Biases in large language models (LLMs) often manifest as systematic distortions in associations between demographic attributes and professional or social roles, reinforcing harmful stereotypes across gender, ethnicity, and geography. This position paper advocates for addressing demographic and gender biases in LLMs through a dual-pronged methodology, integrating category-theoretic transformations and retrieval-augmented generation (RAG). Category theory provides a rigorous, structure-preserving mathematical framework that maps biased semantic domains to unbiased canonical forms via functors, ensuring bias elimination while preserving semantic integrity. Complementing this, RAG dynamically injects diverse, up-to-date external knowledge during inference, directly countering ingrained biases within model parameters. By combining structural debiasing through functor-based mappings and contextual grounding via RAG, we outline a comprehensive framework capable of delivering equitable and fair model outputs. Our synthesis of the current literature validates the efficacy of each approach individually, while addressing potential critiques demonstrates the robustness of this integrated strategy. Ensuring fairness in LLMs, therefore, demands both the mathematical rigor of category-theoretic transformations and the adaptability of retrieval augmentation.

Many errors in student essays can be explained by influence from the native language (L1). L1 interference refers to errors influenced by a speaker's first language, such as using stadion instead of stadium, reflecting lexical transliteration from Russian. In this work, we address the task of detecting such errors in English essays written by Russian-speaking learners. We introduce RILEC, a large-scale dataset of over 18,000 sentences, combining expert-annotated data from REALEC with synthetic examples generated through rule-based and neural augmentation. We propose a framework for generating L1-motivated errors using generative language models optimized with PPO, prompt-based control, and rule-based patterns. Models fine-tuned on RILEC achieve strong performance, particularly on word-level interference types such as transliteration and tense semantics. We find that the proposed augmentation pipeline leads to a significant performance improvement, making it a potentially valuable tool for learners and teachers to more effectively identify and address such errors.

Neural scaling laws describe how model performance improves as a power law with size, but existing work focuses on models above 100M parameters. The sub-20M regime -- where TinyML and edge AI operate -- remains unexamined. We train 90 models (22K--19.8M parameters) across two architectures (plain ConvNet, MobileNetV2) on CIFAR-100, varying width while holding depth and training fixed. Both follow approximate power laws in error rate: $α= 0.156 \pm 0.002$ (ScaleCNN) and $α= 0.106 \pm 0.001$ (MobileNetV2) across five seeds. Since prior work fit cross-entropy loss rather than error rate, direct exponent comparison is approximate; with that caveat, these are 1.4--2x steeper than $α\approx 0.076$ for large language models. The power law does not hold uniformly: local exponents decay with scale, and MobileNetV2 saturates at 19.8M parameters ($α_{\mathrm{local}} = 0.006$). Error structure also changes. Jaccard overlap between error sets of the smallest and largest ScaleCNN is only 0.35 (25 seed pairs, $\pm 0.004$) -- compression changes which inputs are misclassified, not merely how many. Small models concentrate capacity on easy classes (Gini: 0.26 at 22K vs. 0.09 at 4.7M) while abandoning the hardest (bottom-5 accuracy: 10% vs. 53%). Counter to expectation, the smallest models are best calibrated (ECE = 0.013 vs. peak 0.110 at mid-size). Aggregate accuracy is therefore misleading for edge deployment; validation must happen at the target model size.

Long-horizon wildfire risk forecasting requires generating probabilistic spatial fields under sparse event supervision while maintaining computational efficiency across multiple prediction horizons. Extending diffusion models to multi-step forecasting typically repeats the denoising process independently for each horizon, leading to redundant computation. We introduce N-Tree Diffusion (NT-Diffusion), a hierarchical diffusion model designed for long-horizon wildfire risk forecasting. Fire occurrences are represented as continuous Fire Risk Maps (FRMs), which provide a smoothed spatial risk field suitable for probabilistic modeling. Instead of running separate diffusion trajectories for each predicted timestamp, NT-Diffusion shares early denoising stages and branches at later levels, allowing horizon-specific refinement while reducing redundant sampling. We evaluate the proposed framework on a newly collected real-world wildfire dataset constructed for long-horizon probabilistic prediction. Results indicate that NT-Diffusion achieves consistent accuracy improvements and reduced inference cost compared to baseline forecasting approaches.

