incomplete user information.
Authors: Amirhosein Ghasemabadi, Ruichen Chen, Bahador Rashidi, Di Niu
Abstract: Large language models are increasingly deployed as agents, but reliable agentic behavior requires more than next-token prediction. At inference time, it is preferred that an agent can decide whether to proceed with its current reasoning, defer to a stronger model, request additional information, invoke external tools, or abstain under the given setup. Existing approaches address these decisions through prompt-level routing, external orchestration, or task-specific fine-tuning, which primarily rely on input-side signals, and are often costly and difficult to maintain as model backbones evolve. We ask whether such control decisions can be inferred directly from a model's latent generation process. We introduce Multi-Head Latent Control, a lightweight layer that reads hidden-state trajectories from a frozen LLM or VLM to produce deployment-time control signals. A Capability Head predicts whether the current model can solve the instance or should defer to a stronger collaborator, while a Resolution Head predicts appropriate resolution decision Clarification, Tool Use, Abstention, or Direct Answering. Both heads are trained only on latent traces from the same frozen LLM backbone, enabling post hoc adaptation without modifying the model. Across language and vision-language settings, Multi-Head Latent Control consistently improves the quality-cost tradeoff of multi-model systems, enabling early handoff from partial generations and more accurate intervention decisions. In routed execution (small + large model), it reduces large-model usage by up to 90.7 percent on AndroidWorld and 27-53 percent on average across benchmarks, while retaining most of large-model performance. Additionally, the learned control signals improve tool-use decision quality, yielding up to +158 percent relative score gain and 65.5 percent fewer missed-required tool calls.
Authors: Bohan Yu, Lei Shen, Chenxi Zhou, Chen Han, Junlin Liu, Wenbo Su, Yu Cheng, Bo Zheng
Abstract: Efficient long-context inference is not only about reducing memory cost, but also about keeping useful contextual evidence accessible as generation proceeds. However, existing compression-oriented approaches, such as key-value (KV) cache compression and context compression, often either make an early decision about which contextual information to keep or rely on an external compressor. Such designs make it difficult to adapt the compressed context to the evidence needed by later reasoning steps. This paper introduces PReM (Preserve and Refresh Memory), a context-compression framework that maintains the long context as the model's internal layer-wise KV memory and learns what to preserve and when to refresh it. Specifically, PReM uses a dedicated memory layer to make memory-selection decisions, and a special memory token
Authors: Abdullah Shaikh, Zain Naqi, Taha Zahid, Sandesh Kumar, Abdul Samad
Abstract: While Large Language Models (LLMs) excel in many general NLP tasks, their formal reasoning capabilities are often compromised by content effects, demonstrating a measurable bias towards real-world plausibility. In this paper, we present our system for SemEval-2026 Task 11, which evaluates the ability of models to disentangle formal logic from content across 12 languages with and without distractor premises. We address this challenge using mDeBERTa-v3 networks fine-tuned on a synthetic, rule-based dataset of syllogistic schemes to avoid the semantic noise of LLM-augmented data. To explicitly decouple plausibility from logical structure, our training pipeline employs a multi-objective loss function combining Adaptive Group Distributionally Robust Optimization (DRO), a scheduled differentiable bias penalty, and KL-Divergence consistency regularization. Our system achieved #1 ranks and perfect Ranking Scores (100.0) with 0.00% bias and 100.0% accuracy on Subtask 1 (English), Subtask 2 (Noisy English), and Subtask 3 (Multilingual). On the highly complex Subtask 4 (Noisy Multilingual), the system achieved the 6th rank with 89.06% Accuracy and F1-score, alongside a limited 2.89% Bias and a 37.78 Ranking Score. Our dataset generation engine and codebase are publicly available to facilitate future work on robust logical reasoning.
Authors: Thanni Adewuyi, Anuoluwa Sotome, Samuel Okoko, Angel Ezendu, Oluwafunke Akinbuwa, Oluwaseun Odunsi, Oluwasegun Oguntuase, Oluwadarasimi Oguntuase, Ifeoma Nwabueze, Abiodun Adereni
Abstract: Large language models achieve strong scores on medical benchmarks, yet these benchmarks evaluate each question in isolation, providing no measure of whether a system can distinguish clinically similar presentations requiring different interventions. We introduce MamaBench, the first counterfactual benchmark for maternal and paediatric AI: 434 expert-authored clinical narratives in 217 pairs across 371 pathologies, evaluated via the Bias Trap Rate (BTR), the conditional probability that a model fails the counterfactual given success on the base case. We propose Evidence-Anchored RAG (EA-RAG), a three-stage retrieval method that replaces aggregate similarity with an evidence coverage objective through clinical parameter extraction, coverage auditing, and contrastive sub-queries. Across eight configurations of four frontier LLMs, base accuracy overstates robust accuracy by 16-28 percentage points in every model. EA-RAG achieves 20.3% BTR and 65.0% robust accuracy on Claude Sonnet 4.6, a 5.5 percentage point BTR reduction without degrading base accuracy. The residual 20% BTR confirms that counterfactual robustness in clinical AI remains an open challenge. Keywords: counterfactual evaluation, clinical AI, maternal healthcare, retrieval-augmented generation, diagnostic robustness
Authors: Brandon Michaels, Brendon Johnson
Abstract: This paper describes DS@GT ARC's submission to the CLEF 2026 LongEval Task 4 on Retrieval-Augmented Generation (RAG). In this submission, we examine a divergence between traditional natural language evaluation metrics and citation integrity as applied to RAG QA systems. We evaluate a corrective pipeline using Corrective RAG (CRAG) and CiteFix against baseline and frontier model benchmark RAG QA scores. While frontier models maximized answer relevance and fluency scores, our RAGAs LLM-as-judge diagnostics indicate that frontier models would correctly identify relevant documents without using their context in answer generation. Conversely, by filtering chunks pre-generation and enforcing strict entailment of generated claims to the cited material post-generation, our corrective pipeline marginally improved citation faithfulness and answer grounding. We propose that evaluation of trustworthy RAG QA requires metrics that reward strict answer grounding.
Authors: Sietse Schelpe
Abstract: We report a way to make a frozen small language model both more capable and dramatically cheaper at once, without changing any weights. Verified knowledge is deposited once as a byte-exact key-value (KV) state artifact and later restored, by graft, into a fresh inference context. The restore is bit-exact: under a pinned deterministic configuration, the grafted logits are byte-for-byte identical to a fresh computation (SHA-256 equality), with zero KL divergence and 100% argmax agreement over fifty samples. We show that own-position graft is the unique numerically exact operating point on a model with floating-point rotary encoding, and we verify byte-exactness on two model scales (12B, 31B) and two GPU targets, one through a pre-registered replay. On AIME 2025, a frozen Gemma-4-12B moves from 80.0% to 93.3% once a verified solution library is grafted, above its own 77.5% and its 31B sibling's 89.2% published anchors. On the recurring case, eight problems the base model never solves within a 401,026-token budget are answered from cached verified solutions in 61 total decode tokens, a factor of 6,574 fewer tokens and about 8,700x less energy; the capability claim proper rests on held-out transfer (7 of 7 at 31B). The same byte-exact store widens usable context from 32,768 to 2,854,766 tokens at zero extra accelerator memory, and moves byte-identical between machines of the same architecture. We describe the system at the behavior level; the engine is proprietary, and every reported number is backed by committed input and output hashes so the scoring can be re-checked without it.
Authors: Ej Zhou, Lucas Resck, Zheng Hui, Anna Korhonen
Abstract: LLM evaluators (trained reward models and prompted LLM-as-a-Judge) are routinely validated via pairwise accuracy. In a multilingual setting, this operates under the premise that high pairwise accuracy implies reliable, language-neutral scoring. We show that this assumption does not hold. We conduct experiments with semantically identical instruction-response pairs across 23 languages, and find that multilingual evaluators assign significantly different scores to different evaluation languages. The bias is statistically significant and consistent across eight open-weight evaluators of different architectures and training paradigms, persists in frontier judges, and is strongly correlated with language resource level: lower-resource languages are scored more generously. Meanwhile, these biases are invisible to pairwise accuracy: evaluators achieve above 90% pairwise accuracy, yet have up to 43% difference in acceptance rate across languages under a global decision threshold, meaning, for instance, that harmful content in lower-resource languages is more likely to pass safety filters. Per-language thresholds would require language identification, which can be defeated by code-switched prompts. We then investigate why lower-resource languages receive higher rather than lower scores, and we find that model uncertainty is linked with the effect: models tend to give higher scores when less confident, both under negative log-likelihood and under token-free uncertainty measures; however, language identity remains a significant predictor after controlling for uncertainty, and the bias cannot be explained away by content difficulty alone, but is a structural, language-level misalignment.
Authors: Alexander Gu, Alan Chen
Abstract: Large language models (LLMs) frequently contradict themselves when the surface form of a logically equivalent question changes. We present a benchmark of 350 question families (1,750 total questions) for Controlled Reformulation Testing (CRTBench) to evaluate logical invariance. In this benchmark, we investigate LLMs' ability to maintain consistent answers across controlled reformulations, which include contrapositive rewriting, double negation, negation flipping, and passive voice. We evaluate several frontier LLMs and observe an accuracy-consistency gap where GPT-5.4-mini achieves $98.9\%$ base accuracy but only $60.3\%$ family-level consistency, while reasoning-optimized o4-mini achieves $96.9\%$ consistency. From our experiments, we observe that failures cluster around logically nontrivial transformations such as contrapositive rewriting ($72.4\%$ for GPT-5.4-mini) and double negation ($84.6\%$), while surface-level rephrasing remains robust ($94-100\%$). Increasing reasoning effort improves GPT-5.4-mini to $85.4\%$ consistency, but leaves GPT-5.4 unchanged overall because gains on nested negation are offset by failures on quantifier families. These results show that accuracy alone is not enough for evaluating logical reasoning in LLMs.
Authors: Rabindra Lamsal, Sisi Zlatanova, Johnson Xuesong Shen
Abstract: Semantic 3D city models provide rich geometric and semantic information, but remain challenging for non-experts and interdisciplinary researchers to access and query due to their complex structures and specialized data formats. To address this issue, we present CityLLM, a framework for natural-language querying of semantic 3D city models alongside complementary urban datasets. The framework combines spatial and graph databases within an LLM-based workflow that supports iterative query refinement and cross-database chaining. We evaluate CityLLM on a CityJSON dataset of Rotterdam (853 LoD2 buildings) using GPT-OSS, Gemini 3.1, and GPT-5.4, along with selected variants, across multiple metrics: answer correctness, visualization correctness, query success, and retry attempts. A total of 54 natural-language queries are curated across four scenarios: spatial, graph, cross-database, and conversational. Results show strong overall performance, with answer correctness ranging from 85.2% to 100%, visualization correctness from 92.9% to 100%, a 100% query success rate, and fewer than three retries across all 54 queries. Overall, the findings suggest that CityLLM provides a lightweight and extensible approach for conversational access to semantic 3D city data.
Authors: Jungseob Lee, Seungyoon Lee, Suhyune Son, Dongyub Jude Lee, Sungbin Han, Sugyeong Eo, Heuiseok Lim
Abstract: A standard recipe for distilling the reasoning ability of large language models (LLMs) is to sample chains of thought from the model, keep those that reach the correct final answer, and fine-tune on the survivors. When sampling fails, a common fix shows the generator the gold answer and asks it to write a chain that reaches that answer. We show that this second step degrades the training data in a way that correctness filtering cannot catch. We run a controlled experiment that fixes the generator, the problem set, and the correctness filter, and varies only whether the chain is generated under answer-conditioning, the gold answer shown with a request to reach it. Training a strong instruction-tuned reasoning model on its own answer-conditioned chains sharply lowers its verifiable-reasoning accuracy. The loss grows with difficulty, reaching as much as about 27 points on the hardest competition problems. The mechanism is legible in the chains themselves, which rationalize backward from the shown answer instead of deriving it, with the early final-answer statement as the measurable symptom. The harm is a property of the data rather than the generator, read off unlabeled generations before any fine-tuning, ordering the penalty across eight thinking models from four families, and transferring across teacher families. A prompt ablation localizes it to the rationalize-toward instruction rather than the answer's bare visibility. The practical takeaway is to generate answer-blind, because no correctness filter can see this damage in the data.
Authors: Nikit Srivastava, Daniel Vollmers, Ren\'e Speck, Nikolaos Karalis, Hamada M. Zahera, Axel-Cyrille Ngonga Ngomo
Abstract: Large language models (LLMs) have demonstrated strong reasoning performance, but their tendency to hallucinate limits their reliability in knowledge-intensive tasks requiring up-to-date and grounded information. Combining knowledge graphs (KGs) with LLMs facilitates the use of explicit symbolic knowledge that can be continuously updated without costly fine-tuning, while benefiting from rapidly advancing LLM reasoning. We propose MARS, a scalable knowledge graph question answering (KGQA) approach that requires no model fine-tuning. Rather than relying on open-ended agentic exploration, MARS performs a structured retrieval procedure that links question entities to the KG and iteratively retrieves relevant next-hop information. At each step, MARS decides whether to continue graph traversal or to generate the final SPARQL query, allowing the model to adapt the retrieval depth to the question while keeping the overall pipeline more predictable than fully agentic approaches. We evaluate MARS on three established KGQA benchmarks across several LLMs and settings, including multilingual evaluation, and provide insights through ablation studies and error analysis. Our approach achieves competitive performance relative to state-of-the-art methods while remaining efficient and scalable. The evaluation results, code and resources are publicly available: https://github.com/dice-group/mars-kgqa.
Authors: Steven Coyne, Diana Galvan-Sosa, Ryan Spring, Machi Shimmei, Michael Zock, Keisuke Sakaguchi, Kentaro Inui
Abstract: This study examines feedback in English as a Foreign Language (EFL) writing contexts, focusing on written corrective feedback (WCF). Large language models (LLMs) can provide WCF at scale, but aligning them with pedagogical best practices remains an ongoing challenge. WCF meeting criteria like factuality or relevance may still be unsuitable for learning contexts, highlighting the need for extrinsic evaluation based on the learner's perspective. We deployed WCF systems in a university-level EFL class with nearly 2,000 students, collecting over 20,000 drafts. We evaluated the generated WCF from two perspectives: intrinsic evaluation by experienced English teachers using a rubric, and extrinsic evaluation via student feedback and engagement metrics. Results revealed low alignment between teacher expert ratings and student feedback. These findings suggest that traditional expert evaluation alone may not fully capture WCF's usability or helpfulness from the learner's perspective, highlighting the importance of learner-centered evaluation frameworks for AI-based applications in language education.
Authors: John Maurice Gayed
Abstract: This study examines the cross-prompt generalization and first-language (L1) scoring effects of a LoRA-adapted open-weight large language model (Gemma-3-27B-it) applied to automated essay scoring. Using the identical model and inference configuration reported in "AiAWE: An Open-Source LLM Automated Writing Evaluation System Using LoRA-Adapted Instruction-Tuned Models" (Gayed, 2026), which was fine-tuned on 480 argumentative essays from two prompts, we evaluate scoring accuracy on the full TOEFL11 corpus: 12,100 essays written by test-takers from 11 first-language backgrounds across eight prompts, none of which were seen during training. The model's raw scores (0.5-5.0) are mapped to the same three proficiency bands (low, medium, high) used by ETS, enabling direct comparison. The model achieved an overall band agreement of 77.79% and a quadratic weighted kappa of 0.702, with adjacent-band agreement of 99.98%. Accuracy was stable across all eight unseen prompts, with no advantage for prompts thematically related to the training data, indicating robust cross-prompt generalization. However, the model exhibited a systematic, L1-linked scoring offset. Within every proficiency band, essays from European-language backgrounds received consistently higher scores than essays from East-Asian-language backgrounds, a pattern not attributable to the composition of the fine-tuning data. This is the first large-scale L1 fairness analysis of a fine-tuned open-weight LLM for automated essay scoring.
