Generative image models produce striking visuals yet often misrepresent culture. Prior work has examined cultural bias mainly in text-to-image (T2I) systems, leaving image-to-image (I2I) editors underexplored. We bridge this gap with a unified evaluation across six countries, an 8-category/36-subcategory schema, and era-aware prompts, auditing both T2I generation and I2I editing under a standardized protocol that yields comparable diagnostics. Using open models with fixed settings, we derive cross-country, cross-era, and cross-category evaluations. Our framework combines standard automatic metrics, a culture-aware retrieval-augmented VQA, and expert human judgments collected from native reviewers. To enable reproducibility, we release the complete image corpus, prompts, and configurations. Our study reveals three findings: (1) under country-agnostic prompts, models default to Global-North, modern-leaning depictions that flatten cross-country distinctions; (2) iterative I2I editing erodes cultural fidelity even when conventional metrics remain flat or improve; and (3) I2I models apply superficial cues (palette shifts, generic props) rather than era-consistent, context-aware changes, often retaining source identity for Global-South targets. These results highlight that culture-sensitive edits remain unreliable in current systems. By releasing standardized data, prompts, and human evaluation protocols, we provide a reproducible, culture-centered benchmark for diagnosing and tracking cultural bias in generative image models. Project page: https://seochan99.github.io/ECB

Model stitching, connecting early layers of one model (source) to later layers of another (target) via a light stitch layer, has served as a probe of representational compatibility. Prior work finds that models trained on the same dataset remain stitchable (negligible accuracy drop) despite different initializations or objectives. We revisit stitching for Vision Foundation Models (VFMs) that vary in objectives, data, and modality mix (e.g., CLIP, DINOv2, SigLIP 2) and ask: Are heterogeneous VFMs stitchable? We introduce a systematic protocol spanning the stitch points, stitch layer families, training losses, and downstream tasks. Three findings emerge. (1) Stitch layer training matters: conventional approaches that match the intermediate features at the stitch point or optimize the task loss end-to-end struggle to retain accuracy, especially at shallow stitch points. (2) With a simple feature-matching loss at the target model's penultimate layer, heterogeneous VFMs become reliably stitchable across vision tasks. (3) For deep stitch points, the stitched model can surpass either constituent model at only a small inference overhead (for the stitch layer). Building on these findings, we further propose the VFM Stitch Tree (VST), which shares early layers across VFMs while retaining their later layers, yielding a controllable accuracy-latency trade-off for multimodal LLMs that often leverage multiple VFMs. Taken together, our study elevates stitching from a diagnostic probe to a practical recipe for integrating complementary VFM strengths and pinpointing where their representations align or diverge.

Motivated by discrete diffusion's success in language-vision modeling, we explore its potential for multi-view generation, a task dominated by continuous approaches. We introduce ViewMask-1-to-3, formulating multi-view generation as a discrete sequence modeling problem where each viewpoint is represented as visual tokens from MAGVIT-v2. Through discrete diffusion via masked token prediction, our approach enables progressive multi-view generation via iterative token unmasking, unifying language and vision in a shared token space. Importantly, simple random masking combined with self-attention naturally encourages cross-view consistency without specialized architectures or 3D geometric priors. Our method outperforms the baseline on the GSO and 3D-FUTURE benchmarks, ranking first on average across standard image metrics, and achieving a 10.6% higher IoU than continuous diffusion models on 3D-FUTURE. Furthermore, the proposed framework can be naturally extended to support text-to-image generation and multimodal understanding, highlighting its potential toward a more unified paradigm for multimodal understanding and generation.