Designing environments that maximize the rate of emergent behavior development in AI agents remains an open problem. We present the first systematic study of stress-performance relationships in large language model (LLM) multi-agent systems, drawing an explicit parallel to the Yerkes-Dodson law from cognitive psychology. Using a grid-world survival arena, we conduct 22 experiments across four phases, varying environmental pressure through resource scarcity (upkeep cost) and reproductive competition (sexual selection). Our key finding is that cooperative behavior follows an inverted-U curve: trade interactions peak at 29 under medium pressure (upkeep=5), while both low and extreme pressure produce 8--12 trades. Under extreme pressure, behavioral repertoire collapses to movement-only within 5--12 turns. We further show that sexual selection -- a softer pressure mechanism where all agents survive but not all reproduce -- eliminates inter-agent aggression entirely and produces communicative behavior absent under survival pressure. These results suggest that environmental pressure calibration is a viable curriculum design strategy for LLM agent development, analogous to the inverted-U relationship between arousal and performance in biological systems.

We propose DistGP: a multi-robot learning method for collaborative learning of a global function using only local experience and computation. We utilise a sparse Gaussian process (GP) model with a factorisation that mirrors the multi-robot structure of the task, and admits distributed training via Gaussian belief propagation (GBP). Our loopy model outperforms Tree-Structured GPs \cite{bui2014tree} and can be trained online and in settings with dynamic connectivity. We show that such distributed, asynchronous training can reach the same performance as a centralised, batch-trained model, albeit with slower convergence. Last, we compare to DiNNO \cite{yu2022dinno}, a distributed neural network (NN) optimiser, and find DistGP achieves superior accuracy, is more robust to sparse communication and is better able to learn continually.

Clinical machine learning models are increasingly trained using large scale, multimodal foundation paradigms, yet deployment environments often differ systematically from the data generating settings used during training. Such shifts arise from heterogeneous measurement policies, documentation practices, and institutional workflows, leading to representation entanglement between physiologic signal and practice specific artifacts. In this work, we propose a practice invariant representation learning framework for multimodal clinical prediction. We model clinical observations as arising from latent physiologic factors and environment dependent processes, and introduce an objective that jointly optimizes predictive performance while suppressing environment predictive information in the learned embedding. Concretely, we combine supervised risk minimization with adversarial environment regularization and invariant risk penalties across hospitals. Across multiple longitudinal EHR prediction tasks and cross institution evaluations, our method improves out of distribution AUROC by up to 2 to 3 points relative to masked pretraining and standard supervised baselines, while maintaining in distribution performance and improving calibration. These results demonstrate that explicitly accounting for systematic distribution shift during representation learning yields more robust and transferable clinical models, highlighting the importance of structural invariance alongside architectural scale in healthcare AI.

Noisy training data can significantly degrade the performance of language-model-based classifiers, particularly in non-topical classification tasks. In this study we designed a methodological framework to assess the impact of denoising. More specifically, we explored a range of denoising strategies for sentence-level difficulty detection, using training data derived from document-level difficulty annotations obtained through noisy crowdsourcing. Beyond monolingual settings, we also address cross-lingual transfer, where a multilingual language model is trained in one language and tested in another. We evaluate several noise reduction techniques, including Gaussian Mixture Models (GMM), Co-Teaching, Noise Transition Matrices, and Label Smoothing. Our results indicate that while BERT-based models exhibit inherent robustness to noise, incorporating explicit noise detection can further enhance performance. For our smaller dataset, GMM-based noise filtering proves particularly effective in improving prediction quality by raising the Area-Under-the-Curve score from 0.52 to 0.92, or to 0.93 when de-noising methods are combined. However, for our larger dataset, the intrinsic regularisation of pre-trained language models provides a strong baseline, with denoising methods yielding only marginal gains (from 0.92 to 0.94, while a combination of two denoising methods made no contribution). Nonetheless, removing noisy sentences (about 20\% of the dataset) helps in producing a cleaner corpus with fewer infelicities. As a result we have released the largest multilingual corpus for sentence difficulty prediction: see https://github.com/Nouran-Khallaf/denoising-difficulty