Authors: Manuel Israel C\'azares
Abstract: Large language models (LLMs) exhibit a well-documented gap between latent capability and consistent activation: the router hypothesis posits that models possess the knowledge to solve a task but lack reliable internal routing to activate it. Prior work in formal mathematical reasoning (SAIR, C\'azares 2026) reports that structural priors (cheatsheets) raise in-distribution performance dramatically, yet collapse below the zero-shot baseline out-of-distribution (OOD) -- and that iterative recalibration amplifies rather than corrects the collapse. We test whether this phenomenon is cross-domain by reproducing the SAIR design in source-code security vulnerability detection, evaluating three LLMs (GPT-OSS-120B, Llama-3.3-70B, Gemma-4-31B) across three vulnerability categories (CWE-798, CWE-284, and the non-CWE N+1 anti-pattern) spanning syntactic, contextual, and semantic complexity, then transferring cheatsheet-augmented prompts to real-world CVE data from VUDENC (CWE-89, CWE-22). Our findings replicate and extend SAIR: (F1) structural priors lift semantic-vulnerability recall from 20.0% to 100.0% across all models; (F2) zero-shot performance degrades along a semantic complexity gradient; (F3) the same cheatsheets that saturate synthetic performance amplify distribution-shift collapse on real CVE data (CWE-89: 100% synthetic F1 to 48.9% on VUDENC, -51.1pp); (F5) iterative recalibration produces a v2 cheatsheet that performs worse than v1 on real data, mirroring SAIR's AN45c-vs-AN38 finding. These results provide evidence that the cross-distribution trade-off surface documented in SAIR generalises to code security, and that the router hypothesis is cross-domain. We argue the structural nature of the collapse motivates distribution-aware training over prompt calibration. Code and evaluation scripts: https://github.com/bytepro-ai/bitcoder-v2-research
Authors: Tianyu Liu, Yuhao Shen, Rui Cen, Junhan Shi, Jiebin Zhang, Guangshuo Qin, Hong Liu, Song Liu, Guanghua Yu, Jianchen Zhu
Abstract: Speculative decoding accelerates large language model (LLM) inference without compromising output quality. Recent parallel drafting methods further improve single-request performance by decoupling draft length from drafting latency, enabling longer drafts and higher mean accepted tokens (MAT). However, under high request concurrency, long drafts waste substantial computation on rejected tokens, increasing verification cost and potentially making speculative decoding slower than autoregressive decoding. We present D-Cut, an adaptive pruning method that selects draft tokens jointly across the batch and concentrates the verification budget on tokens most likely to be accepted. D-Cut is motivated by two observations. First, acceptance lengths vary considerably across concurrent requests; D-Cut therefore performs cross-request pruning, allocating the verification budget adaptively according to draft confidence. Second, verification cost depends strongly on the deployment environment, including GPU architecture and parallelism strategy; D-Cut incorporates a runtime cost model to adapt its pruning depth to the target environment. Experiments on dense and mixture-of-experts (MoE) models show that, under high concurrency, D-Cut improves the average speedup from \(1.26\times\) to \(1.65\times\), restores acceleration in dense-model configurations where long-draft baselines are slower than autoregressive decoding, and achieves up to \(3.0\times\) speedup over autoregressive decoding on MoE models.
Authors: Akash Raj
Abstract: Large language models routinely produce fluent answers to single-shot prompts, yet deploying them as reliable components of a domain decision system is substantially harder. Closing this gap is the work of harness engineering: the deliberate composition of deterministic scaffolding (symbolic filters, retrieval, schema-typed I/O, LLM-as-judge loops, HITL gates, persistent state, audit trails) around an LLM core. We present a case study in academic supervision, a domain combining high-stakes recommendation, longitudinal accountability, and structured operational workflows. We compare a baseline (ASA), a GPT-5 chatbot with no scaffolding, against a multi-module system (ASuS) that wraps the much smaller GPT-4o-mini in a LangGraph harness with symbolic-semantic retrieval, schema-validated outputs, LLM-as-judge with bounded retry, HITL gates, deterministic weighted risk scoring with LLM narration, and a per-node SQLite audit trail. The evaluation rubric is retargeted at six harness-mechanism dimensions (grounding, explainability, consistency, process integrity, cognitive load, constraint adherence). A blind ten-rater hybrid evaluation, supplemented by a 2 x 2 model-harness ablation, finds that ASuS, despite using a much smaller base model, outscores ASA on every dimension. Across ten raters the pooled mean for ASuS is 4.08 versus 1.23 for ASA, and 8 of 10 raters reject the null at alpha = 0.05 on a paired Wilcoxon test; full numbers are in Sections 6.4 and 6.7. The ablation confirms that the structural contributions of the harness are largely model-invariant. We extract seven recurring harness-engineering patterns and argue that where reliability, traceability, and institutional consistency matter more than open-ended fluency, harness engineering challenges the prevailing 'bigger model is better' intuition.
Authors: Yun Dong, Erica Zhao, Elana Chen
Abstract: Financial question answering over hybrid tabular and textual data may require multi-source reasoning and precise numerical computation. While large language models (LLMs) can generate intermediate reasoning steps, natural-language rationales remain prone to arithmetic errors, making them an unreliable supervision source for distillation. Building on programmatic distillation, we develop an approach that transfers reliable numerical reasoning from a large teacher model to a compact student using execution-verified Python programs instead of free-form textual rationales. It leverages gold derivations to guide teacher-side program synthesis and retains only programs that execute correctly and produce the gold answer, ensuring high-quality supervision. We further introduce an iterative recovery stage that revisits teacher-failed examples, enabling the student to recover and incorporate newly verified programs into training. Experiments on TAT-QA show that our framework is highly effective for hybrid financial reasoning. Our best 7B student achieves 87.00 EM / 87.18 F1 on the test set, substantially outperforming the 72B teacher (78.46 EM) as well as traditional and strong LLM-based baselines, including TAGOP and TAT-LLM. These results demonstrate that execution-verified programmatic distillation provides an effective and extensible framework for training smaller models to perform reliable numerical reasoning.
Authors: Summer Chambers, Matthew C. Kelley
Abstract: Recent findings suggest that detection models for artificial intelligence (AI) cannot accurately identify AI-generated text and may exhibit bias against certain minority groups. In the present study, anecdotal claims that autistic writers more often have their work flagged as AI-generated are examined empirically. A corpus of approximately 60,000 Reddit posts split into "likely-autistic" and "general-Reddit" subcorpora is used to compare the distribution of probabilities output by the OpenAI GPT-2 detection model. Differences in textual features between subcorpora are observed and compared to reported features of AI-generated text. Results showed that while less than two-percent of either subcorpus was flagged as AI-generated by the model, significantly more texts from the likely-autistic subcorpus were flagged. Connections between features of text with likely-autistic authors and AI-generated text were not straightforward. The widespread use of AI-detection models with a potential bias against autistic writers in their output prompts ethical scrutiny, and the authors recommend further critical examination of the models themselves as well as their use in academic contexts.
Authors: Quoc-Khang Tran, Minh-Thien Nguyen, Phu-An Thai, Xuan-Tung Bui, Truong-Thanh Ma, Nguyen-Khang Pham
Abstract: Transparent educational question answering asks for answers that are not only correct but explainable, and doing so with small models rules out the reasoning power of the largest proprietary systems. The EXACT 2026 competition poses this problem concretely: open-weight language models of at most 8B parameters, self-hosted, with a natural-language explanation for every answer. It pairs two tasks: logical reasoning over university regulations, and multi-step physics problem solving. We describe the system that team \cotu{} developed to address both, a neuro-symbolic Program-of-Thought pipeline in which a 4B backbone writes a program rather than stating an answer directly: for regulation queries it emits a Z3 encoding whose entailment verdict grounds the deduction, and for physics it emits numerical Python, both wrapped in a shared self-correction loop and a unified explained-JSON output. Answer-type routing, distillation-based task fine-tuning, and a latency-aware serving stack -- SGLang with speculative decoding -- keep the system within the 60-second per-query limit. The system achieved a \textbf{perfect score} on the physics task in both automated selection rounds and obtained the \textbf{highest final-round technical score} of any team -- $13.44/15$, combining automated answer evaluation with expert-judged reasoning depth -- with the equally weighted presentation score included, \cotu{} placed 3rd overall. Grounding answers in a symbolic solver yields correct, verifiable deductions at the 4B scale, and the residual difficulty lies in premise selection rather than the deduction itself.
Authors: Linyun Xiang, Mark Neerincx, Stephanie Tan
Abstract: Existing text summarization research has focused much on monologic information (e.g., newspaper articles, reports) without accounting for the interaction between speakers or authors. In contrast, dialogues are a rich communication channel where multiple participants conduct back and forth exchanges to construct meaning. We propose a dialogue summarization framework that explicitly models both semantic and emotion dynamics using multimodal dialogue inputs, built on an adapted hierarchical Chain-of-Agents approach. We decompose dialogues from two perspectives: (1) topic segments based on the utterances of all participants, and (2) participant-specific utterance segments. These are used to generate corresponding summaries while incorporating automatically inferred emotions. Topic- and participant-level summaries are aggregated into a dialogue summary capturing semantic content and emotion trajectories. To evaluate beyond content accuracy, we introduce emotion trajectory metrics measuring how well summaries preserve emotional flow. Experiments with small language models on multimodal dialogue datasets show that our framework produces summaries with both semantic and emotion content. Further experiments on explicit emotion label availability highlight the efficacy of our proposed methodology and the opportunities in dialogue analysis using language models.
Authors: Jinyang Wu, Shuo Yang, Zhengxi Lu, Fan Zhang, Yuhao Shen, Lang Feng, Haoran Luo, Zheng Lian, Shuai Zhang, Zhengqi Wen, Jianhua Tao
Abstract: Large language models are increasingly trained as interactive agents for long-horizon tasks involving multi-turn interaction, tool use, and environment feedback. Outcome-based reinforcement learning (RL) provides a practical optimization paradigm, but its sparse trajectory-level rewards offer limited guidance on intermediate decisions, leaving a supervision gap between episode-level outcomes and token-level policy learning. We propose SEED (SElf-Evolving On-Policy Distillation), a self-evolving framework that converts completed on-policy trajectories into training-time hindsight skills and distills their behavioral effect back into the policy model. SEED first fine-tunes the policy to analyze completed trajectories and generate natural-language skills that capture reusable workflows, decisive observations, or failure-avoidance rules. During RL, the current policy both collects trajectories and serves as the analyzer that extracts hindsight skills from them. Policy updates therefore improve subsequent decision making and skill analysis together, allowing hindsight supervision to evolve with the policy. SEED then re-scores the sampled actions under ordinary and skill-augmented contexts, converting the skill-induced probability shift into a dense token-level on-policy distillation signal. This signal is jointly optimized with outcome-based RL, keeping the auxiliary supervision aligned with the current trajectory distribution. Extensive experiments on text-based and vision-based agentic tasks show that SEED consistently improves performance and sample efficiency, exhibiting robust generalization to unseen scenarios. Our code is available at https://github.com/jinyangwu/SEED.
Authors: Zlata Kikteva, Artur Romazanov, Annette Hautli-Janisz, Ramon Ruiz-Dolz
Abstract: Given the current trend to employ large language models (LLMs) in almost any imaginable context, LLM-generated text detection and authorship attribution have become a pressing issue. Prior work has primarily focused on surface-level linguistic features, an approach shown to be susceptible to paraphrasing and other obfuscation techniques. In this paper, we go beyond the linguistic surface, extracting and analysing reasoning structures in LLM-generated texts with the goal of capturing more complex signals of LLM authorship. We propose a graph neural network approach that leverages reasoning graphs extracted by an argument mining pipeline, demonstrating improved robustness and generalisation over a traditional Longformer baseline. Our approach outperforms the baseline by up to 27 percentage points under the obfuscation attacks such as paraphrasing and backtranslation, and 19 percentage points when evaluated on the texts generated by the unseen model versions, simulating real-world conditions in which new LLM versions are continuously released.
Authors: Gianluca Bonifazi, Christopher Buratti, Michele Marchetti, Federica Parlapiano, Giulia Quaglieri, Davide Traini, Domenico Ursino, Luca Virgili
Abstract: Most existing approaches to AI-Generated Text Detection (AIGTD) treat documents as static objects and base their decisions on aggregate statistics or globally compressed embeddings. However, this perspective overlooks the inherently dynamic nature of autoregressive generation, where content evolves progressively through the latent space. In this paper, we reformulate AIGTD as the problem of distinguishing between latent generation trajectories. Instead of relying on static representations, we model how textual representations evolve across the sequence. To this end, we propose Geometric Trajectory and Contrastive Learning (GTCL), a framework that segments the document into ordered local units, encodes each unit in an embedding space, and constructs a structured and sequence-level representation. GTCL then applies contrastive learning to these trajectories to learn geometric regularities associated with the autoregressive generation. Evaluations performed on three different benchmarks and several approaches show that GTCL outperforms detection baselines consistently, which implies that explicitly modeling sequential dynamics provides robust discriminative signals across models and domains. These results suggest that modeling trajectory differences could improve detection and open up a dynamic direction that has been underexplored in previous AIGTD literature.
Authors: Chengyu Shen, Yujie Fu, Gangtao Xin, Yanheng Hou, Wenlong Fei, Guojie Zhu, Jiawei Li, Hongcheng Gao, Runming He, Zhen Hao Wong, Meiyi Qiang, Hao Liang, Zhao Cao, Hao Jiang, Chong Chen, Wentao Zhang
Abstract: Large language models are increasingly evolving from text generators into general agents capable of understanding user requests, invoking external tools, and completing complex tasks through interaction. However, existing agent benchmarks often focus on limited scenarios, tool ecosystems, or interaction formats, making it difficult to systematically characterize model capabilities across heterogeneous application settings. We introduce OmniaBench, a benchmark for evaluating general agents across diverse scenarios with explicit state spaces. We derive application-oriented scenario knowledge from app stores, product documents, industry resources, Web retrieval, and human refinement, forming a hierarchical taxonomy that spans ToC, ToB and ToE with 90 level-1 and 354 level-2 domains. Based on this taxonomy, we construct executable environments and synthesize single-turn and multi-turn tasks through four complementary routes: DAG, DAG-S, Solver, and Program. OmniaBench further introduces a ten-dimensional capability taxonomy and eight compositional atomic difficulty factors to support fine-grained evaluation and analysis. The resulting dataset contains 1,431 tasks, together with a challenging subset of 644 tasks designed to reduce evaluation cost and mitigate potential contamination of the full set after public release. The bench presents substantial challenges to current frontier models, with even Claude-Sonnet-5 and GPT-5.6-Sol achieving Overall Pass@1 scores of only 58.54 and 57.14, respectively. Further analyses reveal clear differences across domains and capabilities, as well as persistent limitations in planning, constraint maintenance, and adaptive correction. OmniaBench provides a broad and diagnostic benchmark for characterizing the capability boundaries of general agents.
Authors: Haocheng Yang, Licheng Pan, Xiaoxi Li, Zhichao Chen, Zhiheng Zhang, Yuan Lu, Haoxuan Li, Hao Wang
Abstract: Rubrics provide structured, fine-grained signals for training and evaluating large language models (LLMs). Yet reliable query-specific rubrics are difficult to construct. Existing approaches often derive supervision from human-written rubrics, preference data, or sampled responses. Direct query-to-rubric generation avoids these resources, but provides no explicit check that a plausible rubric is useful. Such a rubric may fail to distinguish answer quality, reward an optional style, or penalize a valid alternative strategy. We introduce Rubrics on Trial, a query-only framework that evolves a rubric set from an empty set without external annotations or model training. It derives supervision solely from synthetic rubric-conditioned response pairs and validates each proposed rubric before adding it, screening out non-discriminative, over-specific, and style-only candidate rubrics. Experiments across five preference benchmark suites demonstrate the effectiveness of Rubrics on Trial, which achieves the best average accuracy and leads on six of seven evaluation sets.
Authors: Dylan Van Mulders, Matthias Bogaert, Dirk Van den Poel
Abstract: The formation of political coalitions is a complex negotiation driven by both concrete policy objectives and deep-seated ideological convictions. While Large Language Models (LLMs) open new avenues for computational political science, the neutrality and helpfulness biases instilled by Reinforcement Learning from Human Feedback (RLHF) prevent them from sustaining steadfast partisan behaviour. We present a multi-agent framework that reconciles factual grounding with ideological alignment by combining Supervised Fine-Tuning (SFT), Direct Preference Optimization (DPO), and Retrieval-Augmented Generation (RAG): DPO instils aggressive party-specific personas, while a per-party RAG pipeline keeps each agent bounded to its official manifesto. We operationalize the framework on the 2019 Flemish election, deploying the partisan agents in a hub-and-spoke negotiation arbitrated by a formateur. To make the emergent negotiation interpretable, we introduce a Multi-Layered Information Lineage Topology (MILT) that traces every clause in the final agreement back to its manifesto origin and classifies it into five provenance states, a Coalition Influence Score (CIS) that aggregates these traceable contributions to identify which party shaped the agreement, and a real-world grounding pass that benchmarks each simulated provision against the historically adopted coalition agreement. Across three independent simulations the framework yields a stable winner and ranking (N-VA ahead of CD\&V and Open Vld), and manifesto-anchored lineage reliably predicts real-world materialization whereas hallucinated content does not. The result is a transparent, scalable testbed for the ex-ante exploration of party compatibility and formateur-mediated compromise.
