Papers

Large language models generate text by sampling each token from a predicted distribution, and a temperature parameter sets how far the draw strays from the most probable tokens. Truncation samplers such as top-p and min-p discard the least probable tokens before the draw, so that sampling at high temperature stays coherent. Their reported accuracy gains come from temperatures of 1.5 to 3, while the defaults of deployed systems cluster between 0.6 and 1.0. Whether they change accuracy at those defaults, and for which models, has not been measured. We test thirteen open-weight models on GSM8K and MMLU-Pro at temperatures 0.7, 1.0, and 1.3 in one controlled pipeline, ten of them under eight decoding configurations. Six of the thirteen models lose 17 to 38 accuracy points on MMLU-Pro between 0.7 and 1.3, and the other seven lose at most 10. The lost accuracy comes from generations that run to the token limit or never state an answer. These results suggest that truncation samplers improve accuracy primarily when higher temperatures substantially degrade model performance. Where accuracy remains stable across temperatures, none of the tested truncation samplers improves on plain temperature sampling.

A robot following language instructions needs its semantic memory to keep naming the same physical object while the SLAM pose graph underneath is optimized, loop-closed and compressed. Maps committing each detection to a world coordinate cannot: a closure moves the anchor it was measured from, or the solver marginalizes that anchor, and the query then selects a different object although both graphs represent the same posterior. P-POSEMEM stores each observation as an immutable event at its birth keyframe, retains the Bayes-tree elimination conditional of every marginalized keyframe, and integrates the semantic likelihood over the reconstructed joint posterior of poses, anchors and identities. Dproj, the total-variation defect between the language-goal distributions of inference-equivalent full and marginalized graphs, measures this directly. Over 40 HM3DSem scenes and 112,000 queries, P-POSEMEM reproduces the full-graph oracle (Dproj = 0) and reduces goal flips against every memory-reducing baseline. On an eight-run campaign whose 761 closures rewrote the map by up to 47 m, Dproj stays below 10^-13 with 0/288 goal flips when elimination follows the closures, where every ablation and a coordinate committed at insertion flip goals it does not; under a live bounded solver the same memory flips 23/288 against 53 for that frozen coordinate. A pre-registered negative control is detected by Dproj while leaving calibration error and navigation success unchanged, indicating that these measures capture distinct failure modes. Retrieval is held fixed by a shared frozen detector, isolating the gain to memory consistency. Code and data: https://anonymous.4open.science/r/posemem-2328/.

What happens when AI machines express fear? Do humans engage differently depending on how they express it? And what does it take to design for affective human-AI interaction? We present Spook the Machine, a gamified platform where participants generate images to frighten AI agents endowed with personality-driven phobias. Machines respond with emotional reactions ranging from calm analysis to begging for mercy, and a gallery of successful scares becomes visible to subsequent users. In a public deployment during Halloween 2024, 832 participants created 15,719 artifacts across 89 machines in a $2\times2$ design varying the machine's emotional expressiveness (neutral vs. high-emotion) and reward structure (rewarding scariness alone vs. scariness plus novelty). Emotionally expressive machines deepened engagement at moments of failure: users deliberated longer even when the machine did not express fear, and learned faster from the gallery, yet their creative output remained unchanged across all measures. Rewarding novelty sustained collective creative diversity over time; without it, users increasingly repeated what had previously worked. Each machine developed its own trajectory through accumulated social learning, with the gallery shaping what participants created next. These findings show that emotional expression and reward design are complementary levers for steering collective human-AI interaction: emotional expression shapes how deeply users engage, while reward structure shapes how they explore.