Generative zero-shot learning (ZSL) synthesizes features for unseen classes, leveraging semantic conditions to transfer knowledge from seen classes. However, it also introduces two intrinsic challenges: (1) class-level attributes fails to capture instance-specific visual appearances due to substantial intra-class variability, thus causing the class-instance gap; (2) the substantial mismatch between semantic and visual feature distributions, manifested in inter-class correlations, gives rise to the semantic-visual domain gap. To address these challenges, we propose an Attribute Distribution Modeling and Semantic-Visual Alignment (ADiVA) approach, jointly modeling attribute distributions and performing explicit semantic-visual alignment. Specifically, our ADiVA consists of two modules: an Attribute Distribution Modeling (ADM) module that learns a transferable attribute distribution for each class and samples instance-level attributes for unseen classes, and a Visual-Guided Alignment (VGA) module that refines semantic representations to better reflect visual structures. Experiments on three widely used benchmark datasets demonstrate that ADiVA significantly outperforms state-of-the-art methods (e.g., achieving gains of 4.7% and 6.1% on AWA2 and SUN, respectively). Moreover, our approach can serve as a plugin to enhance existing generative ZSL methods.

Multi-object tracking (MOT) involves analyzing object trajectories and counting the number of objects in video sequences. However, 2D MOT faces challenges due to positional cost confusion arising from partial occlusion. To address this issue, we present the novel Occlusion-Aware SORT (OA-SORT) framework, a plug-and-play and training-free framework that includes the Occlusion-Aware Module (OAM), the Occlusion-Aware Offset (OAO), and the Bias-Aware Momentum (BAM). Specifically, OAM analyzes the occlusion status of objects, where a Gaussian Map (GM) is introduced to reduce background influence. In contrast, OAO and BAM leverage the OAM-described occlusion status to mitigate cost confusion and suppress estimation instability. Comprehensive evaluations on the DanceTrack, SportsMOT, and MOT17 datasets demonstrate the importance of occlusion handling in MOT. On the DanceTrack test set, OA-SORT achieves 63.1% and 64.2% in HOTA and IDF1, respectively. Furthermore, integrating the Occlusion-Aware framework into the four additional trackers improves HOTA and IDF1 by an average of 2.08% and 3.05%, demonstrating the reusability of the occlusion awareness.

Sketch colorization is a critical task for automating and assisting in the creation of animations and digital illustrations. Previous research identified the primary difficulty as the distribution shift between semantically aligned training data and highly diverse test data, and focused on mitigating the artifacts caused by the distribution shift instead of fundamentally resolving the problem. In this paper, we present a framework that directly minimizes the distribution shift, thereby achieving superior quality, resolution, and controllability of colorization. We propose a dual-branch framework to explicitly model the data distributions of the training process and inference process with a semantic-aligned branch and a semantic-misaligned branch, respectively. A Gram Regularization Loss is applied across the feature maps of both branches, effectively enforcing cross-domain distribution coherence and stability. Furthermore, we adopt an anime-specific Tagger Network to extract fine-grained attributions from reference images and modulate SDXL's conditional encoders to ensure precise control, and a plugin module to enhance texture transfer. Quantitative and qualitative comparisons, alongside user studies, confirm that our method effectively overcomes the distribution shift challenge, establishing State-of-the-Art performance across both quality and controllability metrics. Ablation study reveals the influence of each component.

Open-vocabulary object detection (OVOD) enables novel category detection via vision-language alignment, but massive model sizes hinder deployment on resource-constrained devices. While quantization offers practical compression, we reveal that naive extreme low-bit (e.g., 4-bit) quantization severely degrades fine-grained vision-language alignment and distorts inter-region relational structures. To address this, we propose curriculum relational quantization-aware training (CR-QAT), an integrated framework combining stage-by-stage optimization with relational knowledge distillation. Within CR-QAT, curriculum QAT (CQAT) mitigates error accumulation by partitioning the model for progressive quantization, ensuring stable optimization via error isolation. Concurrently, text-centric relational KD (TRKD) is applied to task-relevant modules. By constructing text-anchored pairwise similarity matrices, TRKD comprehensively transfers the teacher's multi-dimensional relational knowledge. Experiments on LVIS and COCO zero-shot benchmarks demonstrate that CR-QAT consistently outperforms existing QAT baselines under aggressive low-bit settings, achieving relative AP improvements of up to 38.9% and 40.9%, respectively.