The prevalence of Large Language Models (LLMs) in critical applications highlights the need for controlled language generation methods that are both computationally efficient and enjoy performance guarantees. To address this need, we use a common model of concept semantics as linearly represented in an LLM's latent space. In particular, we take the view that natural language generation traces a trajectory in this continuous semantic space, realized by the language model's hidden activations. This view permits a control-theoretic treatment of text generation in latent space, in which we propose Linear Semantic Control (LiSeCo), a lightweight, gradient-free intervention that dynamically steers trajectories away from regions corresponding to undesired meanings. In particular, we propose to directly intervene, in an online fashion, the activations of the token that is being generated in embedding space. Crucially, LiSeCo does not simply steer activations towards a desirable region. Instead, it relies on classical techniques from control theory to precisely control activations in a context-dependent way, and guarantees that they are brought into a specific pre-defined region of embedding space that corresponds to allowed semantics. The intervention is computed in closed form according to an optimal controller formulation, minimally impacting generation time. This control of the activations in embedding space allows for fine-grained steering of attributes of the generated sequence. We demonstrate that our approach is effective on different tasks -- toxicity, sentiment, and language (English/Spanish) steering -- while maintaining text quality.

Reinforcement Learning with Verifiable Rewards (RLVR) improves reasoning in large language models but treats all correct solutions equally, potentially reinforcing flawed traces that arrive at correct answers by chance. We observe that \emph{better reasoning makes better demonstrations}: high-quality solutions serve as more effective in-context examples than low-quality ones. We term this teaching ability \textbf{Demonstration Utility}, and show that the policy model's own in-context learning ability provides an efficient way to measure it, yielding a quality signal termed \textbf{Evidence Gain}. To leverage this signal during training, we introduce \textbf{In-Context RLVR}, which prepends demonstrations before each rollout. Theoretically, we prove that this simple input modification implicitly reweights rewards by a factor approximately proportional to Evidence Gain, assigning higher weights to high-quality traces without requiring costly computation. Experiments on mathematical reasoning benchmarks demonstrate consistent improvements in both accuracy and reasoning quality over standard RLVR baselines. Our codes and datasets are available at https://github.com/Mithas-114/IC-DAPO.

The adaptation of large-scale vision-language models (VLMs) to downstream tasks with limited labeled data remains a significant challenge. While parameter-efficient prompt learning methods offer a promising path, they often suffer from catastrophic forgetting of pre-trained knowledge. Toward addressing this limitation, our work is grounded in the insight that governing the evolutionary path of prompts is essential for forgetting-free adaptation. To this end, we propose EvoPrompt, a novel framework designed to explicitly steer the prompt trajectory for knowledge-preserving fine-tuning. Specifically, our approach employs a Modality-Shared Prompt Projector (MPP) to generate hierarchical prompts from a unified embedding space. Critically, an evolutionary training strategy decouples low-rank updates into directional and magnitude components, preserving early-learned semantic directions while only adapting their magnitude, thus enabling prompts to evolve without discarding foundational knowledge. This process is further stabilized by Feature Geometric Regularization (FGR), which enforces feature decorrelation to prevent representation collapse. Extensive experiments demonstrate that EvoPrompt achieves state-of-the-art performance in few-shot learning while robustly preserving the original zero-shot capabilities of pre-trained VLMs.

Engine sounds originate from sequential exhaust pressure pulses rather than sustained harmonic oscillations. While neural synthesis methods typically aim to approximate the resulting spectral characteristics, we propose directly modeling the underlying pulse shapes and temporal structure. We present the Pulse-Train-Resonator (PTR) model, a differentiable synthesis architecture that generates engine audio as parameterized pulse trains aligned to engine firing patterns and propagates them through recursive Karplus-Strong resonators simulating exhaust acoustics. The architecture integrates physics-informed inductive biases including harmonic decay, thermodynamic pitch modulation, valve-dynamics envelopes, exhaust system resonances and derived engine operating modes such as throttle operation and Deceleration Fuel Cutoff (DFCO). Validated on three diverse engine types totaling 7.5 hours of audio, PTR achieves a 21% improvement in harmonic reconstruction and a 5.7% reduction in total loss over a harmonic-plus-noise baseline model, while providing interpretable parameters corresponding to physical phenomena. Complete code, model weights, and audio examples are openly available.