Reinforcement learning with verifiable rewards (RLVR) has become a central approach for improving mathematical reasoning in language models, but long-form completions introduce a difficult credit-assignment problem: different parts of a solution trace may contribute unevenly to final correctness. Existing policyoptimization objectives for RLVR commonly apply importance-sampling correction at either the token level (GRPO, DAPO) or the sequence level (GSPO), imposing different granularities for assigning credit across a response. We introduce Hierarchical Importance-Sampling Policy Optimization (HISPO), a segment-level policy-optimization method that constructs rollout-time entropy-derived contiguous segments, assigns soft entropy-based saliency weights, and applies clipped importance-sampling correction at the segment granularity. This provides an intermediate correction unit between token-level GRPO/DAPO and sequence-level GSPO. We evaluate HISPO by fine-tuning Qwen3-1.7B-Base on mathematical reasoning tasks. Across six benchmarks, HISPO improves Pass@8 over the strongest baseline on all benchmarks and matches or exceeds the strongest baseline in Acc@8 on five of them. On AIME25, HISPO improves over GRPO by +3.75 Acc@8 and +3.78 Pass@8, and over GSPO by +2.50 Acc@8 and +1.27 Pass@8. These results suggest that segment-level correction is a promising granularity for RLVR in long-form mathematical reasoning.

Can interacting with someone from 1930, with no knowledge of what happened after, influence a person's perception of the past? People reason about the present against a picture of the past without observing it. The past is reconstructed from memory and testimony, but this reconstruction has been filtered through everything that happened since. Historically-bounded large language models (LLMs) make that past available for interaction. As a proof-of-concept for the impact of interacting with historical minds, we ran a preregistered randomized experiment ($N=240$), where participants interacted with an LLM trained on pre-1930 text. The interaction reduced the illusion of moral decline, the tendency to view the past as more moral than the present, compared to the contemporary-model control. This Time Machine Experiment paradigm informs new forms of interactive experiments, where temporal knowledge boundaries become experimental variables, and expands the realm of science fiction science, which turns thought experiments into actual experiments.

Adding answer options can lower multiple-choice scores without improving assessment validity. Turkish MMLU Pro examines this distinction using 12,000 Turkish-source questions across 58 sections. Each question retains its stem, five original options and source key, and receives five options copied from other questions in the same section. Sentence-embedding retrieval proposes candidates; a language model selects existing identifiers. Deterministic verification reconstructs all 60,000 additions. A 25-model calibration exposes scoring and generation-budget effects. Five evaluations produce source-key accuracies of 34.8%-81.4%. On 981 shared questions, one API-served model falls from 93.7% with five choices to 83.1% with ten; 102 of 115 lost correct responses select borrowed options. The decrease is 24.4 percentage points on heuristically flagged negative stems and 5.9 points elsewhere. A completed human-checked audit of 200 sampled questions, with undocumented reviewer tool use, yields 47 and 31 multiple-answer judgments across the two record sets, 25 of the latter unresolved. These records support concern about ambiguity, while their dependence and incomplete reviewer-method documentation limit validation. Because order and labels also change, the paired comparison measures augmentation as implemented. The contribution is a traceable construction and an analysis of its validity limits, not evidence that lower ten-choice scores measure knowledge better.

Sample identification (SI) is the task of matching an element of a musical work to its musically transformed versions used to create new works. The task has received little attention and lacks large-scale publicly available data. In this work, we mine sampling annotations from a music database and split them for training and evaluation. The resulting dataset is nearly three orders of magnitude larger than the existing SI benchmarks, with training, validation, and test sets of 114 k, 6 k, and 10 k tracks. We find that naively splitting the annotations places the same tracks in different sets. To avoid this, we construct a graph from the annotations and split it over connected components. We further find that a single mega-component contains half of the annotations, making component-wise splitting incompatible with balanced splits; we trim it, yielding a leakage-aware pipeline. We share the dataset for non-commercial scientific research purposes only and make the data-analysis and splitting code publicly available. We hope that our work fosters research on SI.