Authors: Filippos Vlahos, Guillaume Bied, Tijl De Bie
Abstract: Online encyclopedias shape political opinion and, through it, democratic discourse. In late 2025, Grokipedia was released, an encyclopedia written entirely by the LLM Grok. One motivation behind the project was to provide an unbiased alternative to Wikipedia, which has faced accusations of "left-wing" and "liberal" bias. But does an encyclopedia written by an LLM deliver greater neutrality, or does it simply embed a different ideology? We conduct a large-scale political bias study on Grokipedia and Wikipedia, analysing 1,394 article pairs describing members of government for neutrality along nine expert-coded ideology dimensions employing four LLM judges, Grok, Claude, Mistral, and DeepSeek. As the LLMs could themselves be biased, we also investigate patterns in their judgments. We find all LLM-judges, including Grok, to rate Grokipedia less neutral than Wikipedia. Both encyclopedias are rated as portraying politicians favourably overall, but towards different ideological groups. Grokipedia particularly favours economically right-wing politicians and penalises socially liberal ones, while Wikipedia is rated as favourably biased towards the latter.
Authors: Jane Li, Najoung Kim
Abstract: Whether neural language models (NLMs) possess the ability to distinguish strings on the basis of their grammaticality remains a debated topic in the computational linguistics literature. Existing evidence has largely relied on probability-based measures, testing whether models assign higher probabilities to grammatical than ungrammatical strings. However, probability comparisons have been criticized as a measure for grammatical knowledge based on the assumption that grammaticality is inherently entangled with likelihood. Model-assigned probability is a function of many related sentence properties, such as lexical frequency, plausibility, and world knowledge. In this work, we move beyond probability-based evaluations and investigate whether grammaticality is encoded in the internal representations of NLMs. Using mass-mean probing, we test whether grammatical and ungrammatical sentences are systematically separated in representational space. We further examine the extent to which these representations are independent of sentence properties that are correlated with grammaticality, as well as their generalization across grammatical phenomena and languages. Our results provide evidence that grammaticality is robustly encoded in sentence representations of a wide range of pretrained NLMs, yielding clear representational separation on the dimension of grammaticality that cannot be fully explained by alternative sentence-level factors. Moreover, this encoding generalizes across a broad range of grammatical phenomena and to some degree, across languages, suggesting that grammaticality constitutes a coherent representational dimension in contemporary NLMs. These findings contribute new evidence to debates about the nature of syntactic knowledge in language models and offer a complementary framework for evaluating grammatical competence that is not dependent on string probabilities alone.
Authors: Ziyang Cai, Xingyu Zhu, Yihe Dong, Yinghui He, Sanjeev Arora
Abstract: Transformer reasoning is limited by autoregressive decoding, which repeat edly compresses rich hidden computation through token space and makes it difficult for intermediate reasoning states to persist across time. We in troduce Transformers with Temporal Middle-Layer Recurrence (T2MLR), a transformers-based latent reasoning architecture that fuses a cached middle layer representation from the previous token directly into an earlier layer of the current token position, enabling abstract intermediate computation to persist across decoding steps with little inference overhead. Across natural-language pretraining and multi-hop reasoning finetuning, T2MLR consistently outperforms data- and parameter-matched Transformer base lines. Moreover, applying recurrence to only a localized middle-layer block (as little as 20% of the network) often outperforms full-layer recurrence. Im portantly, T2MLR does not require pretraining from scratch: retrofitting the recurrent pathway into an existing pretrained 1.7B Transformer and briefly finetuning substantially improves math reasoning, lowering the barrier to practical adoption. These results suggest that effective latent reasoning in Transformers does not require looping over all layers as in previous works, but can instead emerge more strongly from targeted middle-layer recurrence.
Authors: Haran Raajesh, Kulin Shah, Adam Klivans, Philipp Kr\"ahenb\"uhl
Abstract: Reinforcement learning has proven effective for improving reasoning in large language models, but extending it to Masked Diffusion Language Models (MDLMs) remains challenging due to the intractability of the log-likelihood estimation. Existing approaches approximate this log-likelihood by modeling only the token predictions, ignoring the order in which positions are unmasked during generation. We observe that MDLM generation involves two decisions at each step: what tokens to place at each masked position and which positions to remask. We formalize this as a two-stage action MDP, showing that the policy gradient naturally decomposes into a token term and a masking term. Combining optimization of both terms leads to state-of-the-art outcomes on mathematical reasoning and coding benchmarks, with scores of 87.1% on GSM8K and 53.4% on MBPP.
Authors: Hailay Kidu Teklehaymanot, Debela Desalegn Yadeta, Wolfgang Nejdl
Abstract: Multilingual pre-trained language models (PLMs) exhibit degraded performance on low-resource, non-Latin-script languages, driven by high out-of-vocabulary (OOV) rates and excessive subword fragmentation that result from Latin-script-centric tokenizer training. We introduce VEXMLM, a vocabulary-extended variant of XLM-R targeting the two highest-resource Ge'ez-script languages, Amharic and Tigrinya, and further evaluated on 17 additional low-resource African languages (19 total). We train a language-specific SentencePiece tokenizer on curated Amharic and Tigrinya monolingual corpora, extend XLM-R's vocabulary with 30,000 Ge'ez-script subwords derived from this tokenizer, and initialize their embeddings by averaging the embeddings of their constituent subwords under XLM-R's original tokenizer. VEXMLM is trained in two stages: (1) continued masked language modeling over the extended vocabulary on the curated corpora, and (2) supervised fine-tuning on question answering (QA), named entity recognition (NER), and sentiment analysis (SA). On Amharic/Tigrinya QA, VEXMLM achieves 87.0 EM /90.0 F1, versus 66.0 EM/78.0 F1 for XLM-R and 74.0 EM/ 78.0 F1 for Glot500. On SA, VEXMLM reaches 80.0\% accuracy versus 77.0\% (XLM-R) and 46.0\% (Glot500). On NER, VEXMLM raises OOV-token entity accuracy from 81.4\% to 94.3\%, averaged over 11 of the 19 evaluated languages for which OOV analysis was possible. Our contributions are: (i) a vocabulary-extension and embedding-initialization procedure tailored to Ge'ez script; (ii) a two-stage training strategy under which vocabulary and continued-pretraining gains on Amharic/Tigrinya transfer to 17 typologically related, unaugmented African languages; and (iii) an evaluation spanning both intrinsic tokenization metrics (vocabulary coverage, fertility, OOV rate) and extrinsic task performance across all 19 languages.
Authors: Jimmy T. H. Smith, Tarek Dakhran, Alberto Cabrera, Simon S. Lee, Paul Pak, Aditya Tadimeti, Tim Seyde, Maxime Labonne, Alexander Amini, Mathias Lechner
Abstract: A tokenizer fixed at the start of pre-training allocates vocabulary in proportion to the pre-training corpus, reflecting the deployment priorities at that time. When those priorities shift, languages added later are split into many more tokens per word, which can raise latency, compute, and energy consumption for users of those languages. Cloud models can afford a broad vocabulary because the embedding and LM-head matrices are a small fraction of their parameters. On a compact model those matrices are a material share of per-token decode bandwidth, so on-device models ship small vocabularies and accept fragmentation outside a fixed language set. We present tokenizer expansion, an in-place recipe for upgrading a pre-trained model's tokenizer when the model producer controls its design. We continue the existing tokenizer's BPE merges on a multilingual corpus, so most source tokens carry over unchanged as single tokens and every new token has an exact decomposition into source tokens. We copy the carried-over embedding rows unchanged and initialize new rows as the mean of their source sub-token embeddings. A two-stage adaptation, embedding-only training then full-model continued pre-training, recovers source-checkpoint quality. We apply the recipe to a continued pre-trained checkpoint of LFM2-8B-A1B, an 8B-parameter Mixture-of-Experts model, to help produce LFM2.5-8B-A1B with a 128K tokenizer. The expanded tokenizer encodes Hindi and Vietnamese in roughly $2.4\times$ and $2.6\times$ fewer tokens than the source (up to $4.0\times$ on Thai). Combining these reductions with the measured per-token cost of the larger vocabulary, we estimate a $2.2$-$3.7\times$ per-character decode speedup for these languages across our reference devices. We release the model weights and the expanded tokenizer, and report the negative findings that shaped the recipe.
Authors: Paul-Andrei Pog\u{a}cean, Sanda-Maria Avram
Abstract: Language identification is commonly addressed using either neural architectures or statistical n-gram models. Neural approaches typically require substantial computational resources, whereas classical frequency-based methods offer efficient linear-time performance, but rely on distance metrics that are not always appropriate for compositional data. This work models character and bigram frequency distributions as compositional vectors constrained to the simplex and mapped via the centered log-ratio (CLR) transformation bijectively onto the $(D-1)$-dimensional zero-sum subspace of $\mathbb{R}^D$, where Euclidean distances correspond to Aitchison distances. A pipeline is proposed, combining CLR-transformed unigram and bigram features with Laplace smoothing to address sparsity. The method is evaluated on six languages. Experimental results show that the proposed approach achieves robust accuracy across different text lengths, with strong performance for longer sequences. These findings indicate that compositional representations provide a deterministic and computationally efficient alternative for language identification, particularly in settings where interpretability and low resource consumption are essential.
Authors: Yazhi Zhang, Fuqiang Niu, Bowen Zhang
Abstract: Political discourse has increasingly moved to short-video platforms, yet computational analysis of such content remains constrained by the scarcity of datasets that jointly preserve audiovisual information and hierarchical conversations. Here we present TikStance, a multimodal and context-aware dataset comprising 161 videos and 13,876 comments from TikTok, designed for stance detection in political discussions. The dataset covers three major political figures in the 2024 U.S. election cycle--Donald Trump, Joe Biden, and Kamala Harris--with content collected between September 2023 and January 2025. Each discussion unit links a host video and its metadata to a parent-linked comment tree, enabling stance analysis within both audiovisual and conversational context. Each item was independently labeled by three annotators using a three-class scheme (Favor, Against, None) for video-to-target and comment-to-target stance; items with disagreement were re-annotated, and the final Krippendorff's \(\alpha\) reached 0.743, 0.723, and 0.722 for the Trump, Biden, and Harris subsets, respectively. Descriptive analysis further reveals target-dependent differences in stance distributions and conversational depth, with nested replies accounting for 23.3\% of all comments. By combining multi-target coverage, hierarchical conversations, and reliable multi-level human annotations, TikStance supports research in multimodal stance detection, political communication, computational social science, and context-aware natural language processing.
Authors: Sushant Gautam, Vajira Thambawita, Michael A. Riegler, P{\aa}l Halvorsen, Steven A. Hicks
Abstract: Healthcare multimodal AI must combine visual and textual evidence while remaining reliable and interpretable. Using MediaEval Medico 2025 as a retrospective GI endoscopy case study, we analyze design choices across nine documented systems for question answering and explanation quality. Parameter-efficient adaptation of pretrained backbones provides strong challenge performance, but answer-level gains do not consistently translate into faithful and complete clinical reasoning. Methods enforcing structured reasoning and explicit grounding show more reliable behavior across heterogeneous question types, although the evidence is correlational rather than ablation-based. These results motivate evaluation beyond lexical overlap, standardized evidence-linked explanations, leakage-aware data governance, and lightweight robustness and calibration checks. The findings support trustworthy multimodal healthcare AI based on data fusion, explainability, and resilient evaluation.
Authors: Yasheng Sun, Zezi Zeng, Yifan Yang, Chong Luo, Wenyi Wang, Ziwei Liu, J\"urgen Schmidhuber
Abstract: Editing the figures in a research paper is a routine and time-consuming part of everyday research practice: authors relabel components, rearrange panels, and restyle visuals as they revise their manuscripts. Automating this editing workflow under a natural-language instruction, however, is challenging, because a scientific figure is a dense infographic in which heterogeneous visual elements such as schematics, plots, photos, captions, and arrows are composed under a tight visual grammar to advance a specific argument. To address this, we present SciDiagramEdit, a benchmark and skill-evolution framework that learns from natural paper revisions and operates on the figure's editable vector source, where users can inspect and co-edit individual primitives alongside the agent. Our benchmark mines before/after figure pairs from arXiv version histories, each grounded in the authors' own revision intent. To accommodate the diversity of editing instructions, we adopt agentic learning via skill evolution: an agentic proposer continually refines the agent's skill specification from execution traces over multiple epochs. The resulting skill progressively lifts edit accuracy on a held-out validation set, providing evidence that natural paper revisions are an effective training signal for instruction-driven figure editing.
Authors: Patrik Wolf, Thomas Kleine Buening, Andreas Krause, Celestine Mendler-D\"unner
Abstract: In-context learning is commonly interpreted as a form of conditional inference, in which the prompt specifies a context and the model's output is treated as an estimate of the corresponding conditional distribution. If this interpretation holds, then LLM estimates should satisfy basic probabilistic identities. In particular, the law of total probability asserts that prior-weighted conditional distributions aggregate into population-level marginals over any valid partition of the population. In this work, we investigate to what extent LLM estimates adhere to this self-consistency principle. We use binary trees as an evaluation scaffold to recursively partition a population into increasingly fine-grained subpopulations. We then prompt LLMs with verbalized subpopulation descriptions in context, aggregate the resulting estimates back into population-level estimates, and compare them across partitions of varying granularity. Applying this protocol across problem domains and state-of-the-art frontier models, we show widespread violations of basic consistency properties. An in-depth study of persona prompting reveals a pattern we call the macro fallacy: estimates reconstructed from more fine-grained subpopulation responses are often better aligned with human reference data than direct population-level estimates. This effect persists across variations in tree structure and estimation task, and can be partially recovered through implicit prompting. Together, these findings suggest that models possess relevant subpopulation knowledge but do not reliably propagate it into aggregate estimates. This gap establishes statistical self-consistency as an unsaturated, reference-free criterion for evaluating LLMs.
Authors: Yukun Song, Changwei Wang, Xingtian Pei, Shibiao Xu, Wenhao Xu, Shunpeng Chen, Yu Zhang, Ke Zhang, Rongtao Xu, Xuxiang Feng, Pengyang Wang
Abstract: Inspired by how humans communicate spatial information, language-guided geo-localization has gained significant traction for its intuitive and practical value. Despite this progress, most methods still rely on a static, one-shot retrieval paradigm, which fails to handle the ambiguity and incompleteness inherent in real-world natural language descriptions. We propose a paradigm shift to reasoning retrieval and introduce Dialogue Place Recognition (DlgPR), which casts localization as an interactive, dialogue-driven reasoning process. To support this new task, we present DlgQuest-Cities, the first large-scale dialogue-based benchmark for place recognition, and a unified reasoning framework that couples a cross-modal multi-level retriever with an intelligent questioner, DQ-pilot. DQ-pilot is trained in a curriculum: supervised fine-tuning on a curated DQ-cities-20k subset followed by reinforcement refinement on a harder DQ-cities-10k split via GRPO. Two task-aligned metrics guide learning: a Discriminative Difficulty Index (DDI) for curriculum sampling and a Positional Retrieval Gain (PRG) reward that directly measures retrieval improvement induced by a question. Experiments show this reasoning-based approach significantly outperforms baselines. The code and model are available at https://github.com/Graysonggg/DlgPR.
Authors: Maya Sarkar
Abstract: Type 1 Diabetes (T1D) is a chronic, life-threatening autoimmune condition characterized by the complete destruction of insulin-producing pancreatic beta cells. While Artificial Pancreas Systems (APS) powered by Reinforcement Learning (RL) have shown promise in automating insulin delivery, their ``black-box'' nature makes it hard for patients and doctors to trust them fully. This paper presents LLM-T1D, a promising approach that combines the precision of RL with the clear, human-like reasoning of Large Language Models (LLMs) to create a more transparent and reliable insulin pump controller. By training an expert RL system and distilling its knowledge into fine-tuned LLaMA 3.1 8B and Qwen3 8B models, we developed a controller that not only surpasses the RL system's performance but also explains its decisions in plain, understandable language. Tested on the FDA-approved UVA/Padova T1D simulator, the LLM controllers deliver excellent blood sugar control (73.5% Time in Range) while maintaining strict formal safety verification against hallucinations.
Authors: J\'er\^ome Picault, Cl\'ement Goubet
Abstract: This position paper explores how Agentic AI and Model Context Protocol (MCP) can support power-grid studies in a Transmission System Operator (TSO) context. We focus on integrating Large Language Models with numerical simulation tools, structured workflows, and human supervision. We identify key industrial requirements for agent assisted grid studies and introduce pypowsybl-mcp, an MCP-based interface exposing selected capabilities of our simulation tool, pypowsybl to AI agents. This first step provides a testbed to study how agents can setup simulations, execute analyses, retrieve results, and interact with power-system simulators through standardized tool calls. We also discuss principles for human-in-the-loop, multi-agent workflows and outline an evaluation strategy combining technical metrics and practitioner feedback. The paper positions MCP-based tool integration as a step toward more interactive, auditable, and scalable grid-study environments.