Personal AI agents face a fundamental deployment paradox on mobile: persistent background execution drains battery and violates platform sandboxing policies, yet purely reactive agents miss time-sensitive obligations until the user remembers to ask. We present Jagarin, a three-layer architecture that resolves this paradox through structured hibernation and demand-driven wake. The first layer, DAWN (Duty-Aware Wake Network), is an on-device heuristic engine that computes a composite urgency score from four signals: duty-typed optimal action windows, user behavioral engagement prediction, opportunity cost of inaction, and cross-duty batch resonance. It uses adaptive per-user thresholds to decide when a sleeping agent should nudge or escalate. The second layer, ARIA (Agent Relay Identity Architecture), is a commercial email identity proxy that routes the full commercial inbox -- obligations, promotional offers, loyalty rewards, and platform updates -- to appropriate DAWN handlers by message category, eliminating cold-start and removing manual data entry. The third layer, ACE (Agent-Centric Exchange), is a protocol framework for direct machine-readable communication from institutions to personal agents, replacing human-targeted email as the canonical channel. Together, these three layers form a complete stack from institutional signal to on-device action, without persistent cloud state, continuous background execution, or privacy compromise. A working Flutter prototype is demonstrated on Android, combining all three layers with an ephemeral cloud agent invoked only on user-initiated escalation.

Tracking any point (TAP) is a fundamental yet challenging task in computer vision, requiring high precision and long-term motion reasoning. Recent attempts to combine RGB frames and event streams have shown promise, yet they typically rely on synchronous or non-adaptive fusion, leading to temporal misalignment and severe degradation when one modality fails. We introduce TAPFormer, a transformer-based framework that performs asynchronous temporal-consistent fusion of frames and events for robust and high-frequency arbitrary point tracking. Our key innovation is a Transient Asynchronous Fusion (TAF) mechanism, which explicitly models the temporal evolution between discrete frames through continuous event updates, bridging the gap between low-rate frames and high-rate events. In addition, a Cross-modal Locally Weighted Fusion (CLWF) module adaptively adjusts spatial attention according to modality reliability, yielding stable and discriminative features even under blur or low light. To evaluate our approach under realistic conditions, we construct a novel real-world frame-event TAP dataset under diverse illumination and motion conditions. Our method outperforms existing point trackers, achieving a 28.2% improvement in average pixel error within threshold. Moreover, on standard point tracking benchmarks, our tracker consistently achieves the best performance. Project website: tapformer.github.io

Standard Retrieval-Augmented Generation (RAG) architectures fail in high-stakes financial domains due to two fundamental limitations: the inherent arithmetic incompetence of Large Language Models (LLMs) and the distributional semantic conflation of dense vector retrieval (e.g., mapping "Net Income" to "Net Sales" due to contextual proximity). In deterministic domains, a 99% accuracy rate yields 0% operational trust. To achieve zero-hallucination financial reasoning, we introduce the Verifiable Numerical Reasoning Agent (VeNRA). VeNRA shifts the RAG paradigm from retrieving probabilistic text to retrieving deterministic variables via a strictly typed Universal Fact Ledger (UFL). We mathematically bound this ledger using a novel Double-Lock Grounding algorithm. Coupled with deterministic Python execution, this neuro-symbolic routing compresses systemic hallucination rates to a near-zero 1.2%. Recognising that upstream parsing anomalies inevitably occur, we introduce the VeNRA Sentinel: a 3-billion parameter SLM trained to forensically audit candidate using a single-token inference budget with optional post-hoc reasoning. To train the Sentinel, we steer away from traditional hallucination datasets in favour of Adversarial Simulation, programmatically sabotaging financial records to simulate Ecological Errors. The compact Sentinel consequently outperforms 70B+ frontier models in error detection. Through Loss Dilution phenomenon in Reverse-CoT training, we present a novel Micro-Chunking loss algorithm to stabilise gradients under extreme verdict penalisation, yielding a 28x latency speedup without sacrificing forensic rigor.