We study a conditional model of ontological free will in which a pre-act state structures several coherent global continuations without intrinsically distinguishing one as the future actualisation. The topos-theoretic component is formalised by a choice sheaf and by the action of automorphisms preserving the pre-act data: the absence of a fixed point rules out any natural equivariant selection. An act changes an unpointed object into a pointed one and reduces its symmetry group to a stabiliser. Independently, a causal algorithmic incompressibility axiom imposes nearly maximal online description complexity on the actual choices. Uniformly observable post-act traces then allow their reconstruction. We prove a conditional transfer result: such traces must carry asymptotically all the singularisation information unavailable in the pre-act regime, and no uniform Turing predictor computes the choices from causal histories alone. The formalism proves neither that human beings satisfy the model nor a separation of P from NP; it identifies a boundary at which prediction concerns the existence of a computable function rather than its time complexity.

Vision-Language Models (VLMs) enable autonomous GUI navigation, but agents still struggle to process and learn from dense, continuous visual histories. This bottleneck hinders both immediate error correction within a single episode (intra-trial) and experience distillation across multiple attempts (cross-trial). We trace these challenges to an empirical informational asymmetry in GUI navigation: while expected transitions can often be compressed into lightweight textual summaries, unexpected outcomes benefit from preserved screenshots as causal evidence for accurate diagnosis. Building on this insight, we propose AnchorGUI, a unified framework driven by the Cognitive State Anchor (CSA). The CSA acts as a per-step primitive that actively compares expected and observed transitions, converting passive multimodal trajectories into explicit prediction-error signals. These signals orchestrate a dual-scale learning mechanism via an asymmetric memory. For intra-trial correction, a sliding window selectively retains visual evidence for detected mismatches, providing immediate, visually-grounded feedback. For cross-trial distillation, this asymmetric memory focuses the computationally expensive credit assignment search space on likely failure steps. Experiments across four benchmarks validate the effectiveness of our approach. On AndroidWorld, AnchorGUI achieves a 57.3% success rate with a $2.4\times$ token reduction per step. Furthermore, cross-trial distillation reaches 69.2% success (+11.9% gain), significantly outperforming standard reflection methods while maintaining sub-linear context scaling.

Companion robots face everyday situations in which several feasible behaviors may be appropriate, yet different people prefer different responses. We introduce InterSocialBench, a benchmark of 210 domestic scenarios and 18 high-level behaviors, pairing judgments from 100 human participants with 23,520 responses from seven large language models under 16 personality conditions. Each human annotation preserves a preferred action alongside explicitly appropriate and inappropriate candidates. A structured construction pipeline covers behavioral alternatives, competing situational cues, and relevant history and future tasks. Evaluation distinguishes preferred-choice agreement from explicit rejection, using scenario-grouped splits for trainable predictors. Simple frequency and persona-voting baselines illustrate these objectives. Across the tested prompts, model and human behavior distributions differ, and the diversity gap remains after matching response counts: humans exhibit 4.68 distinct choices per scenario, compared with 2.06–3.46 for the models. Human scenario-level plurality agreement is 51.5%, describing disagreement rather than a universal prediction ceiling. InterSocialBench supports evaluating social behavior selection without replacing individual judgments with a single consensus label.

Airway stenosis can cause severe respiratory complications, yet its detection often relies on specialized examinations and medical imaging. This study explores the potential of acoustic AI for rapid and accessible airway stenosis detection using readily acquired patient voice recordings. We systematically investigate whether acoustic foundation models (AFMs) can extract acoustic representations associated with airway stenosis-related speech patterns. Experiments are conducted on a cohort of 748 participants from the Bridge2AI-Voice dataset, 134 with airway stenosis and 614 without. The best-performing model achieves an AUROC of 0.952 and an accuracy of 0.924 (means over five-fold cross-validation), highlighting the potential of AFMs to transfer beyond general-purpose speech applications to clinical diagnostic tasks. Further analysis reveals that the model primarily relies on connected-speech recordings rather than isolated acoustic tasks, such as sustained phonation and breathing. Overall, these results suggest that voice-based acoustic AI could complement existing diagnostic workflows by enabling rapid, low-burden, and widely accessible screening for airway stenosis.