The growing realism of generative models has blurred the boundary between real and synthetic content, posing significant challenges to reliable AI-generated image detection. Although large-scale pre-trained Vision Foundation Models have advanced detection capability, their generalization to images from unseen generation pipelines remains inadequate. In this paper, we identify, for the first time, a key failure mechanism, termed \emph{semantic fallback}, wherein forensic fine-tuning fails to fully reshape the representation space. Consequently, the resulting representations remain organized along high-level semantic structures rather than manipulation-specific forensic cues. Building on this insight, we propose a \textbf{Geometric Semantic Decoupling (GSD)} framework, which explicitly suppresses semantically dominant directions, thereby promoting invariant forensic representations. Specifically, GSD leverages a frozen CLIP encoder to estimate the dominant semantic subspace via Singular Value Decomposition (SVD). It then suppresses the semantic components through a geometry-constrained formulation with the suppression strength adaptively modulated across samples and layers. We further introduce a mini-batch SVD approximation strategy that amortizes subspace estimation, achieving over a $15 \times$ reduction in computational overhead while preserving effectiveness. Finally, considering practical scenarios spanning both large-scale and online evaluation, we develop three inference protocols, batch, per-sample, and reference-based inference, and demonstrate that they induce consistent semantic decoupling, yielding a stable forgery-oriented feature manifold.

Achieving versatile and natural whole-body humanoid interaction control remains challenging due to the high cost of whole-body teleoperation data. We present ZeroWBC, a teleoperation-free framework that learns humanoid whole-body interaction from human egocentric videos paired with synchronized whole-body motion and text annotations. ZeroWBC adopts a generation-then-tracking formulation to tackle the static scene whole-body interaction control problem. Given an initial egocentric image and a language instruction, a fine-tuned Vision-Language Model generates future human whole-body motion tokens, which are decoded into continuous motions and retargeted to the humanoid. The resulting reference motions, together with root and key body-part trajectories, are then executed by a general interactive motion tracking policy. To improve interaction performance, we introduce an interaction-oriented tracking reward that prioritizes global root and key body-part trajectory alignment while preserving natural whole-body motion. Experiments on the Unitree G1 humanoid robot show that ZeroWBC enables diverse scene-aware behaviors without robot teleoperation demonstrations. These results suggest a scalable paradigm for learning natural humanoid whole-body interaction from human egocentric data.

Planning under partial observability is an essential capability of autonomous robots. The Partially Observable Markov Decision Process (POMDP) provides a powerful framework for planning under partial observability problems, capturing the stochastic effects of actions and the limited information available through noisy observations. POMDP solving could benefit tremendously from massive parallelization on today's hardware, but parallelizing POMDP solvers has been challenging. Most solvers rely on interleaving numerical optimization over actions with the estimation of their values, which creates dependencies and synchronization bottlenecks between parallel processes that can offset the benefits of parallelization. In this paper, we propose Vectorized Online POMDP Planner (VOPP), a novel parallel online solver that leverages a recent POMDP formulation which analytically solves part of the optimization component, leaving numerical computations to consist of only estimation of expectations. VOPP represents all data structures related to planning as a collection of tensors, and implements all planning steps as fully vectorized computations over this representation. The result is a massively parallel online solver with no dependencies or synchronization bottlenecks between concurrent processes. Experimental results indicate that VOPP is at least $20\times$ more efficient in computing near-optimal solutions compared to an existing state-of-the-art parallel online solver. Moreover, VOPP outperforms state-of-the-art sequential online solvers, while using a planning budget that is $1000\times$ smaller.

Learning manipulation policies from human videos could greatly reduce the need for expensive robot demonstrations, but existing approaches typically require restrictive assumptions such as choreographed human motions, predefined keypoints, manual annotations, or known grasp locations. We propose 3PoinTr, a method for pretraining sample-efficient robot policies from unconstrained human videos by predicting dense 3D point tracks. In the unconstrained human demonstration videos, humans are free to follow whatever trajectories and manipulation strategies they see fit, rather than choreographing their motions to mimic a robot. 3PoinTr uses a lightweight visibility-aware transformer to learn how scene points should move from human videos, and then trains a closed-loop multitask robot policy to flexibly extract action-relevant priors from those predicted point tracks. With only 20 action-labeled robot demonstrations, 3PoinTr achieves a 25.0 percentage point higher average success rate than the strongest behavior cloning and video-pretraining baselines on real-world tasks, and a 29.6 percentage point higher average success rate in simulation. Targeted ablations support the key design choices and confirm the benefit of learning from actionless videos. We further show that 3PoinTr's point track prediction transformer outperforms a strong baseline by preserving supervision over partially occluded points. Project page: https://adamhung60.github.io/3PoinTr/.