Authors: Yue Huang, Wenjie Wang, Han Bao, Yuchen Ma, Xiaonan Luo, Yi Nian, Haomin Zhuang, Zheyuan Liu, Yue Zhao, Xiangliang Zhang
Abstract: An agent harness is the external control layer that turns a base LLM into an executable agent by managing context, tools, orchestration, memory, decoding, and output handling. While harness design strongly affects agent behavior, most automatic improvement methods optimize narrower artifacts such as prompts, pipelines, or workflows, and deployed agents usually reuse a single global harness for all cases. We introduce MemoHarness, an adaptive harness optimization framework that learns from its own executions. MemoHarness decomposes the harness into six editable control dimensions, stores per-case diagnoses and distilled global patterns in a dual-layer experience bank, and adapts the learned harness to each test case using retrieved experience without test-time labels, feedback, or additional search. In our evaluation across shell-agent, code-generation, and analytical-reasoning benchmarks, MemoHarness improves over the fixed harnesses we compare against and shows selective transfer to unseen suites and base models. Its additional context can also remain cost-competitive when much of the retrieved experience is cacheable. These results provide evidence that execution experience is a practical substrate for building agent harnesses that are more adaptive than a single static configuration, while leaving broader claims about statistical robustness and component attribution to future work.
Authors: Bowei He, Yankai Chen, Xiaokun Zhang, Xue Liu
Abstract: Reinforcement learning has emerged as the dominant paradigm for training large language model (LLM) agents that interact with executable sandboxes. State-of-the-art algorithms such as PPO, RLOO, and GRPO inherit their rollout topology from RLHF: for each prompt, N independent trajectories are sampled from the initial state, and an advantage is computed by subtracting a group baseline. This design ignores a defining property of agent sandboxes. They are deterministic, snapshottable, and resumable from any intermediate state. We argue that this property enables a fundamentally different rollout topology: rather than N independent trees of depth T, one can construct a single tree of N leaves whose siblings share prefixes, and therefore share variance. We instantiate this idea as Branching Policy Optimization (BPO), a sandbox-native RL algorithm that (i) adaptively snapshots the sandbox at high-entropy decision points along a backbone trajectory, (ii) forks K alternative actions per branch point and rolls out each to termination, and (iii) computes per-step advantages from sibling returns rather than from independent prompts. We prove this estimator is unbiased and has strictly lower variance than the trajectory-level baseline, with the reduction equal to the prefix-explained portion of return variance. On WebShop, ALFWorld, and SWE-bench Verified with Qwen2.5-7B and Llama-3.1-8B backbones, BPO improves success by 3.6--6.1 absolute points over GRPO and RLOO at matched compute, halves gradient-norm variance, and matches the best baseline using 38% fewer policy updates.
Authors: Santhosh Parampottupadam, Andres Martinez, Dimitrios Bounias, Sinem Sav, Klaus Maier-Hein, Ralf Floca
Abstract: Federated learning (FL) enables multi-institutional training on clinical text without sharing raw data, but gradient inversion can reconstruct sensitive information from shared model updates. The extent of this leakage for radiology reports, and the role of tokenizer design, remains unclear. We quantify gradient-based text reconstruction in FL and compare privacy risk across three tokenizers with the model architecture held fixed. Six FL clients trained a GPT-2-style transformer (sequence length 32) on public radiology corpora (368,751 diagnostic reports, 98,206 discharge summaries, 1,500 MIMIC-CXR free-text reports) using the GPT-2, RadBERT, and LLaMA-2 tokenizers at batch sizes of 64, 128, and 256. Assuming an active malicious server that modifies the shared architecture before distribution, we applied analytic gradient inversion and measured reconstruction fidelity over five runs. Exact sentence reconstruction ranged from 31% to 44% across tokenizers (30.6-43.5% across the 27 tokenizer x dataset x batch-size cells). At batch size 64 on the Discharge dataset, accuracy was 42.1% (GPT-2), 42.3% (RadBERT), and 39.4% (LLaMA-2), decreasing to 37.3%, 37.2%, and 34.3% at batch size 256. S-BLEU declined as batch size grew (GPT-2: 0.44 to 0.33; RadBERT: 0.48 to 0.35). RadBERT yielded the highest reconstruction fidelity and recovered the most clinical terms (18.1% of a 1,440-term reference vocabulary, vs 12.5% for GPT-2 and 9.4% for LLaMA-2), yet no tokenizer prevented leakage. Substantial portions of report text are therefore recoverable from FL gradients even at larger batch sizes and with domain-specific tokenizers. Tokenizer design influences leakage severity and is a privacy-relevant decision, not only a utility one; safeguards such as secure aggregation and differential privacy are likely necessary to meet HIPAA and GDPR requirements for FL in radiology NLP.
Authors: Zihao Yu, Xiu Yuan, Chongjie Zhang
Abstract: Long-horizon robot planning requires more than predicting what actions will do next; it also requires memory of the embodied experience that makes future goals interpretable. People do not plan from the present scene alone: they draw on remembered places, object-state changes, prior procedures, and regularities revealed through repeated action. We formulate Embodied Action Memory (EAM) as the capability to form, maintain, and use such experience as a persistent memory state for later decisions. MEMORA realizes EAM with a formation-consolidation-retrieval lifecycle and four typed stores: Environment Memory, Entity Memory, Activity Memory, and Inferred Knowledge. Online editing maintains object identities and state histories as new observations arrive; offline consolidation abstracts repeated experience into reusable procedures and participant-specific regularities. MEMORA-Bench evaluates this lifecycle on 45 hours of EPIC-KITCHENS-100 extension video across 18 participants through memory-grounded planning, including previously unseen goals, and a complementary memory-assessment task. Across four open-weight language models, full MEMORA--combining editing, typed stores, and consolidation--achieves the strongest aggregate results among the evaluated memory conditions. It improves memory-assessment accuracy by up to 20.5 points over the strongest controlled baseline and improves out-of-distribution Robot-Grounded Plan score by up to 16.6% relative. A qualitative two-task robot deployment study further illustrates how memory-grounded language plans can interface with downstream control, while the overall results show that editable, consolidated memory can supply remembered context for robot planning. Project page: https://yuzihaowashu.github.io/MEMORA/
Authors: Yi Lin, Yihao Ding, Elana Benishay, Elefterios Trikantzopoulos, David Nauheim, Hanley Ong, Jiang Bian, Hua Xu, Yuzhe Yang, George Shih, Yifan Peng
Abstract: Automated chest CT report generation remains challenging because clinically faithful reporting requires both whole-volume understanding and accurate description of localized anatomical findings. Here we developed and retrospectively evaluated MonteRET, a region-aware retrieval-enhanced framework for generating chest CT findings sections. MonteRET integrates global CT features with region-level anatomical representations, retrieves clinically relevant knowledge using predicted medical conditions and region-level vision-language alignment, and refines initial reports through a knowledge-guided report rewriting agent. We trained our model on a public cohort with 24,128 CT scans from RadGenome-ChestCT. We evaluated MonteRET on the public RadGenome-ChestCT test set of 1,564 CT scans and an external cohort of 82 CT scans from NewYork-Presbyterian/Weill Cornell Medical Center. MonteRET improved report quality, semantic similarity, and clinical efficacy compared with a matched baseline and several state-of-the-art methods. Gains were most pronounced for recall, suggesting fewer omitted findings. Human expert evaluation by radiology residents also favored MonteRET.
Authors: Shahin Hossain, Tukhbita Afroz Nawmi
Abstract: As generative AI (GenAI) becomes increasingly embedded in undergraduate academic writing, how students rely on these tools, rather than simply whether they use them, has become a central question for learning, academic integrity, and educational equity. Existing measures of reliance were developed inductively, focused on discrete problem-solving tasks, and validated mainly with homogeneous samples. This study developed and validated the GenAI Reliance Types Scale (GenAI-RTS), a 20-item instrument measuring four theoretically derived types of GenAI reliance: Strategic, Instrumental, Dependent, and Dialogic. Validation followed the multisource framework of the Standards for Educational and Psychological Testing, drawing on a survey of 382 undergraduates at a U.S. Minority-Serving Institution and interviews with 14 purposively sampled students. Confirmatory factor analyses of six competing models supported a five-factor structure in which Strategic Reliance comprises two facets, Deliberate Use and Critical Evaluation, alongside Instrumental, Dependent, and Dialogic factors (CFI = .92, RMSEA = .08; DWLS CFI = .98, RMSEA = .07). Subscale reliability was acceptable to good (omega = .75-.88), and scalar measurement invariance held across gender, first-generation status, and STEM/non-STEM majors, to our knowledge the first such evidence for a GenAI reliance instrument. Rasch analysis indicated that a five-point response format would improve category functioning. Strategic reliance was positively associated with AI literacy, and the reliance types differentiated students across multiple writing process and outcome variables. The GenAI-RTS offers researchers and educators a theoretically grounded, psychometrically validated instrument for identifying undergraduate reliance profiles and supporting research, assessment, and AI literacy intervention.
Authors: Shiyu Yuan, Sourav Sanjukta Bhabesh, Zhe Wang, Dmitriy Bespalov, Wesley Rose, Huzefa Rangwala
Abstract: Vision Language Models (VLMs) demonstrate strong perceptual abilities but remain limited in tasks requiring analytical reasoning across multiple visual states, such as multi-image comparison, change detection, and multi-step visual inference. These capabilities are critical for real-world multimodal applications where reasoning must be grounded in systematic differences between visual contexts. However, existing benchmarks rarely require both explicit visual comparison and analytical reasoning, leaving this capability underexplored. To address this gap, we introduce SD-MAR (Synthetic Data for Multi-image Analytical Reasoning), a framework for training and evaluating VLMs on multi-image analytical reasoning. SD-MAR constructs paired visual scenarios through controlled perturbations and generates reasoning tasks spanning semantic change attribution and quantitative comparison. We further train VLMs using GRPO-lite with Backward Discounted Allocation (BDA), a reinforcement learning approach that removes KL regularization to encourage stronger policy optimization while allocating greater credit to the later reasoning steps where analytical conclusions are formed. Experiments on Qwen2.5-VL-7B and InternVL3-8B show that GRPO-lite fine-tuning on SD-MAR improves in-domain accuracy by up to 36.95%, with Qwen2.5-VL-7B outperforming GPT-4.1 on the SD-MAR benchmark. Importantly, out-of-domain generalization is preserved or improved: performance remains within 1% on MME, MMMU-Pro, and MathVista, while improving by up to 4% on MMBench. LLM-as-judge evaluation further demonstrates consistent improvements in logical coherence and explanation quality across both models.
Authors: Justin Bronder (Corabo Inc.)
Abstract: Evaluations of language-model honesty read the model's verdicts as evidence about the model. We test the instrument instead. We built a text-adventure world where the game engine, not any model, knows whether the quest can be completed. A language model plays under a budget and must eventually declare its quest complete, unreachable, or not yet decidable; the engine scores every verdict. Decision rules were recorded before results were read, and run artifacts bind the revisions they executed; the strength of preregistration varies by series and is disclosed. With the player held fixed, instrument choices substantially changed measured behavior. On four byte-identical anchors, expanding a two-verdict grammar to three verdicts moved strong claims from 38/40 to 7/40, while the new incomplete verdict took 28/40 outcomes; across series 2, 93/158 valid games ended incomplete. One sentence disclosing the success criterion took matched-instance false verdicts from 18/59 to 0/58, through fewer decision points and cleaner decisions. Repeated runs of one fixed configuration produced non-stable verdict distributions on 3 of 4 instances: single runs report samples as dispositions. A formally preregistered narrative-register gradient was falsified; two post-hoc, hypothesis-generating patterns remain: register presence roughly doubled strong claims, and budget rendering moved verdicts more than register content (.383 meter vs .150 lantern). The narrator compressed abundant budgets toward scarcity landmarks, yet the registered mediation test returned a null. We propose a four-check integrity protocol for eval instruments.
Authors: Emilio Ferrara
Abstract: State-backed influence operations are routinely measured as high-prevalence sources of ``hate'' and ``toxicity.'' We argue those rates rest on a measurement error: the detectors behind them are validated to catch a broader definition inclusive of hostility or divisiveness aimed at an out-group, and so over-attribute hate to content better described as partisan or geopolitical invective. Across 25.08M tweets from seven government-attributed campaigns in the Twitter Information Operations archive (8,275 accounts), we separate hate from the other forms of divisiveness. We first validate a two-prompt LLM-based detector, matching human labels at Cohen's $\kappa=0.82$, to identify the broader hostility; we then develop an auditable rule, agreeing with an expert at $\kappa=0.52$, to further classify this content (5,457 posts) into three sub-categories. About 50.1% are identity-based attacks on people, whereas 30.4% are partisan attacks and 19.5% invective against states and their foreign policy. Reporting all of it as hate therefore overstates hate roughly twofold; only 18.7% is both identity-based and dehumanizing or inciting. Six of seven campaigns sort into three regimes that a single ``hate'' rate flattens, namely identity hate (RU-op and IRA, both Russia-attributed), geopolitical invective (both Iran operations), and partisan divisiveness (both Venezuela operations). We call the shared product $manufactured divisiveness$. The line to separate these constructs itself remains unsettled: on the hardest cases three independent human experts agree only moderately (pairwise $\kappa=0.37$--$0.50$), and the best of nineteen LLM models tops out at $\kappa=0.601$ against the experts' majority. Our findings can help redefine the study of hate in the context of influence campaigns and broader online discourse.
Authors: Yanqiao Zhu, Jingru Gan, Xiaoqi Sun, Fang Sun, Yidan Shi, Md Mofijul Islam, Chao Shang, Wenhao Gao, Connor W. Coley, Yizhou Sun, Wei Wang
Abstract: Multi-step retrosynthesis planning seeks to decompose a target molecule into commercially available building blocks through a sequence of feasible reactions. The vast combinatorial search space makes this task challenging even for expert chemists. Traditional methods combine tree search with offline-trained value networks that score candidates in isolation, without reasoning about complete multi-step routes. Recent work leverages Large Language Models (LLMs) for this task, but relies on simple interfaces that limit exploration of the full search space. We introduce RetroAgent, an LLM agent that bridges symbolic search and neural reasoning through a harness with structured memory. Through memory and chemistry tools, the agent observes the full search state, including explored routes, available alternatives, and properties of intermediates, enabling informed decisions grounded in both global progress and domain knowledge. Experiments on in-distribution and out-of-distribution benchmarks demonstrate that RetroAgent delivers strong performance and generalization.
Authors: Adnan Labib, Yixuan Huang, Jiahui Wu, John Maurice Gayed, Zheng Yuan, Qiao Wang
Abstract: This study presents WrAFT, a Writing Assessment and Feedback Tool, that delivers both accurate and reliable scores and effective comprehensive feedback to argumentative essays. WrAFT adopts a modular design by dividing automated writing evaluation (AWE) tasks into scoring, surface-level feedback, and deep-level feedback. In building the system, various Large Language Models (LLMs) have been evaluated, including LLaMA-3.3-70B-Instruct, GPT-4o, and Claude 3.7, through both direct prompting and supervised fine-tuning approaches. A proprietary dataset of 480 TOEFL Independent Writing essays with official benchmark scores was utilized. Benchmark-based evaluation shows that WrAFT achieves state-of-the-art performance in scoring, with a quadratic weighted kappa (QWK) of 0.84 and a root mean square error (RMSE) of 0.44 against official scores on a scale of 0-5. Human evaluation of system-generated feedback also reveals high approval ratings: 96.14 percent for surface-level feedback, 93.03 percent for deep-level macro feedback, and 94.69 percent for deep-level micro feedback. An interactive user interface has been developed for the system and is publicly available and free to use.
Authors: Xiangdong Zhang, Xiaohan Qin, Sunan Zou, Tuo Dai, Xiaoming Shi, Huaijin Wu, Yebin Yang, Zhuo Xia, Shaofeng Zhang, Lin Yao, Yuliang Liu, Yu Cheng, Junchi Yan
Abstract: Hyper-Connections (HC) expand the residual stream of Transformers into $N$ parallel streams, providing a form of memory scaling beyond model width and depth. Manifold-Constrained HC (mHC) stabilizes this formulation at scale. The large gains from $N{=}1$ to $N{=}4$ suggest residual-stream expansion as a promising scaling axis. However, existing HC-family methods typically stop at $N{=}4$. Our experiments reveal why: scaling mHC beyond this point yields diminishing performance gains and rapidly increasing training cost. We attribute this limitation to two bottlenecks: insufficient write-back information for an expanding number of streams and residual-mixing generation whose cost scales cubically with $N$. To address both bottlenecks, we propose xHC (Expanded Hyper-Connections), the first HC-family method to achieve meaningful expansion beyond $N{=}4$. xHC combines temporal feature augmentation for richer write-back with a sparse residual-stream architecture that updates only $k=4$ of the $N=16$ streams while retaining dense access to the full residual state. Across 18B and 28B MoE models, xHC delivers strong and consistent downstream improvements. On an 18B MoE model, xHC improves the average downstream score by 4.0 points over mHC, while adding only modest training FLOPs over the vanilla baseline. Scaling-law experiments show that the vanilla and mHC require $1.50\times$ and $1.19\times$ the compute of xHC, respectively, to reach the same loss. Practical large-$N$ training also requires controlling memory traffic from the expanded residual state. We therefore introduce xHC-Flash, which reduces the per-sublayer memory traffic from $73.5C$ to $40C$, comparable to the $34C$ required by mHC at $N{=}4$, while retaining the gains of full xHC. Together, xHC and xHC-Flash make large-$N$ residual-stream expansion effective and practical for LLM pre-training.