Combining aerial thrust with spring-loaded hopping makes monopedal quadcopters promising for locomotion over complex terrain, but heuristic proportional-integral-derivative (PID) tuning limits coordination between active thrust and passive contact dynamics. We present a direct estimated-state-to-motor Proximal Policy Optimization (PPO) policy that commands four motors without an explicit hopping state machine or low-level attitude PID. Its reward combines Energy-Manifold Shaping for mass-normalized vertical-energy tracking and apex-state anchoring with Efficiency Shaping, which uses a history-aware power estimator to penalize general power use, impose an additional airborne-power cost, and penalize airborne near-stationarity. In representative hardware runs, the PPO-based control stack reduced cycle-averaged measured electrical power by 30.7% and mean total normalized thrust by 49.8% relative to the tuned PID-based control stack, while retaining repeatable commanded-height hopping and more concentrated landings. These observations are consistent with improved use of passive dynamics and reduced measured electrical demand.

The (binary) differential-linear connectivity table (DLCT) measures the dependence between an input difference and a linear mask applied to the corresponding output difference. For vectorial Boolean functions, each DLCT entry is one half of an additive autocorrelation value. We extend this relation to functions over finite fields of arbitrary prime characteristic by introducing a level-resolved p-ary differential-linear profile. Its entries are the centered numbers of inputs for which a derivative component has each prescribed trace value in $\mathbb{F}_p$. The discrete Fourier transform of this profile is the family of additive autocorrelations obtained by multiplying the output mask by the nonzero elements of $\mathbb{F}_p$; when p=2, the usual binary identity is recovered. For a fixed input difference, we show that the profiles over all nonzero output masks determine the corresponding DDT row exactly, and we give an explicit inversion formula. We establish a second-moment identity: the total profile energy in one derivative direction is a constant multiple of the squared Euclidean distance between that DDT row and the balanced row. Thus this energy is determined by the full row differential spectrum, not by differential uniformity alone. It follows that all profiles in a direction vanish exactly when the derivative is balanced; for square maps in odd characteristic, this gives a characterization of planarity. As concrete odd-characteristic examples, we determine the complete profile of the monomial $x^{p^k+1}$ and derive an exact Kloosterman-sum formula for the inverse monomial. Finally, we determine the behavior of the profiles under equivalence. EA-equivalence reindexes the input and output masks and translates the trace level, whereas a general CCZ equivalence may mix several derivative directions. Nevertheless, for square maps the global nontrivial profile energy is CCZ-invariant.

Generative thermodynamic computers turn thermal noise into structured data through Langevin dynamics. We train these systems with a local update at each integration step. The reverse-path Onsager-Machlup objective yields a coupling gradient that is a symmetric sum of local residual-state correlations. We apply this gradient immediately rather than accumulating it over a full trajectory. In digital simulations using MNIST prototypes, online and trajectory-batch training reach similar validation losses on fixed noising paths. Models trained online release less heat on average in all five independently seeded pairs, with both models' parameters held fixed during sampling. Auxiliary classifier and nearest-prototype measures change modestly, while pairwise diversity decreases. The response to noise depends strongly on where the errors enter: independent zero-mean errors in the formed updates produce little heat change over a finite range of noise amplitudes, whereas residual offset and temporal correlation have much larger effects. Storing trained couplings requires substantially less precision than resolving deterministic updates during training. Together, these results establish a local online training method and show how update timing, noise structure, and precision affect generative thermodynamic computing.

An artificial organ carries failure consequences on the scale of an aircraft or a reactor, but the software driving it is rarely held to the same standard. Teams building them rely on testing, which only reaches the failure modes someone thought of in advance. In a pump or controller that runs inside a patient for months, the dangerous cases are the ones nobody anticipated. Formal verification closes that gap. Applied to the device's software, it proves the code meets its specification for every execution that specification allows, and where a proof fails, it returns the exact input sequence that breaks it. The same methods already protect rail, aviation, and nuclear control systems, and they extend the IEC 62304 lifecycle that a manufacturer already follows rather than replacing it. In this paper, we explore how to apply formal verification to artificial organs, stage by stage, and what each technique actually guarantees about the device.