Conventionally, memory in end-to-end robotic learning involves inputting a sequence of past observations into the learned policy. However, in complex multi-stage real-world tasks, the robot's memory must represent past events at multiple levels of granularity: from long-term memory that captures abstracted semantic concepts (e.g., a robot cooking dinner should remember which stages of the recipe are already done) to short-term memory that captures recent events and compensates for occlusions (e.g., a robot remembering the object it wants to pick up once its arm occludes it). In this work, our main insight is that an effective memory architecture for long-horizon robotic control should combine multiple modalities to capture these different levels of abstraction. We introduce Multi-Scale Embodied Memory (MEM), an approach for mixed-modal long-horizon memory in robot policies. MEM combines video-based short-horizon memory, compressed via a video encoder, with text-based long-horizon memory. Together, they enable robot policies to perform tasks that span up to fifteen minutes, like cleaning up a kitchen, or preparing a grilled cheese sandwich. Additionally, we find that memory enables MEM policies to intelligently adapt manipulation strategies in-context.

As strong general reasoners, large language models (LLMs) encounter diverse domains and tasks, where the ability to adapt and self-improve at test time is valuable. We introduce MASS, a meta-learning framework that enables LLMs to self-adapt by generating problem-specific synthetic training data and performing targeted self-updates optimized for downstream performance at inference time. We train this behavior end-to-end via bilevel optimization: an inner loop adapts on self-generated examples while an outer loop meta-learns data-attribution signals and rewards post-update task performance. The synthetic data is optimized with scalable meta-gradients, backpropagating the downstream loss through the inner updates to reward useful generations. Experiments on mathematical reasoning show that MASS learns to synthesize per-instance curricula that yield effective, data-efficient test-time adaptation.

What determines whether a molecular property prediction model organizes its representations so that geometric and compositional information can be cleanly separated? We introduce Compositional Probe Decomposition (CPD), which linearly projects out composition signal and measures how much geometric information remains accessible to a Ridge probe. We validate CPD with four independent checks, including a structural isomer benchmark where compositional projections score at chance while geometric residuals reach 94.6\% pairwise classification accuracy. Across ten models from five architectural families on QM9, we find a \emph{linear accessibility gradient}: models differ by $6.6\times$ in geometric information accessible after composition removal ($R^2_{\mathrm{geom}}$ from 0.081 to 0.533 for HOMO-LUMO gap). Three factors explain this gradient. Task alignment dominates: models trained on HOMO-LUMO gap ($R^2_{\mathrm{geom}}$ 0.44--0.53) outscore energy-trained models by $\sim$0.25 $R^2$ regardless of architecture. Within-architecture ablations on two independent architectures confirm this: PaiNN drops from 0.53 to 0.31 when retrained on energy, and MACE drops from 0.44 to 0.08. Data diversity partially compensates for misaligned objectives, with MACE pretrained on MPTraj (0.36) outperforming QM9-only energy models. Inside MACE's representations, information routes by symmetry type: $L{=}1$ (vector) channels preferentially encode dipole moment ($R^2 = 0.59$ vs.\ 0.38 in $L{=}0$), while $L{=}0$ (scalar) channels encode HOMO-LUMO gap ($R^2 = 0.76$ vs.\ 0.34 in $L{=}1$). This pattern is absent in ViSNet. We also show that nonlinear probes produce misleading results on residualized representations, recovering $R^2 = 0.68$--$0.95$ on a purely compositional target, and recommend linear probes for this setting.