Computational engine sound modeling is central to the automotive audio industry, particularly for active sound design applications and virtual prototyping. Emerging data-driven engine sound synthesis methods require large volumes of standardized, clean audio recordings with precisely time-aligned operating-state annotations: data that is difficult to obtain due to high costs, specialized measurement equipment requirements, and inevitable noise contamination. We present an analysis-driven framework for generating engine audio with sample-accurate control annotations. The method extracts harmonic structures from real recordings through pitch-adaptive spectral analysis, which then drive an extended parametric harmonic-plus-noise synthesizer. With this framework, we augment 5-10 min of source audio per engine 15-30x via diverse control trajectories and parametric variation, producing the Procedural Engine Sounds Dataset (19.0 h, 5,935 files): a set of engine audio signals with sample-accurate RPM and torque annotations spanning a wide range of operating conditions, signal complexities, and harmonic profiles. Comparison against real recordings validates that the synthesized data preserves characteristic harmonic structures, and a baseline differentiable synthesis network trained on the dataset confirms its suitability for data-driven engine sound modeling. The dataset is released publicly to support research on engine timbre analysis, control parameter estimation, and neural generative synthesis.

Reliable deployment of large language models (LLMs) requires accurate uncertainty estimation. Existing methods are predominantly answer-first, producing confidence only after generating an answer, which measure the correctness of a specific response and limits practical usability. We study a confidence-first paradigm, where the model outputs its confidence before answering, interpreting this score as the model's probability of answering the question correctly under its current policy. We propose CoCA(Co-optimized Confidence and Answers), a GRPO reinforcement learning framework that jointly optimizes confidence calibration and answer accuracy via segmented credit assignment. By assigning separate rewards and group-relative advantages to confidence and answer segments, CoCA enables stable joint optimization and avoids reward hacking. Experiments across math, code, and factual QA benchmarks show improved calibration and uncertainty discrimination while preserving answer quality, thereby enabling a broader range of downstream applications.

Time series foundation models (TSFMs) are revolutionizing the forecasting landscape from specific dataset modeling to generalizable task evaluation. However, we contend that existing benchmarks exhibit common limitations in four dimensions: constrained data composition dominated by reused legacy sources, compromised data integrity lacking rigorous quality assurance, misaligned task formulations detached from real-world contexts, and rigid analysis perspectives that obscure generalizable insights. To bridge these gaps, we introduce TIME, a next-generation task-centric benchmark comprising 50 fresh datasets and 98 forecasting tasks, tailored for strict zero-shot TSFM evaluation free from data leakage. Integrating large language models and human expertise, we establish a human-in-the-loop benchmark construction pipeline to ensure high data integrity and redefine task formulation by aligning forecasting configurations with real-world operational requirements and variate predictability. Furthermore, we propose a novel pattern-level evaluation perspective that moves beyond traditional dataset-level evaluations based on static meta labels. By leveraging structural time series features to characterize intrinsic temporal properties, this approach offers generalizable insights into model capabilities across diverse patterns. We evaluate 12 TSFMs and establish a multi-granular leaderboard to facilitate in-depth analysis and visualized inspection. The leaderboard is available at https://huggingface.co/spaces/Real-TSF/TIME-leaderboard.