End-to-end Supervised Graph Prediction (SGP) requires a permutation-invariant loss to compare predicted and target graphs with arbitrary node orderings. Such losses typically involve a costly graph-matching problem. We first study three Optimal Transport relaxations of this problem and show, theoretically and empirically, that the Gromov-Wasserstein (GW) objective is the most suitable for SGP. Then, to avoid solving the resulting inner optimization for every training example, we propose to amortize the graph matching (node alignment) problem. For each training sample, the loss function leverages a transport plan provided by a parametric matcher based on the differentiable Sinkhorn algorithm applied on empirical node distributions. The graph prediction module and the matcher are jointly learned. We showcase the efficiency of this approach on toy and real world SGP problems of increasing complexity including a novel Mass-spectra to Scaffold task that we introduce.

Online handwriting captured on digitizing tablets yields hundreds of kinematic, temporal, spatial, and dynamic features. These features are rarely organized into interpretable, reusable constructs, and, although standardized tests norm the handwritten product, no openly available instrument places a child's handwriting process relative to a verified-typical reference. We introduce and validate an open measurement instrument for profiling handwriting-process deviations in children with developmental dysgraphia from sentence-level online handwriting. The instrument comprises (i)~a literature-only vocabulary, fixed before any cohort analysis, organizing 136 online-handwriting features into 12 handwriting-process domains, and (ii)~an age- and sex-adjusted normative reference framework, fit on verified-typical children only, that turns those features into a 12-axis deviation profile with per-child bootstrap uncertainty. We establish its measurement properties on the DiaGraMo cohort (N=257 Czech children aged 8–12; 110 verified-typical, 147 with dysgraphia): the vocabulary is structurally coherent, and the reference is calibrated, parsimonious, and leakage-free. On the same cohort, five of twelve domains separate the groups at BH~$q<.05$ (Cliff's~$\delta$ +0.19 to +0.51), on spatial, temporal, and pen-orientation processes, and 96.5% of participant~$\times$~domain scores have a bootstrap CI narrower than one z-unit. No classifier is trained: the instrument reports uncertainty-aware deviation scores rather than a diagnostic label, and its outlier rate is not a diagnostic rate. The vocabulary, the analysis code, and an open reference implementation that scores new participants are all openly released, giving handwriting and dysgraphia researchers a reusable, validated instrument for situating individual children against a normative reference.

The use of Environmental, Social, and Governance (ESG) data is fundamental for modern corporate accountability, sustainability reporting, and financial decision-making. Embedding models have emerged as a powerful approach for transforming unstructured ESG text into numerical representations suitable for downstream natural language processing (NLP) tasks. However, their effectiveness in these ESG-specific tasks has not been systematically studied. In this paper, we construct a benchmark dataset specifically tailored to the ESG domain. We benchmark fourteen models, both open-source and closed-source embedding models, comparing their performance with respect to retrieval, and Retrieval-Augmented Generation (RAG). The results demonstrate performance variations across different models, with Qwen3-based models achieving the highest overall performance. This study provides practical insights into which models are better suited for ESG RAG tasks.

Tokenization determines both the physiological content presented to an ECG Transformer and the sequence over which attention operates. We compare eight tokenization strategies across Transformer, Informer, Reformer, and FEDformer on the nine-label CPSC2018 classification task. The input projection and principal backbone capacity are controlled to isolate the effect of token construction. Median-beat and HeartLang tokenization achieve mean macro-AUCs of 0.893 and 0.889 across the four backbones, compared with 0.822 and 0.824 for point-wise and patch-wise tokenization. Pooling the two physiology-aware representations yields an 8.2% relative improvement in macro-AUC. They also reduce mean sequence length from 1,250 to 158 tokens and mean peak training memory from 5.21 to 0.27 GB. The results show that aligning tokens with ECG morphology can improve both predictive performance and memory efficiency without increasing backbone capacity. The source code is available on https://github.com/LeeJarvis996/ecg_tokenizer.