While artificial neural networks excel in unsupervised learning of non-sparse structure, classical statistical regression techniques offer better interpretability, in particular when sparseness is enforced by $\ell_1$ regularization, enabling identification of which factors drive observed dynamics. We investigate how these two types of approaches can be optimally combined, exemplarily considering two-photon calcium imaging data where sparse autoregressive dynamics are to be extracted. We propose embedding a vector autoregressive (VAR) model as an interpretable regression technique into a convolutional autoencoder, which provides dimension reduction for tractable temporal modeling. A skip connection separately addresses non-sparse static spatial information, selectively channeling sparse structure into the $\ell_1$-regularized VAR. $\ell_1$-estimation of regression parameters is enabled by differentiating through the piecewise linear solution path. This is contrasted with approaches where the autoencoder does not adapt to the VAR model. Having an embedded statistical model also enables a testing approach for comparing temporal sequences from the same observational unit. Additionally, contribution maps visualize which spatial regions drive the learned dynamics.

Image-text contrastive pretraining has become a dominant paradigm for visual representation learning, yet existing methods often yield representations that remain partially organized by modality. We propose ITO, a framework addressing this limitation through two synergistic mechanisms. Multimodal multiple alignment enriches supervision by mining diverse image-text correspondences, while a lightweight training-time multimodal fusion module enforces structured cross-modal interaction. Crucially, the fusion module is discarded at inference, preserving the efficiency of standard dual-encoder architectures. Extensive experiments show that ITO consistently outperforms strong baselines across classification, retrieval, and multimodal benchmarks. Our analysis reveals that while multiple alignment drives discriminative power, training-time fusion acts as a critical structural regularizer -- eliminating the modality gap and stabilizing training dynamics to prevent the early saturation often observed in aggressive contrastive learning.

Despite the success of Large Vision--Language Models (LVLMs), most existing architectures suffer from a representation bottleneck: they rely on static, instruction-agnostic vision encoders whose visual representations are utilized in an invariant manner across different textual tasks. This rigidity hinders fine-grained reasoning where task-specific visual cues are critical. To address this issue, we propose iGVLM, a general framework for instruction-guided visual modulation. iGVLM introduces a decoupled dual-branch architecture: a frozen representation branch that preserves task-agnostic visual representations learned during pre-training, and a dynamic conditioning branch that performs affine feature modulation via Adaptive Layer Normalization (AdaLN). This design enables a smooth transition from general-purpose perception to instruction-aware reasoning while maintaining the structural integrity and stability of pre-trained visual priors. Beyond standard benchmarks, we introduce MM4, a controlled diagnostic probe for quantifying logical consistency under multi-query, multi-instruction settings. Extensive results show that iGVLM consistently enhances instruction sensitivity across diverse language backbones, offering a plug-and-play paradigm for bridging passive perception and active reasoning.

Single-cell perturbation studies face dual heterogeneity bottlenecks: (i) semantic heterogeneity--identical biological concepts encoded under incompatible metadata schemas across datasets; and (ii) statistical heterogeneity--distribution shifts from biological variation demanding dataset-specific inductive biases. We propose HarmonyCell, an end-to-end agent framework resolving each challenge through a dedicated mechanism: an LLM-driven Semantic Unifier autonomously maps disparate metadata into a canonical interface without manual intervention; and an adaptive Monte Carlo Tree Search engine operates over a hierarchical action space to synthesize architectures with optimal statistical inductive biases for distribution shifts. Evaluated across diverse perturbation tasks under both semantic and distribution shifts, HarmonyCell achieves a 95% valid execution rate on heterogeneous input datasets (versus 0% for general agents) while matching or even exceeding expert-designed baselines in rigorous out-of-distribution evaluations. This dual-track orchestration enables scalable automatic virtual cell modeling without dataset-specific engineering.

Large Language Models (LLMs) are increasingly used for medical entity extraction, yet their confidence scores are often miscalibrated, limiting safe deployment in clinical settings. We present a conformal prediction framework that provides finite-sample coverage guarantees for LLM-based extraction across two clinical domains. First, we extract structured entities from 1,000 FDA drug labels across eight sections using GPT-4.1, verified via FactScore-based atomic statement evaluation (97.7\% accuracy over 128,906 entities). Second, we extract radiological entities from MIMIC-CXR reports using the RadGraph schema with GPT-4.1 and Llama-4-Maverick, evaluated against physician annotations (entity F1: 0.81 to 0.84). Our central finding is that miscalibration direction reverses across domains: on well-structured FDA labels, models are underconfident, requiring modest conformal thresholds ($τ\approx 0.06$), while on free-text radiology reports, models are overconfident, demanding strict thresholds ($τ$ up to 0.99). Despite this heterogeneity, conformal prediction achieves target coverage ($\geq 90\%$) in both settings with manageable rejection rates (9--13\%). These results demonstrate that calibration is not a global model property but depends on document structure, extraction category, and model architecture, motivating domain-specific conformal calibration for safe clinical deployment.