Interactive world models continually generate video by responding to a user's actions, enabling open-ended generation capabilities. However, existing models typically lack a 3D representation of the environment, meaning 3D consistency must be implicitly learned from data, and spatial memory is restricted to limited temporal context windows. This results in an unrealistic user experience and presents significant obstacles to downstream tasks such as training agents. To address this, we present PERSIST, a new paradigm of world model which simulates the evolution of a latent 3D scene: environment, camera, and renderer. This allows us to synthesise new frames with persistent spatial memory and consistent geometry. Both quantitative metrics and a qualitative user study show substantial improvements in spatial memory, 3D consistency, and long-horizon stability over existing methods, enabling coherent, evolving 3D worlds. We further demonstrate novel capabilities, including synthesising diverse 3D environments from a single image, as well as enabling fine-grained, geometry-aware control over generated experiences by supporting environment editing and specification directly in 3D space. Project page: https://francelico.github.io/persist.github.io

Agentic language models operate in a fundamentally different safety regime than chat models: they must plan, call tools, and execute long-horizon actions where a single misstep, such as accessing files or entering credentials, can cause irreversible harm. Existing alignment methods, largely optimized for static generation and task completion, break down in these settings due to sequential decision-making, adversarial tool feedback, and overconfident intermediate reasoning. We introduce MOSAIC, a post-training framework that aligns agents for safe multi-step tool use by making safety decisions explicit and learnable. MOSAIC structures inference as a plan, check, then act or refuse loop, with explicit safety reasoning and refusal as first-class actions. To train without trajectory-level labels, we use preference-based reinforcement learning with pairwise trajectory comparisons, which captures safety distinctions often missed by scalar rewards. We evaluate MOSAIC zero-shot across three model families, Qwen2.5-7B, Qwen3-4B-Thinking, and Phi-4, and across out-of-distribution benchmarks spanning harmful tasks, prompt injection, benign tool use, and cross-domain privacy leakage. MOSAIC reduces harmful behavior by up to 50%, increases harmful-task refusal by over 20% on injection attacks, cuts privacy leakage, and preserves or improves benign task performance, demonstrating robust generalization across models, domains, and agentic settings.

This paper presents ShareVerse, a video generation framework enabling multi-agent shared world modeling, addressing the gap in existing works that lack support for unified shared world construction with multi-agent interaction. ShareVerse leverages the generation capability of large video models and integrates three key innovations: 1) A dataset for large-scale multi-agent interactive world modeling is built on the CARLA simulation platform, featuring diverse scenes, weather conditions, and interactive trajectories with paired multi-view videos (front/ rear/ left/ right views per agent) and camera data. 2) We propose a spatial concatenation strategy for four-view videos of independent agents to model a broader environment and to ensure internal multi-view geometric consistency. 3) We integrate cross-agent attention blocks into the pretrained video model, which enable interactive transmission of spatial-temporal information across agents, guaranteeing shared world consistency in overlapping regions and reasonable generation in non-overlapping regions. ShareVerse, which supports 49-frame large-scale video generation, accurately perceives the position of dynamic agents and achieves consistent shared world modeling.

In modern applications such as ECG monitoring, neuroimaging, wearable sensing, and industrial equipment diagnostics, complex and continuously structured data are ubiquitous, presenting both challenges and opportunities for functional data analysis. However, existing methods face a critical trade-off: conventional functional models are limited by linearity, whereas deep learning approaches lack interpretable region selection for sparse effects. To bridge these gaps, we propose a sparse Bayesian functional deep neural network (sBayFDNN). It learns adaptive functional embeddings through a deep Bayesian architecture to capture complex nonlinear relationships, while a structured prior enables interpretable, region-wise selection of influential domains with quantified uncertainty. Theoretically, we establish rigorous approximation error bounds, posterior consistency, and region selection consistency. These results provide the first theoretical guarantees for a Bayesian deep functional model, ensuring its reliability and statistical rigor. Empirically, comprehensive simulations and real-world studies confirm the effectiveness and superiority of sBayFDNN. Crucially, sBayFDNN excels in recognizing intricate dependencies for accurate predictions and more precisely identifies functionally meaningful regions, capabilities fundamentally beyond existing approaches.