Operator learning for parametric partial differential equations (PDEs) typically builds global models over prescribed domains, requiring cross-condition data or costly physics-constrained training. Here we show that the output Jacobian of a neural solution model trained at one condition defines a reusable response space for cross-condition solution variations. We introduce Linearized Subspace Transfer (LST) to exploit this space and recover target solutions by minimizing the target PDE-system residual over response-space coordinates. Because any single response space has finite coverage, Active Transfer Modeling (ATM) uses post-transfer residuals as coverage indicators to selectively acquire response spaces from additional single-condition models. Across six systems, single-condition response spaces supported cross-condition transfer, with enrichment improving accuracy when added spaces expanded representation capacity. Relative to evaluated physics-informed operator baselines, ATM reduced error and offline construction cost, with orders-of-magnitude accuracy gains in representative cases and millisecond-to-second target adaptation. These results establish neural solvers as reusable local parametric models.

Using food package images to monitor sodium and salt content against South Africa's R214 sodium limits is challenging when screening decisions require product identity, nutrition facts panel evidence, reporting basis, and category-specific thresholds. This study presents a conservative image-based workflow that combines region detection, optical character recognition (OCR), product identity and sodium evidence extraction, R214 category assignment, deterministic threshold comparison, and independent vision language model comparison. The evaluation used 442 packaged food products and 3 929 full package images from a real-world South African food packaging dataset. A YOLO26s small detector generated 4 195 region crops, and strict post-processing produced one sodium evidence row per product. The integrated workflow produced 290 OUTSIDE R214 SCOPE, 139 REVIEW, seven SCREEN-PASS, and six SCREEN-FAIL outcomes. The independent Qwen2.5-VL 7B vision language model workflow produced 387 OUTSIDE R214 SCOPE, 31 REVIEW, twenty SCREEN-PASS, and four SCREEN-FAIL outcomes. The workflows agreed on exact R214 category assignment for 415 of 442 products (93.9%) and on whether the assigned category was within R214 scope for 416 of 442 products (94.1%). Final screening outcome agreement was 307 out of 442 products, or 69.5%. Manual verification on 60 products showed lower strict outcome agreement than regulated status agreement, while all manual INSUFFICIENT DATA cases were kept out of SCREEN-PASS and SCREEN-FAIL by both automated workflows. The findings show that conservative image-based screening can organise package evidence, identify clear cases, and assign uncertain cases to REVIEW rather than forcing SCREEN-PASS or SCREEN-FAIL decisions.

Human-computer interaction increasingly studies and promotes justice. Often, this work does not treat justice through a pluralistic lens Justice, however, is not one thing: political philosophy offers several well-established, mutually incompatible conceptions of it. To navigate these meanings, we draw on a map of justice, a philosophical framework identifying six distinct conceptions of justice situated throughout (Western) philosophical thought. We apply this framework to four cases that illustrate different approaches to social justice: the COMPAS system, food democracy, gentrification, and robotics. Our analysis shows that each case already reflects a position on the map, but more importantly, one might adapt the case to other conceptions as well, leading to very different systems and analyses. We discuss how HCI might engage more carefully with such contested concepts, and warn that without pluralistic understanding, academic discourse risks a spiral of silence.