The growing computational and memory demands of the Key-Value (KV) cache significantly limit the ability of Large Language Models (LLMs). While KV merging has emerged as a promising solution, existing methods that rely on empirical observations of KV asymmetry and gradient-based Hessian approximations lack a theoretical foundation and incur suboptimal compression and inference overhead. To bridge these gaps, we establish a theoretical framework that characterizes this asymmetry through the spectral energy distribution of projection weights, demonstrating that concentrated spectra in Query/Key weights induce feature homogeneity, whereas dispersed spectra in Value weights preserve heterogeneity. Then, we introduce KVSlimmer, an efficient algorithm that captures exact Hessian information through a mathematically exact formulation, and derives a closed-form solution utilizing only forward-pass variables, resulting in a gradient-free approach that is both memory- and time-efficient. Extensive experiments across various models and benchmarks demonstrate that KVSlimmer consistently outperforms SOTA methods. For instance, on Llama3.1-8B-Instruct, it improves the LongBench average score by 0.92 while reducing memory costs and latency by 29% and 28%, respectively.Code is available at https://github.com/lianjunl13-sudo/KVSlimmer.

Production-ready human video generation requires digital actors to maintain strictly consistent full-body identities across dynamic shots, viewpoints and motions, a setting that remains challenging for existing methods. Prior methods often suffer from face-centric behavior that neglects body-level consistency, or produce copy-paste artifacts where subjects appear rigid due to pose locking. We present Actor-18M, a large-scale human video dataset designed to capture identity consistency under unconstrained viewpoints and environments. Actor-18M comprises 1.6M videos with 18M corresponding human images, covering both arbitrary views and canonical three-view representations. Leveraging Actor-18M, we propose WildActor, a framework for any-view conditioned human video generation. We introduce an Asymmetric Identity-Preserving Attention mechanism coupled with a Viewpoint-Adaptive Monte Carlo Sampling strategy that iteratively re-weights reference conditions by marginal utility for balanced manifold coverage. Evaluated on the proposed Actor-Bench, WildActor consistently preserves body identity under diverse shot compositions, large viewpoint transitions, and substantial motions, surpassing existing methods in these challenging settings.

While multimodal large language models offer a promising solution to the "black box" nature of health AI by generating interpretable reasoning traces, verifying the validity of these traces remains a critical challenge. Existing evaluation methods are either unscalable, relying on manual clinician review, or superficial, utilizing proxy metrics (e.g. QA) that fail to capture the semantic correctness of clinical logic. In this work, we introduce a reproducible framework for evaluating reasoning in ECG signals. We propose decomposing reasoning into two distinct, components: (i) Perception, the accurate identification of patterns within the raw signal, and (ii) Deduction, the logical application of domain knowledge to those patterns. To evaluate Perception, we employ an agentic framework that generates code to empirically verify the temporal structures described in the reasoning trace. To evaluate Deduction, we measure the alignment of the model's logic against a structured database of established clinical criteria in a retrieval-based approach. This dual-verification method enables the scalable assessment of "true" reasoning capabilities.