Large Language Models (LLMs) are prone to generating plausible yet incorrect responses, known as hallucinations. Effectively detecting hallucinations is therefore crucial for the safe deployment of LLMs. Recent research has linked hallucinations to model uncertainty, suggesting that hallucinations can be detected by measuring dispersion over answer distributions obtained from multiple samples drawn from a model. While drawing from the distribution over tokens defined by the model is a natural way to obtain samples, in this work, we argue that it is suboptimal for the purpose of detecting hallucinations. We show that detection can be improved significantly by taking into account model uncertainty in the Bayesian sense. To this end, we propose a very simple, training-free approach based on perturbing an appropriate subset of model parameters, or equivalently hidden unit activations, during sampling. We demonstrate that our approach significantly improves inference-time hallucination detection over standard sampling across diverse datasets, model architectures, and uncertainty metrics.

In the social sciences, it is often necessary to debias studies and surveys before valid conclusions can be drawn. Debiasing algorithms enable the computational removal of bias using sample weights. However, an issue arises when only a subset of features is highly biased, while the rest is already representative. Algorithms need to strongly alter the sample distribution to manage a few highly biased features, which can in turn introduce bias into already representative variables. To address this issue, we developed a method that uses feature weights to minimize the impact of highly biased features on the computation of sample weights. Our algorithm is based on Maximum Representative Subsampling (MRS), which debiases datasets by aligning a non-representative sample with a representative one through iterative removal of elements to create a representative subsample. The new algorithm, named feature-weighted MRS (FW-MRS), decreases the emphasis on highly biased features, allowing it to retain more instances for downstream tasks. The feature weights are derived from the feature importance of a domain classifier trained to differentiate between the representative and non-representative datasets. We validated FW-MRS using eight tabular datasets, each of which we artificially biased. Biased features can be important for downstream tasks, and focusing less on them could lead to a decline in generalization. For this reason, we assessed the generalization performance of FW-MRS on downstream tasks and found no statistically significant differences. Additionally, FW-MRS was applied to a real-world dataset from the social sciences. The source code is available at https://github.com/kramerlab/FeatureWeightDebiasing.

Plug-and-Play diffusion prior (PnPDP) frameworks have emerged as a powerful paradigm for solving imaging inverse problems by treating pretrained generative models as modular priors. However, we identify a critical flaw in prevailing PnP solvers (e.g., based on HQS or Proximal Gradient): they function as memoryless operators, updating estimates solely based on instantaneous gradients. This lack of historical tracking inevitably leads to non-vanishing steady-state bias, where the reconstruction fails to strictly satisfy physical measurements under heavy corruption. To resolve this, we propose Dual-Coupled PnP Diffusion (DC-PnPDP), which restores the classical dual variable to provide integral feedback, progressively enforce agreement between the data-consistency and prior. However, this rigorous geometric coupling introduces a secondary challenge: the accumulated dual residuals exhibit spectrally colored, structured artifacts that violate the Additive White Gaussian Noise (AWGN) assumption of diffusion priors, causing severe hallucinations. To bridge this gap, we introduce Spectral Homogenization (SH), a frequency-domain adaptation mechanism that modulates these structured residuals into statistically compliant pseudo-AWGN inputs. This effectively aligns the solver's rigorous optimization trajectory with the denoiser's valid statistical manifold. Extensive experiments on CT and MRI reconstruction demonstrate that our approach resolves the bias-hallucination trade-off, achieving state-of-the-art fidelity with significantly accelerated convergence. The code is available at https://github.com/duchenhe/DC-PnPDP

Bubeck and Selke (2021) propose the connection between the Law of Robustness and robust generalization error as an open problem. The Law of Robustness states that overparameterization is necessary for models to interpolate robustly, i.e., the interpolating function is required to be Lipschitz. Wu et al. (2023) extend this law to arbitrary data distributions, proving that the Lipschitz constant satisfies $L = Ω(n^{1/d})$. Robust generalization, on the other hand, asks whether small robust training loss implies small robust test loss. This can be studied using statistical learning techniques such as Rademacher complexities, where a bound on the Rademacher complexity of the robust loss class implies a bound on the Lipschitzness of the function class. We use this connection to explicitly link the two for arbitrary data distributions. (i) We prove that the order of the Lipschitz bound remains the same when considering the global Rademacher complexity of robust loss classes. (ii) At the local scale, i.e., for subsets of functions with small empirical error, the order of the Lipschitz bound changes with the perturbation radius $ρ$ and the localized concentration term $\sqrt{r/n}$.