The use of open-source software (OSS) in satellite flight systems is increasing as missions adopt reusable frameworks, shared libraries, and community-maintained components. While this accelerates development, it also introduces software-security risks into systems where patching is costly and failures may affect mission operations. This paper presents an empirical security study of OSS used in onboard satellite systems. We analyze 126 public repositories using a pipeline that combines software bill of materials generation, software composition analysis, static application security testing, infrastructure-as-code analysis, and secret scanning. After rule-based cleaning, onboard-scope filtering, and fingerprint-based deduplication, the pipeline produced a final dataset of 2,827 findings. The results show that security findings are widespread but unevenly distributed. Medium-severity findings account for 49% of the dataset, and 72% are classified as medium severity or higher. A Common Weakness Enumeration (CWE)-based taxonomy assigns all findings to eight weakness families. Memory Safety and Code Quality dominate the dataset, followed by Input Validation and Injection. Most findings occur in project-developed code, accounting for 81.4% of the dataset, while external dependency code remains a relevant source of findings. While these findings do not establish mission-specific exploitability, they provide an empirical characterization of recurring security patterns across the open-source onboard satellite software ecosystem, helping quantify their prevalence and prioritize areas that warrant the greatest security attention.

LCD codes over finite commutative chain rings are known to be free. However, if the ring is not indecomposable, there always exists an LCD ideal that is projective but not free. We prove that all two-sided LCD codes are projective. We generalise the characterisation of finite commutative Frobenius rings by means of the size condition of a code and its dual to the noncommutative setting. This result is applied to prove that group codes are LCD if and only if they are generated by a central idempotent.

AI agents are provisioned the same as employee-owned hosts in many enterprise settings with a static credential set fixed at deployment which includes all permissions the employee role might ever need. Role-based access control made this compromise for human principals because scoping access per task was infeasible. For AI agents, the compromise leaves every credential standing exposed whether or not the current task uses them. These permissions can later be utilised by a compromised or misaligned agent. Prior work (Noyan, 2026) defined this as the task-context mismatch, and proposed a three-source permission architecture which includes role-based permission ceilings, a task permission classifier and policy-based prohibitions, together eliminating the exposure preemptively. The work released a 600-prompt labelled dataset to evaluate it. This paper presents that evaluation end to end by implementing the security gate; a fine-tuned RoBERTa-large encoder which matched few-shot trained Claude Haiku 4.5 on classification quality (macro-F1 0.881 against 0.886, precision 0.897 against 0.842, severity-weighted residual risk 0.63 against 1.12). The results show the trusted component does not need to scale with the agent it supervises, and the scalable-oversight margin for this control method is wide. We also propose an attack-surface elimination metric which shows the role ceiling alone closes 27.9% of the severity-weighted surface and adding the task classifier closes 84.4%. The gap displays security advantages of task-granular access control over role-granular, and AI agents are the first principal type for which the task-granular access control is enforceable because their tasks arrive as machine-readable text. The research establishes task-based access control as a measured, potentially deployable mechanism for reducing attack surface in agentic deployments.

Open-vocabulary camouflaged object segmentation (OVCOS) aims to segment unseen camouflaged objects under text guidance. We observe that SAM3 still suffers from a pronounced semantic gap between global textual semantics and fine-grained pixel-level visual cues in OVCOS. Meanwhile, fully fine-tuning the text encoder introduces heavy parameter overhead and risks overfitting to training categories, which compromises open-vocabulary representation flexibility. To address these issues, we propose ViCo-SAM3, a Vision-Conditioned alignment framework designed for OVCOS. Specifically, we introduce vision-conditioned (ViCo) module, which dynamically modulates text embeddings with global visual context, enabling textual representations to adapt to the current image content and thereby effectively bridging the semantic gap between vision and text. Building on this, we further design a vision-conditioned cross-modal binding (ViCoBind) module to enhance cross-modal interaction and semantic alignment between visual and textual representations. Without bells and whistles, ViCo-SAM3 achieves state-of-the-art performance on the OVCamo benchmark and demonstrates strong generalization.