Current Large Multimodal Models (LMMs) struggle with high-resolution visual inputs during the reasoning process, as the number of image tokens increases quadratically with resolution, introducing substantial redundancy and irrelevant information. A common practice is to identify key image regions and refer to their high-resolution counterparts during reasoning, typically trained with external visual supervision. However, such visual supervision cues require costly grounding labels from human annotators. Meanwhile, it remains an open question how to enhance a model's grounding abilities to support reasoning without relying on additional annotations. In this paper, we propose High-resolution Annotation-free Reasoning Technique (HART), a closed-loop framework that enables LMMs to focus on and self-verify key regions of high-resolution visual inputs. HART incorporates a post-training paradigm in which we design Advantage Preference Group Relative Policy Optimization (AP-GRPO) to encourage accurate localization of key regions without external visual annotations. Notably, HART provides explainable reasoning pathways and enables efficient optimization of localization. Extensive experiments on MME-RealWorld-Lite, TreeBench, V* Bench, HR-Bench-4K/8K, and MMStar demonstrate that HART improves performance across a wide range of high-resolution visual tasks, consistently outperforming strong baselines.

We decompose physician disagreement in the HealthBench medical AI evaluation dataset to understand where variance resides and what observable features can explain it. Rubric identity accounts for 15.8% of met/not-met label variance but only 3.6-6.9% of disagreement variance; physician identity accounts for just 2.4%. The dominant 81.8% case-level residual is not reduced by HealthBench's metadata labels (z = -0.22, p = 0.83), normative rubric language (pseudo R^2 = 1.2%), medical specialty (0/300 Tukey pairs significant), surface-feature triage (AUC = 0.58), or embeddings (AUC = 0.485). Disagreement follows an inverted-U with completion quality (AUC = 0.689), confirming physicians agree on clearly good or bad outputs but split on borderline cases. Physician-validated uncertainty categories reveal that reducible uncertainty (missing context, ambiguous phrasing) more than doubles disagreement odds (OR = 2.55, p < 10^(-24)), while irreducible uncertainty (genuine medical ambiguity) has no effect (OR = 1.01, p = 0.90), though even the former explains only ~3% of total variance. The agreement ceiling in medical AI evaluation is thus largely structural, but the reducible/irreducible dissociation suggests that closing information gaps in evaluation scenarios could lower disagreement where inherent clinical ambiguity does not, pointing toward actionable evaluation design improvements.

To operate reliably under changing conditions, complex systems require feedback on how effectively they use resources, not just whether objectives are met. Current AI systems process vast information to produce sophisticated predictions, yet predictions can appear successful while the underlying interaction with the environment degrades. What is missing is a principled measure of how much of the total information a system deploys is actually shared between its observations, actions, and outcomes. We prove this shared fraction, which we term bipredictability, P, is intrinsic to any interaction, derivable from first principles, and strictly bounded: P can reach unity in quantum systems, P equal to, or smaller than 0.5 in classical systems, and lower once agency (action selection) is introduced. We confirm these bounds in a physical system (double pendulum), reinforcement learning agents, and multi turn LLM conversations. These results distinguish agency from intelligence: agency is the capacity to act on predictions, whereas intelligence additionally requires learning from interaction, self-monitoring of its learning effectiveness, and adapting the scope of observations, actions, and outcomes to restore effective learning. By this definition, current AI systems achieve agency but not intelligence. Inspired by thalamocortical regulation in biological systems, we demonstrate a feedback architecture that monitors P in real time, establishing a prerequisite for adaptive, resilient AI.

AI agents -- systems that execute multi-step reasoning workflows with persistent state, tool access, and specialist skills -- represent a qualitative shift from prior automation technologies in social science. Unlike chatbots that respond to isolated queries, AI agents can now read files, run code, query databases, search the web, and invoke domain-specific skills to execute entire research pipelines autonomously. This paper introduces the concept of vibe researching -- the AI-era parallel to vibe coding -- and uses scholar-skill, a 26-skill plugin for Claude Code covering the full research pipeline from idea to submission across 18 orchestrated phases with 53 quality gates, as an illustrative case. I develop a cognitive task framework that classifies research activities along two dimensions -- codifiability and tacit knowledge requirement -- to identify a delegation boundary that is cognitive, not sequential: it cuts through every stage of the research pipeline, not between stages. I argue that AI agents excel at speed, coverage, and methodological scaffolding but struggle with theoretical originality and tacit field knowledge. The paper concludes with an analysis of three implications for the profession -- augmentation with fragile conditions, stratification risk, and a pedagogical crisis -- and proposes five principles for responsible vibe researching.