Advanced reasoning typically requires Chain-of-Thought prompting, which is accurate but incurs prohibitive latency and substantial test-time inference costs. The standard alternative, fine-tuning smaller models, often sacrifices interpretability while introducing significant resource and operational overhead. To address these limitations, we introduce Prompt-Level Distillation (PLD). We extract explicit reasoning patterns from a Teacher model and organize them into a structured list of expressive instructions for the Student model's System Prompt. Evaluated using Gemma-3 4B, PLD improved Macro F1 scores on StereoSet (57\% to 90.0\%) and Contract-NLI (67\% to 83\%), while increasing LogiQA accuracy to 70\%. Similar results on Mistral Small 3.1 demonstrate cross-architecture generalizability, enabling these compact models to match frontier performance with negligible latency overhead. These expressive instructions render the decision-making process transparent, allowing for full human verification of logic, making this approach ideal for regulated industries such as law, finance, and content moderation, as well as high-volume use cases and edge devices.

Large language models (LLMs) such as GPT-5 and Gemini 3 have pushed the frontier of automated reasoning and code generation. Yet current benchmarks emphasize accuracy and output quality, neglecting a critical dimension: efficiency of token usage. The token efficiency is highly variable in practical. Models solving the same problem with similar accuracy can exhibit up to a \textbf{5.0$\times$} difference in token length, leading to massive gap of model reasoning ability. Such variance exposes significant redundancy, highlighting the critical need for a standardized benchmark to quantify the gap of token efficiency. Thus, we introduce OckBench, the first benchmark that jointly measures accuracy and token efficiency across reasoning and coding tasks. Our evaluation reveals that token efficiency remains largely unoptimized across current models, significantly inflating serving costs and latency. These findings provide a concrete roadmap for the community to optimize the latent reasoning ability, token efficiency. Ultimately, we argue for an evaluation paradigm shift: tokens must not be multiplied beyond necessity. Our benchmarks are available at https://ockbench.github.io/.

Continuum robots are flexible, slender manipulators well suited for confined surgical environments. In these settings, unknown interaction forces and model uncertainty significantly affect robot shape, motivating state estimation from external observations. Existing estimation methods either neglect actuation modeling or rely on simplified deterministic actuation models. In contrast, we jointly estimate robot shape, external loads, and actuation inputs using mechanically principled actuation priors. To achieve this, we present a discrete Cosserat rod formulation with piecewise-linear strain integration that provides high numerical accuracy while inducing a sparse factor graph structure for efficient nonlinear optimization. We extend the framework to tendon-driven and parallel robots in simulation and validate it experimentally on a surgical concentric tube robot. Overall, our approach enables principled real-time estimation across multiple robot architectures while providing direct access to manipulator Jacobians through the linearized factor graph.

Value alignment of Large Language Models (LLMs) requires us to empirically measure these models' actual, acquired representation of value. Among the characteristics of value representation in humans is that they distinguish among value of different kinds. We investigate whether LLMs likewise distinguish three different kinds of good: moral, grammatical, and economic. By probing model behavior, embeddings, and residual stream activations, we report pervasive cases of value entanglement: a conflation between these distinct representations of value. Specifically, both grammatical and economic valuation was found to be overly influenced by moral value, relative to human norms. This conflation was repaired by selective ablation of the activation vectors associated with morality.