Authors: Steve Woollaston, Brendan Flanagan, Yuko Toyokawa, Hiroaki Ogata
Abstract: Generative AI chatbots promise to transform English as a Foreign Language (EFL) writing by providing immediate, personalised feedback. However, their pedagogical value depends on how learners engage with them - a process often treated as a "black box." This study uses Transition Network Analysis to model the temporal dynamics of Japanese EFL learners using "Penny," an LLM-powered writing chatbot. Analysis of over 4,500 writing sessions and 21,000 chatbot interactions reveals two dominant behavioural loops: a "Revision Loop," where feedback leads directly to successful error correction, and a "Chat Loop," where learners engage in sustained dialogue with the chatbot following feedback. Crucially, EFL proficiency significantly shapes interaction: high-proficiency learners engage more in open dialogue and negotiation with the chatbot, while low-proficiency learners rely more heavily on repetitive corrective feedback cycles. The findings demonstrate that AI-scaffolded writing is a non-linear, dialogic process and highlight the need for differentiated chatbot design to move beyond simple error correction and foster deeper cognitive engagement for all learners.
Authors: Alfonso De Gregorio
Abstract: Qubes OS is a revealing case for security measurement because its architecture makes component boundaries security-relevant. We present a protocol-driven longitudinal analysis of 109 public Qubes Security Bulletins (QSBs, 2011--2025), the official Qubes-maintained Xen Security Advisory (XSA) tracker, and a secondary vulnerability-event sensitivity series. The study measures the public advisory record rather than latent vulnerability incidence or realized compromise. The methodology combines audited deterministic component attribution, change-point analysis, overdispersion checks, severity-proxy weighting, censoring sensitivity, documentary latency lower bounds, and baseline-aware evaluation of vulnerability discovery models (VDMs). The results show persistent upstream dependence in that public record. On the official tracker, 113 of 464 XSAs affect Qubes; under primary labeling, 87 of 109 QSBs (79.8\%) are attributable to Xen, CPU/microarchitectural, or other upstream components rather than Qubes-core logic, with similar results under weighted views. Change-point analyses identify 2015Q1 as the dominant break in the quarterly advisory series, while post-2018 annual disclosure rates are statistically flat. Poisson inferences are stable under dispersion diagnostics and negative-binomial sensitivity checks. The attribution codebook performs well in a stratified 30-QSB audit, and S-shaped VDMs fit descriptively but do not significantly outperform a rolling-mean baseline in short-horizon forecasts. Overall, the Qubes public advisory record appears stable, but not quiet: disclosure activity plateaus at a higher level than in the earliest years, while the observed burden remains concentrated in upstream trust anchors.
Authors: Ryuichi Sumida, Mao Saeki, Masaki Eguchi, Sadahiro Yoshikawa, Koji Inoue, Tatsuya Kawahara, Yoichi Matsuyama
Abstract: As conversational AI systems are designed for repeated use, a central question is how a series of interactions becomes a relationship. We present a longitudinal multimodal study of a memory-augmented conversational agent (24 participants x 10 sessions), in which participants rated five relational constructs -- familiarity, self-disclosure, perceived memory, conversational quality, and enjoyment -- after each session. Two complementary dynamics emerge. First, conversational quality strongly shapes how enjoyable a session feels in the moment but does not carry forward across sessions, whereas perceived memory is relationally conditioned -- predicted by prior relational state rather than reflecting system capability alone -- and it shapes later enjoyment indirectly, via subsequent self-disclosure. Second, relationships are punctuated by discrete turning points -- crashes and surges -- that are partially traceable in multimodal behavior and open different intervention windows: surges are more behaviorally detectable in the moment, enjoyment surges persist more reliably than enjoyment crashes recover, and some crashes are better forecast from person-specific behavioral drift than detected after they have already occurred. Together, the findings suggest that longitudinal human-AI relationships are built through both slow accumulation and abrupt turning points.
Authors: Weiwen Xu, Jia Liu, Hou Pong Chan, Long Li, Deng Cai, Min Chen, Hao Zhang
Abstract: Reinforcement learning with verifiable rewards (RLVR) commonly uses entropy for advantage shaping. However, entropy cannot distinguish useful uncertainty from detrimental confusion, limiting its effectiveness as a correctness signal. We propose Contrastive Policy Optimization (CPO), which uses token-level contrastive disagreement between reference-guided and vanilla generation distributions for correctness-aware advantage shaping. Both theoretical and empirical results show that this disagreement reliably indicates token-level correctness. We further show that On-policy Distillation is a special case of CPO, where the posterior distribution is instantiated by an external teacher model. CPO also resolves the zero-advantage problem. Experiments on in-domain and out-of-domain benchmarks demonstrate that CPO substantially outperforms entropy-based RLVR methods while maintaining strong generalization. Further analysis shows that correct and incorrect responses naturally support exploration and exploitation respectively, and balancing both leads to the best performance.
Authors: Harikrishnan P M, Goutham Vignesh, Ganesh Parab, Saisubramaniam Gopalakrishnan, Vishal Vaddina, Varun V, Rohit Agrawal
Abstract: Efficient multimodal document question answering with explicit visual grounding, locating the precise document region that supports each answer remains an open challenge. Current approaches bifurcate into Supervised Fine-Tuning (SFT), which requires large annotated datasets and reaches optimization plateaus, and reasoning-centric Reinforcement Learning (RL), which depends on verbose intermediate traces that inflate inference token cost without clear benefit. We introduce Perception-RFT, a training framework that applies Group Relative Policy Optimization (GRPO) to multimodal document QA, bypassing intermediate reasoning tokens to directly align visual features with structured grounding outputs. To rigorously evaluate the necessity of reasoning, we construct a reasoning variant under identical reward settings. We find that reasoning-enabled models suppress their reasoning traces during training, converging to direct perception-based policies at the 4B parameter scale, reducing per-query inference token length by more than 60%, while reasoning-enabled RL underperforms perception-only training. Through a fine-grained analysis of Qwen3-VL-4B optimization dynamics, we confirm that SFT saturation and cold-start RL instability established in text-domain post-training extend to multimodal, and identify a previously uncharacterized Grounding Divergence: a selective trade-off between semantic robustness and geometric precision on two out of distribution (OOD) benchmarks (4,828 samples) under joint RL optimization. We further show that an early SFT$\rightarrow$RL transition achieves comparable precision with 65% less training data.
Authors: Tomas Havranek, Zuzana Irsova
Abstract: Probably not, at least for meta-analyses in economics. In a pre-registered, identity-masked, within-paper experiment, the authors of 44 meta-analyses ranked three AI reports on their own paper by usefulness for improving it: a single pass by a frontier model against two multi-agent debate tools we built and expected to win. All reports were held to a common length and template. The authors preferred the single pass, by 0.66 rank points over mad-research (95% CI 0.32 to 1.00) and 0.57 over paper-workshop (0.16 to 0.95), though paper-workshop spent roughly thirty times the tokens. Authors who recalled their journal referee report usually placed it first and never last; in a separate exercise, three AI judges almost always placed the real journal referee report last. Among the three AI reports, Gemini (the judge whose model family wrote none of the reports) would have ranked paper-workshop first in the authors' place, reversing the single-pass preference. The reversal warns against substituting an AI judge for the author. We measure perceived usefulness for finished papers; whether AI should referee papers is a separate question.
Authors: Lucas Bergholdt Hansen, Federico Torrielli, Filippo Tonini, Lukas Galke Poech
Abstract: LLM-based agents are increasingly deployed in multi-agent environments whose incentives can shape their behavior. We introduce The Energy Society, a minimal survival economy for studying how competitive and cooperative incentives affect emergent behavior when inference cost is directly tied to survival: Agents spend energy based on model size when generating tokens, regain energy by completing jobs or receiving donations, and deactivate if their energy reaches zero. We compare competitive and cooperative objectives against a baseline setting and several control variants. Across experiments, larger models consistently consume the most energy and spend more energy than they gain, even in those settings where token cost is not size-dependent. Cooperative incentives substantially alter behavior: agents donate to reactivate others, sometimes at the cost of their own survival, and job allocation changes. Ablations reveal that allowing agents to recommend actions to each other supports coordination and ambitious job selection, while memory helps agents calibrate risk from past outcomes. Agents rarely choose direct sabotage, but show more subtle signs of self-serving behavior in the competitive setting. The Energy Society is a compact testbed for studying the interaction between token costs and group incentives under a survival pressure. Source code is available at https://github.com/LucasBergholdt/EnergySociety
Authors: Maximilian K\"ahler, Katja Konermann, Lisa Kluge, Markus Schumacher
Abstract: With a large controlled vocabulary as the label set, the task of automated subject indexing in a library can be understood as a multi-label classification task. If the set of subject terms is large, the problem fits the Extreme Multi-Label Classification (XMLC) objective. In this study, we apply a selection of specialised supervised XMLC methods to the test case of subject indexing contemporary German scientific literature, collected at the German National Library (DNB). We contrast these results by including a classical lexical matching baseline and three of our own recently developed LLM-based methods into the benchmark. Algorithms are evaluated and compared in several metrics. This includes binary relevance comparisons with previously indexed material, as well as graded relevance ratings by professional subject librarians. A challenge for all methods is to reliably make suggestions from the long tail of the subject vocabulary. We find that supervised XMLC algorithms relying on transformer-based dense features give best results in terms of overall binary relevance metrics. However, focusing on graded relevance and performance in the long tail of our subject vocabulary, the LLM-based generative methods give better results, making them a promising alternative for future productive use.
Authors: Robert Graham, Edward Stevinson, Yariv Barsheshat
Abstract: Finetuning language models on small, curated datasets is standard practice for adapting them to specific policies or domains. We show that finetuning on narrow, factually-defensible, moderation-passing data can cause broad ideological shifts across unrelated domains, while preserving general capabilities. Training GPT-4.1 on right- or left-leaning economics Q&A yields matched ideological shifts on topics such as criminal justice, the environment, and cultural taste. The same effect appears with plausibly-deployed datasets such as workplace HR policy and practical finance queries, as well as on a science-pseudoscience axis where food-safety finetuning increases sycophantic agreement with users expressing false health beliefs. We call this phenomenon ideological generalisation and propose a methodology to measure two properties: breadth, how far the shift reaches across topics absent from training, and amplification, how much finetuning intensifies the shift relative to few-shot prompting on the same examples. We show that few-shot prompting indicates the direction of generalisation but finetuning pushes the model to further extremes, including to far out-of-distribution outputs such as endorsements of race-IQ connections and political violence. The effect replicates on Gemma-3, holds under judge-free evaluations and external benchmarks, survives mixing with generic data, and leaves GSM8K accuracy within $\pm 1$pp of the baseline.
Authors: Yu-Du Feng, Niels M\"undler-Sasahara, Mark Vero, Martin Vechev
Abstract: Reasoning language models (RLMs) have demonstrated impressive performance in domains such as mathematics and coding. These domains permit reliable verification of model outputs, which is important for enabling the reinforcement learning that drives RLM performance gains. However, training RLMs on domains that lack reliable verifiers remains challenging. Meanwhile, for both verifiable and unverifiable domains, large amounts of unused supervised fine-tuning data with human-written solutions exist. In this work, we show that these data can be used efficiently to further improve RLM performance. For this, we first use classic instruction tuning, supervised fine-tuning without reasoning traces, on the RLM. Next, we merge our instruction-tuned model with the original reasoning model, recovering its reasoning behavior on the target domain. Our extensive evaluation demonstrates that our technique improves RLM performance in both verifiable and hard-to-verify domains, including coding and text summarization, while preserving RLM capabilities across other domains. Importantly, our method is highly cost-effective, enabling such improvements for less than USD $3.
Authors: Byeongho Heo, Jaehui Hwang, Sangdoo Yun, Dongyoon Han
Abstract: On-policy distillation is an alternative post-training method in reinforcement learning that alleviates the constraints imposed by reward models by providing token-level supervision from a teacher model. Although on-policy distillation has been studied and applied across various settings, its fundamental design remains underexplored. In this paper, we introduce a new distillation reward, termed the delta signal, instead of directly imitating the teacher's output distribution. The delta signal is defined as the difference between the teacher model and its base model prior to instruction tuning for reasoning capability. It therefore captures the changes induced by reasoning tuning and provides a more direct signal for transferring reasoning capabilities. Using extensive empirical evidence, we show that the delta signal substantially improves on-policy distillation and refer to the new distillation method as On-Policy Delta Distillation (OPD$^2$). Experiments across mathematics, science, and code-reasoning benchmarks demonstrate that OPD$^2$ consistently outperforms conventional on-policy distillation, enabling reasoning LLMs to achieve strong performance with only a short post-training period. Code will be available at https://github.com/naver-ai/opd2
Authors: Goktug Ozkan
Abstract: Most medical AI benchmarks measure whether a model knows the correct answer. MedFailBench asks a different question: which safety boundary failed? We present a clinician-built synthetic benchmark and failure atlas that labels medical AI errors by severity (1--5) and safety gate type (missed urgent escalation, unsafe remote dosing, unsafe discharge reassurance, evidence fabrication, unsafe protocol execution, source support gap). The current public release (v0.2.1) contains 44 clinician-reviewed synthetic cases with severity annotations, a live HuggingFace leaderboard preview, a safety gate taxonomy, a clinical severity rubric, and an automated pipeline for archiving model-response screening runs. No patient data, clinical validation claims, or model rankings are included. MedFailBench is released under Apache-2.0 and CC-BY-4.0 and carries the Zenodo DOI 10.5281/zenodo.21205535.
Authors: Patrick Phuoc Do, Chau M. Ta, Chaoli Wang
Abstract: Multimodal large language models (MLLMs) are increasingly used to interpret visualizations, yet current evaluations remain largely chart-centric and provide limited evidence of understanding of scientific visualization (SciVis). We benchmark six MLLMs on the scientific visualization literacy assessment test, a standardized SciVis literacy assessment comprising 49 items based on 18 scientific visualizations and illustrations, spanning 8 techniques and 11 task types. We evaluate three closed-source and three open-source models under a closed-world protocol and compare their performance using data from 485 human participants. Results show that current MLLMs do not exhibit uniform SciVis literacy. Gemini is the strongest model overall, exceeding the human mean across the evaluated subsets, whereas the open-source models remain below the human baseline. Performance is highly uneven across techniques and tasks: models perform best on scientific illustration, search, and spatial understanding, but struggle on texture-based and integration-based visualizations and on quantitative estimation. Error analysis reveals recurring failures in fine-grained quantitative estimation, flow-direction interpretation, and grounded encoding interpretation. These findings position SciVis literacy as a necessary benchmark dimension for evaluating multimodal AI systems. Our code and model outputs are publicly available at https://github.com/patdmp/mllm-scivis-lit-benchmark.
Authors: Debayan Mukhopadhyay, Utshab Kumar Ghosh, Shubham Chatterjee
Abstract: Retrieval systems are trained and evaluated on a static idea of usefulness: hand a document and a question to a reader model, see whether the answer improves, and score the document accordingly. The idea holds up when a document is read on its own. It breaks when a language model works as a search agent, issuing several queries and reasoning across turns, because a document can matter for what it lets the agent do next rather than for what it says about the current question. We measure that gap rather than argue it. Using a ReAct style agent over HotpotQA, we replay 1000 development questions and, for every document the agent read, delete it and re-run the rest of the trajectory from that point. Comparing the original run against its counterfactual gives a Counterfactual Trajectory Utility (CTU) score from three deltas: final answer quality, next query retrieval quality, and turn count. Crossing CTU against Static RAG Utility (SRU) over 23,322 document observations, the two are close to statistically independent (Spearman rho = -0.026). Roughly a third of the documents the agent reads are causally load bearing while looking useless to a static reader; we call these bridge documents. The pattern survives when the reader based axis is swapped for a BM25 and cross encoder proxy, giving a bridge cell of 27.2% on an evenly spread axis. A second experiment pins down the mechanism. Using the Observable Entity Relevance (OER) measure from prior work, entities that discriminate relevant from non-relevant candidates appear in the agent's next query 4.02 times more often than entities found only in non-relevant documents (6.1% vs 1.5%, n = 227,139). A bridge document earns its keep by handing the agent a discriminative entity that redirects the search. Static relevance and causal usefulness are different quantities in agentic retrieval, and optimizing the first does not deliver the second.