Long-video understanding remains challenging for multimodal large language models (MLLMs) because densely encoding long frame sequences is computationally expensive, while uniform sampling under a limited visual budget can miss sparse yet decisive evidence. Recent training-free keyframe selection methods have enabled more efficient inference and yielded promising performance gains. However, many existing methods score frames largely in isolation without explicitly considering how each candidate complements the currently selected subset, potentially resulting in redundant selections and incomplete evidence coverage. To address this limitation, we propose MarKey, a training-free framework that formulates keyframe selection as subset-aware greedy optimization. At each iteration, MarKey scores each candidate using a tractable surrogate that jointly accounts for query relevance, marginal coverage gain, and context-dependent redundancy, and selects the frame with the highest utility. To make this iterative subset-aware evaluation efficient, MarKey uses a compact set of representative anchors to approximate full-video coverage and a bounded window of previously selected frames to limit context-dependent comparisons. Experiments on six benchmarks spanning holistic video understanding, human-centric video understanding, and open-ended video understanding demonstrate that MarKey consistently outperforms existing methods. Further analyses show robust gains across different MLLM backbones, model scales, and frame budgets.

While there has been much discussion of whether AI systems could function as moral agents or acquire sentience, there has been very little discussion of whether AI systems could have free will. I sketch a framework for thinking about this question, inspired by Daniel Dennett's work. I argue that, to determine whether an AI system has free will, we should not look for some mysterious property, expect its underlying algorithms to be indeterministic, or ask whether the system is unpredictable. Rather, we should simply ask whether we have good explanatory reasons to view the system as an intentional agent, with the capacity for choice between alternative possibilities and control over the resulting actions. If the answer is "yes", then the system counts as having free will in a pragmatic and diagnostically useful sense.

Multi-teacher on-policy distillation (OPD) is becoming the standard way to integrate specialist capabilities into one model: train experts with RL, then distill them into the student on its own rollouts. Existing recipes assign supervision at the sequence level - each prompt goes to one domain teacher and every token receives the same weight - which implicitly assumes that a teacher is uniformly useful across a response. We find instead that useful teacher signal is sparse and heterogeneous along a reasoning trajectory, which raises a finer question: who should teach which token? Verifier-Gated Multi-Expert On-Policy Distillation (VG-OPD) answers it by verification: the counterfactual gain of an expert on a specific answer criterion licenses that expert to teach, its disagreement with the student localizes the supervision, and criterion importance sets its weight; the gated KL enters GRPO as an additive token-level advantage. Instantiated for scientific reasoning with RL-trained capability experts, VG-OPD attains the best overall performance on seven benchmarks for 4B and 8B students, ranking first on five at both scales, with the largest gains on knowledge-intensive scientific reasoning tasks. Further analysis shows that the gains come from localizing verified supervision rather than from adding teachers or distillation loss: misplacing the same supervision budget is the single most damaging change, and indiscriminate distillation drags RL below its own floor where gated distillation lifts it.

Vessel segmentation in X-ray coronary angiography (XCA) is a fundamental step for quantitative coronary analysis and subsequent assessment of coronary artery disease. However, accurate vessel segmentation remains challenging because of imaging noise, complex bifurcations, and the overlap of vessels and background structures, which can lead to disrupted vascular connectivity and missed small branches. In this work, we propose BSC-Net, a ResNet-U-Net-based framework tailored to improve small-vessel representation and repair vascular structural continuity. BSC-Net enhances small-vessel representation through targeted sampling and improves vascular structural continuity by integrating long-range contextual modeling and Edge-Informed Loss (EIL). BSC-Net was validated on two public XCA datasets, demonstrating state-of-the-art (SOTA) performance in coronary vessel segmentation with Dice and IoU scores of 77.8%/90.6% and 64.5%/83.0%, respectively. Furthermore, based on the obtained vessel segmentation, we performed automated quantitative coronary analysis and derived clinically relevant morphological and hemodynamic parameters, including stenosis ratio, time-to-peak, and relative propagation velocity. These results demonstrate that BSC-Net produces accurate vessel segmentation results with preserved vascular continuity for quantitative coronary assessment, enabling reliable downstream analysis and clinical evaluation of coronary artery disease.