Scalable methods for networked multi-agent reinforcement learning let each agent plan using only a small neighborhood of the agent graph. This works only when the system is value-local, meaning a perturbation at one agent affects the long-run value at another agent weakly when the two are far apart. In the average-reward setting, the standard way to certify locality is the Dobrushin row-sum bound on a single matrix $C^π$ that captures how each agent's next state depends on each other agent's current state. To make this matrix easy to work with, prior work bounds it by a supremum over joint actions. The resulting bound is independent of the policy, but it is loose whenever the policy never picks the worst-case action. We split $C^π$ into pieces that separately track environment sensitivity and policy sensitivity, $C^π\preceq E^{\mathrm s}+E^{\mathrm a}Π(π)$, where $E^{\mathrm s}$ measures how the next state moves with the current state, $E^{\mathrm a}$ measures how it moves with the current action, and $Π(π)$ measures how reactive the policy is to changes in state. The spectral radius of $H^π:= E^{\mathrm s}+E^{\mathrm a}Π(π)$ then controls the decay of the average-reward Poisson solution, and the spectral certificate $ρ(H^π)<1$ is strictly weaker than the row-sum condition $\|H^π\|_\infty<1$ on the same matrix and applies in regimes where policy-independent action-supremum bounds used in prior Dobrushin-style work cannot. For temperature-$τ$ softmax policies we get $Π(π)\le L/(2τ)$, so the softmax temperature directly controls locality. We use this decay result to give a deterministic oracle guarantee for a block-coordinate KL-proximal policy-improvement template whose truncation bias decays exponentially in the message-passing radius $κ$.

Recent work leverages the capabilities and commonsense priors of generative models for robot control. In this paper, we present an agentic control system in which a reasoning-capable language model plans and executes tasks by selecting and invoking robot skills within an iterative planner and executor loop. We deploy the system on two physical robot platforms in two settings: (i) tabletop grasping, placement, and box insertion in indoor mobile manipulation (Mobipick) and (ii) autonomous agricultural navigation and sensing (Valdemar). Both settings involve uncertainty, partial observability, sensor noise, and ambiguous natural-language commands. The system exposes structured introspection of its planning and decision process, reacts to exogenous events via explicit event checks, and supports operator interventions that modify or redirect ongoing execution. Across both platforms, our proof-of-concept experiments reveal substantial fragility, including non-deterministic suboptimal behavior, instruction-following errors, and high sensitivity to prompt specification. At the same time, the architecture is flexible: transfer to a different robot and task domain largely required updating the system prompt (domain model, affordances, and action catalogue) and re-binding the same tool interface to the platform-specific skill API.

Recent robot foundation models largely rely on large-scale behavior cloning, which imitates expert actions but discards transferable dynamics knowledge embedded in heterogeneous embodied data. While the Unified World Model (UWM) formulation has the potential to leverage such diverse data, existing instantiations struggle to scale to foundation-level due to coarse data usage and fragmented datasets. We introduce LDA-1B, a robot foundation model that scales through universal embodied data ingestion by jointly learning dynamics, policy, and visual forecasting, assigning distinct roles to data of varying quality. To support this regime at scale, we assemble and standardize EI-30k, an embodied interaction dataset comprising over 30k hours of human and robot trajectories in a unified format. Scalable dynamics learning over such heterogeneous data is enabled by prediction in a structured DINO latent space, which avoids redundant pixel-space appearance modeling. Complementing this representation, LDA-1B employs a multi-modal diffusion transformer to handle asynchronous vision and action streams, enabling stable training at the 1B-parameter scale. Experiments in simulation and the real world show LDA-1B outperforms prior methods (e.g., $π_{0.5}$) by up to 21\%, 48\%, and 23\% on contact-rich, dexterous, and long-horizon tasks, respectively. Notably, LDA-1B enables data-efficient fine-tuning, gaining 10\% by leveraging 30\% low-quality trajectories typically harmful and discarded.

Test-time alignment of large language models (LLMs) attracts attention because fine-tuning of LLMs requires high computational costs. In this paper, we propose a new test-time reward-guided alignment method called adaptive importance sampling on pre-logits (AISP) on the basis of the sampling-based model predictive control with the stochastic control input. AISP applies the Gaussian perturbation into pre-logits, which are outputs of the penultimate layer, so as to maximize expected rewards with respect to the mean of the perturbation. We demonstrate that the optimal mean is obtained by importance sampling with sampled rewards. AISP outperforms best-of-n sampling in terms of rewards over the number of used samples and achieves higher rewards than other reward-based test-time alignment methods.