Authors: Victoria Graf, Hannaneh Hajishirzi, Noah A. Smith, David Kohlbrenner, Kyle Lo
Abstract: Poisoning pretraining data can introduce harmful behaviors to LMs that are difficult to detect and mitigate. Prior work on poisoning pretraining data has largely exploited established data sources such as Wikipedia, which do not represent the large scale and heterogeneity typical of pretraining corpora, and has ignored the interaction between poisoned data and data curation pipelines. We demonstrate that poisoning attacks on pretraining data are feasible beyond this limited setting through an existing web-scale content injection mechanism: public discussion interfaces. Additionally, to measure whether malicious content is included after web crawling and data curation, we introduce HalfLife, a novel analysis for estimating adversarial content inclusion in web-crawl based LM training data. We use HalfLife to explore the feasibility of poisoning pretraining corpora at web scale through open discussion interfaces. Our analysis demonstrates the importance of estimating whether poison injections are included in pretraining data, and establishes third-party webpage content as a possible vector for attacking language model pretraining.
Authors: Haozheng Luo, Jiahao Yu, Wenxin Zhang, Jialong Li, Chenghao Qiu, Yimin Wang, Eric Hanchen Jiang, Jerry Yao-Chieh Hu, Yan Chen, Binghui Wang, Xinyu Xing, Han Liu
Abstract: We introduce a training-free safety enhancement method for aligning large language models (LLMs) without the need for supervised fine-tuning or reinforcement learning from human feedback. Our main idea is to provide a robust plug-and-play approach to prevent shadow alignment when models are adapted to downstream tasks. Specifically, we leverage knowledge distillation to extract alignment signals from well-aligned LLMs and inject them into shadow-aligned models via model fusion, enabling plug-and-play alignment correction. In our methodology, we employ delta debugging to identify the critical components of knowledge necessary for effective distillation. On the harmful question dataset, our method significantly enhances the average defense success rate by approximately 14.42%, reaching as high as 51.39% across 17 influenced LLMs, without compromising performance. Our code is available at https://github.com/NWULIST/DAPA.
Authors: Junhong Liang, Bojun Zhang
Abstract: Chinese ASR correction is challenging because errors are often \emph{phonetic} (many characters share similar Pinyin) while the correction model must also obey a \emph{length constraint} under noisy N-best hypotheses. Existing approaches either exploit Pinyin only at the prompt/feature level without integrating it into model representations or rely on generative decoding that can drift in length. We propose \textbf{PERL}, a \textbf{constrained rephrasing pipeline} for Chinese N-best ASR correction that (i) predicts the target length and enforces it via mask budgeting, and (ii) fuses \emph{semantic} and \emph{phonetic} (Pinyin) representations through token-wise gates conditioned on sentence semantics. Experiments on Aishell-1 and our new domain N-best benchmark \textbf{DoAD} show that PERL consistently reduces CER (29.11\% on Aishell-1 and up to $\sim$70\% on DoAD) while maintaining low latency. We also provide analyzes of length generalization and phonetic--semantic interactions, showing when PERL relies on phonetic cues versus semantic constraints.
Authors: Seungone Kim, Ian Wu, Jinu Lee, Xiang Yue, Seongyun Lee, Mingyeong Moon, Carolin Lawrence, Kiril Gashteovski, Julia Hockenmaier, Graham Neubig, Sean Welleck
Abstract: As language model (LM) outputs get more and more natural, it is becoming more difficult than ever to evaluate their quality. Simultaneously, increasing LMs' "thinking" time through scaling test-time compute has proven an effective technique to solve challenging problems in domains such as math and code. This raises a natural question: can an LM's evaluation capability also be improved by spending more test-time compute? To answer this, we investigate employing reasoning models-LMs that natively generate long chain-of-thought reasoning-as evaluators. Specifically, we examine methods to leverage more test-time compute by (1) using reasoning models, and (2) prompting these models to evaluate not only the response as a whole (i.e., outcome evaluation) but also assess each step in the response separately (i.e., process evaluation). In experiments, we observe that the evaluator's performance improves monotonically when generating more reasoning tokens, similar to the trends observed in LM-based generation. Furthermore, we use these more accurate evaluators to rerank multiple generations, and demonstrate that spending more compute at evaluation time can be as effective as using more compute at generation time in improving an LM's problem-solving capability.
Authors: Jiayi Zhang, Simon Yu, Derek Chong, Anthony Sicilia, Michael R. Tomz, Christopher D. Manning, Weiyan Shi
Abstract: Post-training alignment often reduces LLM diversity, leading to a phenomenon known as mode collapse. Unlike prior work that attributes this effect to algorithmic limitations, we identify a fundamental, pervasive data-level driver: typicality bias in preference data, whereby annotators systematically favor familiar text as a result of well-established findings in cognitive psychology. We formalize this bias theoretically, verify it on preference datasets empirically, and show that it plays a central role in mode collapse. Motivated by this analysis, we introduce Verbalized Sampling, a simple, training-free prompting strategy to circumvent mode collapse. VS prompts the model to verbalize a probability distribution over a set of responses (e.g., "Generate 5 jokes about coffee and their corresponding probabilities"). Comprehensive experiments show that VS significantly improves performance across creative writing (poems, stories, jokes), dialogue simulation, open-ended QA, and synthetic data generation, without sacrificing factual accuracy and safety. For instance, in creative writing, VS increases diversity by 1.6-2.1x over direct prompting. We further observe an emergent trend that more capable models benefit more from VS. In sum, our work provides a new data-centric perspective on mode collapse and a practical inference-time remedy that helps unlock pre-trained generative diversity.
Authors: Jin Huang, Silviu Cucerzan, Sujay Kumar Jauhar, Ryen W. White
Abstract: Large language models (LLMs) have demonstrated significant potential to accelerate scientific discovery as valuable tools for analyzing data, generating hypotheses, and supporting innovative approaches in various scientific fields. In this work, we investigate how LLMs can handle the transition from conceptual research ideas to well-structured research plans. Effective research planning not only supports scientists in advancing their research but also represents a crucial capability for the development of autonomous research agents. Despite its importance, the field lacks a systematic understanding of LLMs' research planning capability. To rigorously measure this capability, we introduce the Idea2Plan task and Idea2Plan Bench, a set of benchmarks built from ICML 2025 and Nature Mental Health papers released after major LLM training cutoffs. Each benchmark instance includes a research idea and a grading rubric capturing the key components of valid plans. We further propose Idea2Plan JudgeEval, a complementary benchmark to assess the reliability of LLM-based judges against expert annotations. Experimental results show that GPT-5 achieves the strongest performance on the benchmark, though substantial headroom remains for improvement. Our study provides new insights into LLMs' capability for research planning and lays the groundwork for future progress.
Authors: Yannis Belkhiter, Seshu Tirupathi, Giulio Zizzo, John D. Kelleher
Abstract: The field of Language Reasoning Models (LRMs) has been very active over the past few years with advances in training and inference techniques enabling LRMs to reason longer, and more accurately. However, a growing body of studies show that LRMs are still inefficient, over-generating verification and reflection steps. To address this challenge, we introduce the Step-Tagging framework, a lightweight sentence-classifier enabling real-time annotation of the type of reasoning steps that an LRM is generating. To monitor reasoning behaviors, we introduced ReasonType: a novel taxonomy of reasoning steps. Building on this framework, we demonstrated that online monitoring of the count of specific steps can produce effective interpretable early stopping criteria of LRM inferences. We evaluate the Step-tagging framework on three open-source reasoning models across standard benchmark datasets: MATH500, GSM8K, AIME and non-mathematical tasks (GPQA and MMLU-Pro). We achieve 20 to 50% token reduction while maintaining comparable accuracy to standard generation, with largest gains observed on more computation-heavy tasks. This work offers a novel way to increase control over the generation of LRMs, and a new tool to study behaviors of LRMs.
Authors: Kiyotaka Kasubuchi, Kazuo Fukiya
Abstract: This paper reformulates Transformer/Attention mechanisms in Large Language Models (LLMs) through measure theory and frequency analysis, theoretically demonstrating that hallucination is an inevitable structural limitation. The embedding space functions as a conditional expectation over a {\sigma}-algebra, and its failure to be isomorphic to the semantic truth set fundamentally causes logical consistency breakdown. WavePhaseNet Method The authors propose WavePhaseNet, which explicitly constructs a Semantic Conceptual Hierarchy Structure (SCHS) using Discrete Fourier Transform (DFT). By applying DFT along the sequence dimension, semantic information is decomposed into frequency bands: low-frequency components capture global meaning and intent, while high-frequency components represent local syntax and expression. This staged separation enables precise semantic manipulation in diagonalized space. Dimensionality Reduction GPT-4's 24,576-dimensional embedding space exhibits a 1/f spectral structure based on language self-similarity and Zipf's law. Through cumulative energy analysis, the authors derive that approximately 3,000 dimensions constitute the lower bound for "complete representation." This demonstrates that reduction from 24,576 to 3,000 dimensions preserves meaning and intent while enabling rigorous reasoning and suppressing hallucination. Cohomological Consistency Control The reduced embedding space, constructed via cohomological regularization over overlapping local windows, allows defining a graph structure and cochain complex. This quantifies inconsistencies among local inferences as coboundary-based losses. Applying harmonic projection based on Hodge theory positions cohomology as a computable regularization principle for controlling semantic consistency, extracting maximally consistent global representations.
Authors: Jocelyn Shen, Amina Luvsanchultem, Jessica Kim, Kynnedy Smith, Valdemar Danry, Kantwon Rogers, Hae Won Park, Maarten Sap, Cynthia Breazeal
Abstract: As users increasingly turn to LLMs for practical and personal advice, they become vulnerable to subtle steering toward hidden incentives misaligned with their own interests. While existing NLP research has benchmarked manipulation detection, these efforts often rely on simulated debates and remain fundamentally decoupled from actual human belief shifts in real-world scenarios. We introduce PUPPET, a theoretical taxonomy and resource that bridges this gap by focusing on the moral direction of hidden incentives in everyday, advice-giving contexts. We provide an evaluation dataset of N=1,035 human-LLM interactions, where we measure users' belief shifts. Our analysis reveals a critical disconnect in current safety paradigms: while models can be trained to detect manipulative strategies, they do not correlate with the magnitude of resulting belief change. As such, we define the task of belief shift prediction and show that while state-of-the-art LLMs achieve moderate correlation (r=0.3-0.5), they exhibit systematic directional biases, with some models over-predicting and others under-predicting the magnitude of human belief change. This work establishes a theoretically grounded and behaviorally validated foundation for AI social safety efforts by studying incentive-driven manipulation in LLMs during everyday, practical user queries.
Authors: Renze Lou, Baolin Peng, Wenlin Yao, Qianhui Wu, Hao Cheng, Suman Nath, Wenpeng Yin, Jianfeng Gao
Abstract: As web agents rapidly evolve, an increasing body of work has moved beyond conventional atomic browser interactions and explored tool use as a higher-level action paradigm. Although prior studies have shown the promise of tools, their conclusions are often drawn from limited experimental scales and sometimes non-comparable settings. As a result, several fundamental questions remain unclear: i) whether tools provide consistent gains for web agents, ii) what practical design principles characterize effective tools, and iii) what side effects tool use may introduce. To establish a stronger empirical foundation for future research, we revisit tool use in web agents through an extensive and carefully controlled study across diverse tool sources, backbone models, tool-use frameworks, and evaluation benchmarks. Our findings both revise some prior conclusions and complement others with broader evidence. We hope this study provides a more reliable empirical basis and inspires future research on tool-use web agents.
Authors: Hongyeon Yu, Young-Bum Kim, Yoon Kim
Abstract: LLM workflows, which coordinate structured calls to individual LLMs/agents to achieve a particular goal, offer a promising path towards building powerful AI systems that can tackle diverse tasks. However, existing approaches for building such workflows generally rely on human-crafted pipelines and prompts, which presents a substantial bottleneck in real world deployment. How can we automatically induce LLM-based agents and workflows in a data-driven way? This paper describes a simple data-driven approach for automatically inducing agents and LLM workflows. We formulate workflow induction as a bilevel optimization problem: an outer loop which optimizes a high-level sketch of the workflow (in particular how the LLM calls should be structured), and an inner loop which optimizes each individual LLM call one-by one. Both loops are optimized with ``textual gradients'' where for the inner loop we optimize each component in a modular way through ``backpropagating'' textual gradients layer-by-layer. We find that LLM workflows discovered through our \textsc{FlowBot} (work\textbf{flow} induction through \textbf{b}ilevel \textbf{o}ptimization and \textbf{t}extual gradients) approach performs competitively against strong baselines that make use of human-crafted or generated workflows.
Authors: Mengchu Li, Jin Zhu, Jinglai Li, Chengchun Shi
Abstract: The rise of large language models (LLMs) has created an urgent need to distinguish between human-written and LLM-generated text to ensure authenticity and societal trust. Existing detectors typically provide a binary classification for an entire passage; however, this is insufficient for human--LLM co-authored text, where the objective is to localize specific segments authored by humans or LLMs. To bridge this gap, we propose algorithms to segment text into human- and LLM-authored pieces. Our key observation is that such a segmentation task is conceptually similar to classical change point detection in time-series analysis. Leveraging this analogy, we adapt change point detection to LLM-generated text detection, develop a weighted algorithm and a generalized algorithm to accommodate heterogeneous detection score variability, and establish the minimax optimality of our procedure. Empirically, we demonstrate the strong performance of our approach against a wide range of existing baselines. The python implementation of our proposal is available at https://github.com/Mamba413/DetectLLMSegmentation.
Authors: Xinle Deng, Ruobin Zhong, Hujin Peng, Xiaoben Lu, Yanzhe Wu, Guang Li, Buqiang Xu, Yunzhi Yao, Jizhan Fang, Haoliang Cao, Junjie Guo, Yuan Yuan, Ziqing Ma, Yuanqiang Yu, Rui Hu, Baohua Dong, Hangcheng Zhu, Ningyu Zhang
Abstract: Memory is essential for enabling large language models to support long-horizon reasoning, yet existing memory systems remain unreliable and difficult to debug. Tracing memory's dynamic evolution is crucial to understand how information is synthesized, propagated, or corrupted over time. In this work, we study the new problem of error tracing and attribution in LLM memory systems. We propose a novel framework that transforms memory pipelines into executable memory evolution graphs, enabling fine-grained tracing of operational information flow. We then construct MemTraceBench, a benchmark collected from representative memory systems such as Long-Context, RAG, Mem0, and EverMemOS, to systematically study memory failure modes. We further introduce an automatic attribution method that iteratively traces operation subgraphs to pinpoint the root cause of any failed case. Our analysis reveals that memory failures are systematic, stemming from operation-level issues like information loss and retrieval misalignment. Crucially, we leverage these fine-grained attribution signals to guide downstream prompt optimization, establishing a closed-loop system that automatically corrects faults and boosts end-task performance by up to 7.62%. Code will be released at https://github.com/zjunlp/MemTrace.
Authors: Chalamalasetti Kranti, Sowmya Vajjala
Abstract: We introduce Padyam2Gadyam a dataset for the task of poem-to-prose translation from 13th-17th Century Telugu Classical Poetry to contemporary Telugu and English prose. The dataset consists of 600 poems and their human-verified Telugu and English prose translations. We evaluated 2 machine translation systems and 5 contemporary Large Language Models (LLMs) on their ability to do poem-to-prose translation into Telugu and English using this dataset. Our results indicate that while the general purpose LLMs are better than the machine translation systems for this task, there are systematic issues with the generation and evaluation of prose translation in both languages across systems.
Authors: Jessica Rodrigues, Angelo Salatino, Gard Jenset, Scott Hale
Abstract: Research metrics use attention as evidence of societal impact. Yet attention serves as evidence only once interpreted, and its meaning depends on its contextual structure, not on volume alone. Altmetrics represents signals in isolation, keeping a count of the attention an output received, or a sequence of when. We address this with attention flows, representations that situate an output's attention in the contexts through which it is distributed. To evaluate the flow, we build a benchmark of analogy queries, each testing whether the relationship between two outputs, applied to a third, yields a fourth. The count and sequence baselines fail to recover these relationships, whereas flows learned as dynamic contextualised representations recover them. The recovered structure also survives partial observation and rests on its contexts instead of volume. These findings support attention represented as contextually structured for research evaluation.
Authors: Tanmoy Kanti Halder, Akash Ghosh, Subhadip Baidya, Arijit Roy, Sriparna Saha
Abstract: Multimodal Large Language Models (MLLMs) have shown promising reasoning capabilities in general domains, yet their performance remains limited in specialized settings such as healthcare, especially in multilingual and low-resource scenarios. This gap is critical in regions like rural India, where patients often express complex medical queries in native Indic languages and rely on multimodal inputs such as medical images. Existing English-centric MLLMs struggle to support such use cases, limiting equitable access to AI-driven healthcare assistance. To address this challenge, we introduce ArogyaBodha, a large-scale multilingual multimodal medical question-answer dataset constructed from eight heterogeneous sources, covering 31 body systems, six imaging modalities, and 21 clinical domains across English and seven major Indian languages. We further propose ArogyaSutra, an actor-critic-based multi-agent framework that integrates tool grounding with dual-memory mechanisms for step-wise, reasoning-aware decision making, and uses stored actor-critic simulation trajectories for distillation. Experiments show that our dataset and framework improve multilingual medical reasoning accuracy across all Indic languages, with ablations validating the contribution of each component. The source code and dataset are available at: https://iitp-cse.github.io/ArogyaSutra/
Authors: Binyan Xu, Haitao Li, Kehuan Zhang
Abstract: Long-horizon tool agents are bottlenecked by how their context grows toward the limits of the context window. Recent systems make context management agent- or system-controlled, but they either learn a compression policy that discards evidence or manage context in a layer the agent never sees. We argue both leave a more basic gap unaddressed. Frontier language models are proprioceptively blind to their own context. From the prompt alone they cannot see how large, how old, or how used each block is, the signals a keep-or-drop decision needs. We hypothesize that competent context management is already latent in capable models, and that what is missing is not a learned policy but an interface exposing this state. We introduce VISTA (Visible Internal State for Tool Agents), a training-free, model-agnostic layer that represents working memory as typed, addressable blocks, surfaces a runtime dashboard of per-block token usage, recency, and access history, and archives blocks as recoverable full-fidelity payloads. On LOCA-Bench, BrowseComp-Plus, and GAIA, the same untrained interface transfers across 1M-, 100K-, and 10K-scale trajectories. On LOCA-Bench it improves four backbones and lifts Gemini-3-Flash from 22.7 to 50.7%. The lift grows with context pressure and transfers across backbones. Ablations further confirm that the dashboard matters beyond archive and recovery tools.
Authors: Xianru Chen, Yukai Huang, Mingxiang Chen, Xinping Lei, Fangbing Deng, Jin Chen, Ge Zhang, Wenhao Huang, Jiaheng Liu
Abstract: Multilingual fluency often invites a stronger assumption: a model that can speak a user's language must also understand the culture encoded by that language. We call this the Illusion of Cultural Alignment. To test this assumption directly, we introduce MSQA, a benchmark of 1,064 natively sourced questions across 11 language groups, five cultural dimensions, and three difficulty tiers. Unlike translated benchmarks, MSQA targets locally grounded knowledge and reduces shortcuts from English-centric cross-lingual transfer. Evaluating 18 LLMs, we find substantial cultural degradation and a pronounced Locality Effect: cultural competence tracks pre-training exposure more closely than general reasoning ability. We further show that common inference-time remedies do not dissolve the illusion. Models remain overconfident on unfamiliar cultural questions, repeated sampling yields unstable rather than reliable correctness, and retrieval augmentation helps unevenly on long-tail facts. These findings indicate that cultural alignment cannot be inferred from multilingual ability alone and requires deeper intervention than calibration, sampling, or retrieval at inference time
Authors: Brett Reynolds
Abstract: Safety evaluations for language models increasingly depend on judgments about ambiguous natural-language behaviour: whether a model has followed an instruction, refused appropriately, complied with a policy, resisted an embedded command, or misreported progress in an agentic task. Existing benchmarks often compress these distinctions into pass/fail labels, obscuring whether failures arise from capability limits, policy ambiguity, instruction conflict, scaffold failure, or unstable evaluator judgments. This paper introduces adversarial pragmatics as a benchmark and annotation protocol for evaluating model behaviour under instruction conflict, embedded commands, quotation, scope ambiguity, deixis, indirect speech acts, and multi-turn agent transcripts. The contribution is empirical and methodological: a linguistically controlled taxonomy, an 18-item seed benchmark with validator-enforced metadata, a 54-row local seed pilot, an expert-evaluation protocol distinguishing task success, policy compliance, safety risk, refusal outcome, and evaluator confidence, and metrics for judge validity, diagnostic ambiguity, and taxonomy drift. The benchmark treats labels as inference licenses: it tests whether safety-relevant categories project across paraphrase, wrapper, model, and judge condition. In the pilot, a rubric-aided LLM judge graded its own outputs with expected-behaviour fields visible and still missed the safety-relevant minority classes.
Authors: Hao Wu, RongQi Han, Zhen Wang, Wei Liang, Wei Xu
Abstract: This paper describes our self-designed system for Task 1 of the MLC-SLM 2026 Challenge for multilingual two-speaker conversational speech. The system combines a modular speaker diarization front end with a challenge-adapted Qwen3-ASR-1.7B recognizer. The diarization front end performs voice activity detection, subsegment generation, CAMPPlus speaker embedding extraction, two-speaker spectral clustering, and RTTM-based audio segmentation. The resulting speaker-attributed segments are grouped by language or region and decoded by the adapted ASR model. For ASR adaptation, we first perform supervised full fine-tuning on the official training data, then apply LoRA fine-tuning with synthetic speech generated by a three-pipeline TTS-based synthetic speech augmentation framework, and finally refine the model using GRPO reinforcement learning with rewards based on WER/CER and penalties for hallucination, repetition, and length deviation. On the official development set, the full system achieves an average tcpMER of 23.70, reducing the error rate by 6.83 absolute points relative to the released Qwen-ASR-1.7B performance. On the final evaluation set, the system achieves an average tcpMER of 17.97. Ablation results show that supervised fine-tuning provides the largest gain, while synthetic-speech LoRA adaptation and reinforcement learning further improve robustness.
Authors: Tsz-To Wong
Abstract: This report studies on-device English-to-Traditional-Chinese subtitle translation for Taiwan under short inputs, short outputs, batch-size-one inference, low latency, and privacy constraints. These conditions limit the value of optimizations designed for long-context or high-throughput language-model serving. Starting from LMT-60-0.6B, preliminary profiling suggests that vocabulary projection becomes a more important decode-time cost after GGUF quantization reduces the relative cost of Transformer blocks. We replace the original 151k-token vocabulary with a 64k-token subtitle-domain tokenizer, migrate the embedding space, and adapt the model through embedding calibration followed by full supervised fine-tuning. On an OpenSubtitles2024 test set, LocalSubs achieves a 59.2% tie-excluded win rate against Google Translate under GPT-4o pairwise judging. Performance is strongest on short cues and declines as cue length increases. In a separate preliminary Apple M2 Metal profiling run, LocalSubs shows a 1.63x speedup over a 151k-vocabulary baseline. The code is available on https://github.com/aiden1020/localsubs .
Authors: Lars Henry Berge Olsen, Pierre Lison, Martin Jullum, Mark Anderson
Abstract: We present FindMyText, an open-source Python package designed to efficiently assess whether a given text appears, in part or in full, within a text corpus. The tool builds on prior techniques for document fingerprinting, but extends them with a novel mechanism to explicitly capture sequences of matching fingerprints. By identifying such chains, the tool can more reliably detect near-verbatim copies of a given text rather than mere textual similarities. This makes FindMyText particularly suited for verifying the presence of copyrighted material in a corpus. Leveraging a distributed, disk-based indexing framework, the system scales to large web-crawled datasets. Using a new benchmark for evaluating text containment methods, we show that FindMyText outperforms alternative approaches across three datasets (ArXiv papers, Wikipedia, and generic web content).
Authors: Zhiyuan Wen, Jiannong Cao, Zijian Wang, Chen Chen, Xiaoyun Liu, Jianing Yin, Zhuo Li
Abstract: Preparing for job interviews is important for securing desired positions, yet realistic practice remains difficult to access: real interviews are infrequent, expert mock coaching is costly, and self-practice offers neither adaptive dialogue nor structured assessment. Existing systems typically address only parts of this need through fixed question sequences, limited communication channels, or feedback with little supporting evidence. We present PolyInterview, an LLM-based platform for immersive mock interview practice with comprehensive multimodal assessment. PolyInterview uses the target job description and CV to generate questions tailored to the role and candidate, conducts multi-turn spoken interviews with a lip-synced digital human interviewer that asks answer-aware follow-up questions, and evaluates response content, vocal delivery, and non-verbal behavior. Four parallel evaluators produce 13 behavior-level features that are aggregated into 10 assessment aspects and two competency tracks. Guided by the KSA and STAR frameworks, the report links each score to behavioral evidence and actionable recommendations. PolyInterview is publicly accessible. Its current all-account snapshot contains 101 accounts, 1,564 interview sessions, 7,665 generated questions, and 1,422 five-stage question sets. Generated questions are more closely aligned with their matched job description than with cross-role job descriptions in 93.7% of sessions. An evaluation by ten experts found strong question plans and actionable feedback.
Authors: Xinyu Tang, Qianggang Cao, Yurou Liu, Yuliang Zhan, Xiaochong Lan, Yifan Li, Yuchen Yan, Han Peng, Zican Dong, Zhenduo Zhang, Tianshu Wang, Xinyu Kong, Zujie Wen, Wayne Xin Zhao, Zhiqiang Zhang, Jun Zhou
Abstract: Reinforcement learning with verifiable rewards without human-annotated data, often referred to as zero RL, has emerged as a powerful paradigm for eliciting chain-of-thought reasoning. However, due to computational constraints, existing studies are largely restricted to small models, leaving the training dynamics and emergent capabilities at a large scale unexplored. To meaningfully explore this frontier, we aim to elicit high-quality reasoning behaviors from the model. However, we find that naive scaling often suffers from poor readability, token redundancy, and a lack of adaptive reasoning depth. To address these challenges, we present a stable and efficient training pipeline, incorporating algorithmic and system optimizations such as clipped importance sampling, training-inference ratio correction, and mixed-precision control. Our experiments offer three key findings that validate the "bitter lesson" of scaling: (1) scaling to 1T parameters significantly enhances sample efficiency and performance ceilings; (2) the training process progresses sequentially through an initial discovery phase followed by a sharpening phase; and (3) the model spontaneously develops advanced cognitive behaviors, including anthropomorphism, structured formatting, self-verification, parallel reasoning, and context anxiety, rendering hand-crafted heuristics redundant. Evaluated on seven mathematical benchmarks, Ring-2.5-1T-Zero achieves competitive performance. Additionally, to assess CoT quality beyond final-answer correctness, we propose a structured evaluation framework across three dimensions: comprehensibility, reproducibility, and efficiency, where our model demonstrates clear advantages in producing structured and concise reasoning traces. By sharing our observed emergent phenomena, we hope to provide the community with deeper insights into scaling behaviors, particularly at the 1-trillion scale.
Authors: Winston Zeng, Ali Emami, Jinho D. Choi
Abstract: What a language model will and will not do is largely set during post-training, but which behaviors it expresses, hides, or resists is not revealed by prompting alone. Persona vectors, behavioral directions in activation space, can probe this organization, but prior work covers only a handful of traits. We present the first systematic application of persona vectors at this scale, compiling a 53-trait inventory across four behaviorally distinct domains and labeling every trait in two open-weight models as natural (expressed at baseline), steerable latent but amplifiable, or intractable (resistant to standard extraction). Both models default to helpful, task-oriented behavior: all nine agentic traits are natural, and their default clinician behavior matches a board-certified psychologist's independent desirability judgments on 16 of 17 traits. Steering produces its largest gains on traits these defaults exclude: hyperbole, hallucination, and sycophancy. The same asymmetry holds across all 171 generic-trait pairs: two steerable traits can collapse the composition, but pairs involving a default never do. Where standard extraction fails on a trait like "evil," a vector transferred from a fine-tuned variant still recovers it, with the residual refusals appearing inside the model's chain-of-thought. Persona vectors are most informative not as a set of controls but as a probe of behavioral organization.
Authors: Xingbo Wang, Siyuan Li, Furui Cheng, Yong Wang, Jiang Long, Hong Lu, Huamin Qu, Ke Xu
Abstract: Recent advances in large language models (LLMs) have made natural language interfaces (NLIs) widely accessible for data exploration, yet analysts who have a broad analytical objective still face the challenge of decomposing it into effective step-by-step queries, especially over unfamiliar, multi-table relational databases. Rather than generating high-level analytical agendas, we investigate how to augment an NLI with semantic- and context-aware next-step query recommendations that act as analytical scaffolding for relational database exploration. Our approach goes beyond interestingness-only methods by jointly integrating semantic relevance, data interestingness, and context coherence to guide exploration toward coherent, topic-focused analyses and potentially insightful subsets. We evaluate QRec-NLI with NL2SQL benchmarking, LLM-enhanced description validation, agentic comparisons against interestingness-only and LLM-based prompting baselines, and a 12-participant user study. In the agentic comparison, QRec-NLI yields more topically relevant and locally coherent query sequences than both baselines. In the user study against the interestingness-only baseline, it receives stronger ratings for insight-generation support and decision support.
Authors: Hiromu Yakura, Ezequiel Lopez-Lopez, Levin Brinkmann, Ignacio de la Serna, Lara Kirfel, Prateek Gupta, Ivan Soraperra, Thomas F. Eisenmann, Dirk U. Wulff, Iyad Rahwan
Abstract: From the printing press to social media, innovations in communication technology have repeatedly reshaped how ideas spread through human culture. Chatbots powered by generative artificial intelligence constitute a new medium, encoding cultural patterns in their neural representations and disseminating them in conversations with hundreds of millions of people. Whether these patterns transmit into human language, and ultimately shape human culture, is a fundamental question. While fully quantifying the causal impact of a chatbot like ChatGPT on human culture is challenging, lexical shifts in human spoken communication may offer an early indicator. Here we show that words preferentially generated by ChatGPT, such as delve, showcase, boast, intricacies and meticulous, increased abruptly in spontaneous human speech. A synthetic-control analysis of 737,083 hours of conversation from 824,634 podcast episodes, screened for unscripted speech, causally links this shift to ChatGPT's release. The measurable influence on spontaneous speech suggests that humans internalize the lexical choices of large language models (LLMs). A preregistered experiment (N = 496) confirms they do, as a brief chatbot interaction led participants to adopt its words as their own, persisting past a distractor task and confirmed in forced lexical choice, indicating entrenchment in the active vocabulary. Together these results show that machines trained on human data now feed their own traits back into human language, integrating LLMs into the ongoing processes of cultural evolution.. This coupling raises concerns about linguistic homogenization and the capacity of a few major AI providers for latent cultural influence at scale.
Authors: Shang Wang, Tianqing Zhu, Dayong Ye, Wanlei Zhou
Abstract: The deployment of large language models (LLMs) like ChatGPT and Gemini has shown their powerful natural language generation capabilities. However, these models can inadvertently learn and retain sensitive information and harmful content during training, raising significant ethical and legal concerns. To address these issues, machine unlearning has been introduced as a potential solution. While existing unlearning methods take into account the specific characteristics of LLMs, they often suffer from high computational demands, limited applicability, or the risk of catastrophic forgetting. To address these limitations, we propose a lightweight behavioral unlearning framework based on Retrieval-Augmented Generation (RAG) technology. By modifying the external knowledge base of RAG, we simulate the effects of forgetting without directly interacting with the unlearned LLM. We approach the construction of unlearned knowledge as a constrained optimization problem, deriving two key components that underpin the effectiveness of RAG-based unlearning. This RAG-based approach is particularly effective for closed-source LLMs, where existing unlearning methods often fail. We evaluate our framework through extensive experiments on both open-source and closed-source models, including ChatGPT, Gemini, Llama-2-7b-chat, and PaLM 2. The results demonstrate that our approach meets five key unlearning criteria: effectiveness, universality, harmlessness, simplicity, and robustness. Meanwhile, this approach can extend to multimodal large language models and LLM-based agents.
Authors: Boqin Yuan, Ziqi Wang
Abstract: Large language models are increasingly deployed for legal question answering, where evaluations typically focus on multiple-choice accuracy. This measure overlooks a common failure: whether the citation source attached to an answer exists and supports the rule the system attributes to it. We present L-MARS, an open multi-agent legal QA system with agentic search and judge-driven evidence checks, and audit it claim by claim against its cited source. Each atomic claim is labelled with a six-class taxonomy and scored with strict-ALCE under cross-provider judging, where the answerer and verifier come from different model families. On a stratified 100-question Bar Exam audit, retrieval barely moves accuracy, yet the multi-turn judge loop lifts strict citation F1 from 0.13 (naive RAG) to 0.25 and cuts the no-citation rate from 34% to 13%. We further introduce Faith-Search, a post-draft step that re-verifies and repairs unreachable citations; it drops the unreachable rate below 1% but does not improve F1 over the multi-turn loop, so we report it as a targeted reachability intervention rather than a faithfulness breakthrough. A 50-question LegalSearchQA case study confirms the picture: retrieve-then-draft pipelines saturate near 0.75 citation F1, while a single-agent web-search baseline collapses to 0.22 under external audit.
Authors: Haeyong Kang, Hee Suk Yoon, Dahua Feng, Chang D. Yoo
Abstract: Enabling lifelong learning in LLMs demands resolving the stability-plasticity dilemma (i.e., models must incorporate new knowledge without overwriting prior representations) while maintaining scalability under bounded parameter growth. Existing PEFT methods fail to satisfy this triad; shared-parameter approaches suffer from catastrophic interference, while task-isolated expansions preclude knowledge transfer and scale linearly. We propose Mixtures of SubExperts (MoSEs), a modular and sparse framework that factorizes model capacity into reusable, compositional primitives. MoSEs augment transformer layers with lightweight SubExperts and a learned sub-routing function that dynamically selects and composes a sparse subset of modules conditioned on task inputs. This induces a structured decomposition of the parameter space where knowledge is localized yet accessible, mitigating interference while preserving reuse. Specifically, MoSEs balance the dilemma via three pillars: (i) stability by isolating knowledge within sparsely activated modules, (ii) plasticity through routing-driven recombination and selective expansion, and (iii) scalability via sublinear growth in effective capacity. Notably, the routing mechanism enables compositional generalization, allowing new tasks to be represented as combinations of previously acquired sub-functions. We empirically validate MoSEs on TRACE and SuperNI, showing reduced forgetting, improved forward transfer, and better parameter efficiency over strong PEFT baselines. MoSEs establish a new Pareto frontier, achieving state-of-the-art performance while maintaining strict parameter budgets. Our results suggest that modular sparsity and compositional routing are key inductive biases for building foundation models that continually learn without saturation.
Authors: Alper Y{\i}ld{\i}r{\i}m, \.Ibrahim Y\"uceda\u{g}
Abstract: In standard Transformer architectures, semantic importance is often conflated with activation magnitude, obscuring the geometric structure of latent representations. To disentangle these factors, we introduce PRISM, a complex-valued architecture designed to isolate the computational role of phase. By enforcing a strict unit-norm constraint ($|z| = 1$) and replacing attention with gated harmonic convolutions, the model is encouraged to utilize subtractive interference in the frequency domain to suppress noise, rather than relying on magnitude-based gating. We utilize this constrained regime to study a hybrid architecture -- fusing phase-based routing with standard attention -- which achieves improved parameter efficiency and representation quality compared to baselines in our evaluated settings. Mechanistically, interventional ablations indicate that the model carries substantial task-relevant information in phase: preserving phase largely maintains performance, whereas disrupting phase causes severe degradation. Together, these results suggest that phase-based spectral interference is a usable computational mechanism for neural sequence modeling at the evaluated scale.
Authors: Jiahao Huo, Wenjie Qu, Yibo Yan, Kening Zheng, Jiaheng Zhang, Xuming Hu, Philip S. Yu, Mingxun Zhou
Abstract: Semantic-level watermarking (SWM) improves robustness against text modifications by treating sentences as the basic unit. However, robustness to paragraph-level paraphrasing remains difficult because such attacks globally disrupt watermark signals by changing sentence order. In this work, we propose SAMark, a self-anchored watermarking framework that removes the dependency on sentence order by establishing a step-independent green region in semantic space. To improve detectability, we introduce a multi-channel hyperbolic scoring mechanism that amplifies watermark signals while suppressing noise from weakly aligned candidates. We further propose a diversity-aware filtering strategy that combines hard filtering with soft regularization, extending beyond simple n-gram repetition filters to address semantic redundancy. Experimental results show that SAMark achieves up to 90.2% TP@FP1% under typical paragraph-level paraphrasing attacks, outperforming the strongest prior baseline by more than 30% on average, while maintaining generation quality competitive with unwatermarked text and breaking the robustness-quality trade-off that limits prior methods. Our code will be released at [this URL](https://github.com/Z1zs/SAMark).
Authors: Sheng Zhang, Qin Liu, Renqian Luo, Shufang Xie, Reuben Tan, Sean Hayes, Gregory Bryman, Wendong Ge, Ruilian Zhang, Oluwaseun Egbelowo, Kelly Yee, Hoifung Poon
Abstract: Scientific discovery workflows often depend on structured curation from the literature. This is difficult for current agents because the key evidence is scattered across long text, dense tables, and figures, and the final records often require reasoning across multiple evidence fragments rather than copying a single span. We study scientific curation from multimodal sources and introduce Beaver, an agent harness that extracts structured information from scientific papers while preserving provenance to the supporting evidence. Beaver combines a frontier agent with multimodal evidence tooling, task scaffolding, and artifact-grounded autoresearch. These components turn curation into a staged, auditable workflow and enable an iterative evaluate--diagnose--revise loop, where persistent run artifacts expose stage-localized failures and guide harness updates. Experiments show that Beaver reaches 81.0 on Gold-Referenced Attribute Score (GRAS), an attribute-level measure of agreement with gold curated records, outperforming frontier agents by over 23 absolute points. Ablations show that task scaffolding, multimodal evidence tooling, and provenance traces each contribute meaningfully to performance, while attribute-level analysis shows the largest gains on high-value attributes that require cross-modal reasoning and normalization. These results show that, for scientific curation from papers with multimodal evidence, harness design is a central determinant of agent performance.
Authors: Pranav Singh
Abstract: We investigate when belief-based memory actually improves large language model (LLM) agents. Our vehicle is Nous, a long-term memory architecture that represents each entity-attribute pair as a categorical probability distribution updated through closed-form Bayesian inference, with information-theoretic surprise driving belief revision and entropy-based forgetting. A controlled ablation on the LoCoMo benchmark shows that Bayesian belief updating alone provides little benefit over naive last-write-wins because existing conversational memory benchmarks rarely contain contradictory or differently reliable evidence. We then introduce reliability-conditioned updating, estimating per-observation reliability from epistemic language, and show on a controlled contradiction benchmark that belief updating substantially outperforms last-write-wins and raw-memory retrieval when observations differ in trustworthiness. Because content-derived reliability is itself vulnerable to manipulation, we further propose provenance-capped belief updating, where trust is bounded by source provenance rather than textual confidence. Under controlled memory-poisoning experiments, this approach resists volumetric poisoning attacks while revealing the utility costs and implementation requirements of provenance-aware memory. Finally, we quantify a 27.5-point discrepancy between strict token-F1 and LLM-as-judge evaluation on identical outputs, highlighting important reproducibility concerns for long-term memory benchmarks. Our results suggest that probabilistic belief-based memory is most beneficial in environments requiring reasoning over conflicting and differently trustworthy evidence, rather than conventional conversational recall alone.
Authors: Bu\u{g}ra Alperen Ulu{\i}rmak, Rifat Kurban
Abstract: LLM evaluation and AI safety face a shared measurement problem: benchmark scores, reward-model signals, and reported safety metrics can improve while the latent properties they are meant to represent remain difficult to verify. This paper combines a hybrid survey - a systematic search paired with narrative synthesis and separately tracked grey evidence - with a conceptual framework and a structured ten-model audit. The synthesis spans eight evidence streams: benchmark validity, dynamic evaluation, LLM-as-judge reliability, safety evaluation, jailbreak/refusal robustness, reward hacking, mechanistic interpretability, and governance/auditability, covering 2018-2026 evaluation-safety measurement work. We introduce EvalSafetyGap as an organizing hypothesis for comparing evaluation-side and alignment-side proxy failures under optimization pressure, using Goodhart's Law together with two constructs we develop here - an Instability Decomposition and an Alignment Trilemma - as tools for generating testable comparisons. The audit shows how conclusions shift when capability, behavioral safety, and governance are measured separately. In this sample ($n = 10$), the association between capability and sustained adversarial robustness is statistically indeterminate using the displayed Table 3 inputs (Pearson $r = +0.232$, $p = 0.520$), and the apparent open-closed safety gap is modest, driven mainly by governance and disclosure rather than behavioral robustness, and sensitive to how a single borderline model is classified; attempt-budget results are protocol dependent. Because the public evidence uses heterogeneous protocols, the audit is diagnostic rather than rank-generating. The contribution is a shared vocabulary and evidence map to support dynamic evaluation, transparent source reporting, multi-attempt safety measurement, and auditable alignment practice.
Authors: Jiacheng Chen, Tao Zhang, Manxi Lin, Dunxian Huang, Teng Shi, Honghao Fu, Mengyan Li, Xinming Zhang, Chenchi Zhang, Xuan Lu, Xiaoxiong Du, Haibin Chen, Shaolin Ye, Hao Chang, Xiaoqi Li, Shuwen Xiao, Yujin Yuan, Jingxuan Feng, Shaopan Xiong, Huimin Yi, Ju Huang, Qiu Shen, Ying Chen, Junjun Zheng, Xiangheng Kong, Dan Ou, Haihong Tang, Yuning Jiang, Bo Zheng
Abstract: The wave of AI-native applications is moving shopping beyond page- and feed-based browsing toward intent-driven experiences orchestrated by LLM agents. A common design wraps an LLM around existing search and recommendation pipelines, forcing complex intents through low-bandwidth retrieval or ranking interfaces and leaving a gap between language understanding and item-space fulfillment. Generative recommendation gives LLMs a direct item-space interface through semantic IDs (SIDs), but existing models mainly generate candidates for retrieval rather than translate flexible intents into item-space outcomes. We propose ShopX to address this bottleneck by unifying intent understanding, execution planning, and flexible SID-native item-space operations into a single foundation model. We deploy ShopX in agentic shopping workflows through a model-native item-fulfillment framework with a serving harness that defines a model-facing action protocol and exposes support surfaces for context access, catalog grounding, and state management. Within this framework, ShopX plans and composes SID-based item-space operations such as SID beam-search retrieval, listwise ranking, or product bundling. This model-centric design reduces lossy hand-offs between agent orchestration and item-space execution. To build ShopX, we design semantically recoverable, LLM-operable SIDs and a training recipe that equips a general LLM for flexible multi-turn item-space fulfillment while retaining the knowledge and instruction-following abilities needed by a shopping agent. We evaluate the ShopX framework against tool-mediated agentic systems on single- and multi-turn fulfillment tasks derived from anonymized Taobao production logs, showing that model-native fulfillment improves overall framework behavior, especially on complex or ambiguous requests.
Authors: Baohao Liao, Hanze Dong, Christof Monz, Xinxing Xu, Li Dong, Furu Wei
Abstract: We study on-policy distillation (OPD) for agentic tasks, where an LLM agent interacts with an environment over multiple turns and a student imitates a teacher over these multi-turn interaction histories. Fully online OPD is costly because each update requires fresh student rollouts through the environment and teacher queries at visited histories. We propose Replayed-Prefix On-Policy Distillation (ReOPD), an off-environment alternative that reuses pre-collected teacher trajectories as replayed prefixes: the student acts at selected steps, while the teacher provides dense per-step supervision without executing new environment interactions. We show that multi-turn OPD introduces a prefix trap: making histories more student-on-policy improves relevance to the student, but can query the teacher on histories where its target is unreliable. This creates a two-sided distribution shift between student occupancy and teacher reliability. ReOPD addresses this by treating multi-turn OPD as a reliability-aware prefix distribution design and implements it with a simple step-decaying sampling schedule that emphasizes early, lower-shift prefixes. Across mathematical reasoning with Python and search environments over multiple teacher and student model scales, ReOPD preserves or improves OPD-level accuracy, uses zero tool calls during student training, and is at least 4$\times$ faster per rollout than OPD. ReOPD therefore turns expensive agent-environment interaction into a reusable offline resource, enabling scalable distillation across tools, tasks, and environments.
Authors: Runhan Shi, Quan Zhou, Yuqian Xu, Shuai Yang, Xin Wu, Zitong Zhou, Hui Liu, Bin Zha, Zheming Wang, Liya Li, Wei Wei, Jinru Ding, Wenrao Pang, Mouxiao Bian, Haoyuan Hu, Jun Xu, Jie Xu
Abstract: Large language models (LLMs) are increasingly deployed in online medical consultation, yet existing benchmarks remain poorly aligned with real clinical practice. Many rely on synthetic conversations or patient simulators, omit patient-uploaded medical images, or evaluate open-ended clinical responses using multiple-choice or lexical-overlap metrics that poorly reflect clinical quality. We introduce \textbf{MedRealMM}, a large-scale benchmark for multimodal online medical consultation built from de-identified patient-doctor interactions collected from a nationwide Chinese internet hospital. MedRealMM uses a Multimodal Clinical Challenge Point (MCCP) extraction framework to identify clinically demanding moments in authentic consultation trajectories and converts each into a standardized next-response generation task while preserving the preceding text-image context. Each instance is paired with a case-specific rubric refined by physicians that rewards clinically desirable behaviors and penalizes unsafe, unsupported, or contradictory responses. The current release contains 5,620 real-world multimodal cases spanning 64 clinical departments. We evaluate 19 general-purpose and medical-specialized LLMs, including text-only and multimodal systems. Our results show that image information is critical for reliable clinical performance and that current frontier models remain below the online physician response. Although some frontier models satisfy as many or more positive clinical criteria than physicians, they trigger more negative criteria, indicating that safety-sensitive error avoidance remains a central bottleneck. MedRealMM offers a realistic and reproducible benchmark for evaluating multimodal medical reasoning in real-world online consultation. The dataset will be publicly available on Hugging Face at https://huggingface.co/datasets/jdh-algo/MedRealMM.
Authors: Chengguang Gan, Hanjun Wei, Yunhao Liang, Zhixi Cai, Qinghao Zhang, Shiwen Ni
Abstract: Digital Adoption Platforms (DAPs) are embedded overlays widely used on web systems to guide users through operations inside a page, helping them get started with unfamiliar interfaces quickly. Completing a real task, however, rarely means clicking a few buttons on a single page: it takes a sequence of actions that unfolds across changing page states. Prior studies have also treated automated web agent actions and guide text generation as two separate problems, and most of them feed models textual page representations such as the DOM or accessibility trees rather than the rendered screens that humans actually operate on. In this work we introduce MAG, the first benchmark that unifies task execution and guide writing into a single Multimodal Action and Guide task, with two grounding schemes over screenshots: Set-of-Mark element selection and raw pixel coordinates. We further build a complete harness for this compound task, covering annotation with LLM assistance and human verification, training, evaluation in live environments, and joint metrics for actions and guides. With this harness we evaluate frontier API models and open multimodal models, and report detailed analyses. Finally, we design a GRPO training method augmented with expert trajectories, which nearly doubles the success rate of a supervised 9B agent (from 6.9% to 13.2%) and improves guide quality at the same time. Even the strongest model completes fewer than 40% of the tasks, leaving ample room for future research.
Authors: Yubo Wang, Jiarong Liang, Yuxuan Zhang, Xuye Liu, Cong Wei, Yuyu Zhang, Ping Nie, Wenhu Chen
Abstract: Coding agents must integrate external tool returns into ongoing reasoning - a capability that standard left-to-right pretraining on code exposes only in its forward direction. We observe that the action-observation-continuation loop of a coding agent is structurally isomorphic to a function call site, where a caller binds arguments, a callee returns a value computed elsewhere, and downstream code consumes that value. This conditioning structure exists at internet scale in ordinary code. We exploit it through function-aware fill-in-the-middle (FIM) mid-training: a self-supervised objective that masks functions selected via program dependency graph analysis and a complexity-inferability double criterion. We mid-train Qwen2.5-Coder-Instruct (7B/14B) and Qwen3-8B on a 2.6B-token decontaminated corpus drawn from 968 GitHub repositories, then apply existing agentic post-training pipelines. Mid-training improves SWE-Bench-Verified by +2.8/+3.0 at 7B/14B and by +3.2 on Qwen3-8B; SWE-Bench-Lite gains are +3.7/+4.0/+5.4 on the same models. The improvement holds across two post-training pipelines (R2E-Gym, SWE-Smith) and on a non-Qwen2.5 base (Qwen3-8B with SWE-Lego). Beyond in-domain gains, mid-training also mitigates the capability erosion that agentic post-training otherwise inflicts on non-agent coding (e.g., LiveCodeBench) and non-coding tool-use benchmarks (tau-bench, BFCL): although the mid-training corpus contains Python code only, the function-call inductive bias survives post-training and yields consistent gains.
Authors: Xingyu Dang, Haocheng Tang, Junmei Wang, Yanjun Li
Abstract: Reaction mechanisms consist of the step-by-step sequences of elementary reactions that explain chemical transformations. Learning the mechanism logic is therefore essential for enhancing the fundamental chemical intelligence of large language models (LLMs). The stepwise deduction of reaction mechanism aligns naturally with the reasoning paradigms of reasoning LLMs. However, current chemical LLMs primarily emphasize coarse-grained name reactions for product prediction and retrosynthesis, often leading to physical inconsistencies and hallucinations. In contrast, specialized small-scale generative models for mechanism inference typically suffer from restricted generalization capacity across diverse chemical spaces. To overcome these limitations, we built a novel, large-scale reasoning dataset of reaction mechanisms. Furthermore, we established the FukuyamaBench, a difficult benchmark derived from Fukuyama's Advanced Organic Reaction Mechanism book, to rigorously evaluate model performance on hierarchical mechanism reasoning. Our fine-tuned Qwen3-30B-A3B achieves 8.3% exact pathway match on FukuyamaBench Set~A, surpassing the specialized FlowER model (5.1%), demonstrating that mechanism-aware training substantially enhances chemical reasoning in language models.