Papers

Modern machine learning models are typically trained on data assumed to be representative of the predictive tasks the model will face in the future, but practical datasets violate this assumption. We focus on outcome-dependent truncation in which an observation appears in the training data depending on its label. Our work connects two largely distinct literatures: efficient learning under truncation and survival analysis. Existing methods for truncated survival analysis do not characterize the optimization landscape and instead rely on iterative procedures such as expectation maximization. We address this gap and prove that Cox regression can be efficiently learned under truncated and right-censored data using first-order optimization when the baseline hazard is known (despite nonconvexity). In contrast, when the baseline hazard is learned jointly, we construct identifiable and correctly specified models whose negative log-likelihood contains strict suboptimal local minima. On experiments with four survival datasets, we show that accounting for truncation reduces the prediction bias induced by naive Cox regression and extends to neural networks. We also show orders of magnitude speedups compared to standard expectation maximization.

When a trader executes a large order, the relevant question is not what will the price be? but what would the price have been had I not traded? Market impact is thus a causal quantity, yet classical models (Almgren–Chriss, propagators) regress price on flow, which is biased whenever informed flow moves prices through information as well as trading. We present a proof-of-concept Causal Market Simulator: a continuous-time causal prior-data fitted network (PFN) trained on a synthetic market prior in which a hidden informational flow confounds the flow–price relation by construction, with confounding strength an exact per-episode knob. On this prior, the naive estimator's bias follows a closed form we confirm empirically (up to 68% overestimation), while the PFN tracks the unbiased true-parameter oracle at high confounding. Its do-ablated twin, the same model with the intervention inputs removed, collapses onto the pre-window mean. Zero-shot, the model responds at the ground-truth scale to metaorders in an ABIDES limit-order book ($+0.38$ pre-onset standard deviations against seed-replay counterfactuals), and its intervention response replicates on about 2,300 detected flow bursts in BTC/ETH/SOL perpetuals and 3,800 on Nasdaq equities, evidence consistent with, though not proof of, causal transfer. Off-prior the response degrades gracefully, and split-conformal recalibration repairs interval calibration. Transfer hinges on encoding the metaorder as added flow rather than a clamped price level, not on more training. A recurrent state-space variant closes the remaining shape gap: its impact accrues over execution duration where the transformer's stays flat, raising the correlation with measured metaorder impact from $+0.68$ to between $+0.92$ and $+0.96$, at the cost of weaker on-prior deconfounding that traces to the training recipe rather than to recurrence.

EliteTable is a self-improving program-search algorithm guided by large language models (LLMs) that uses a fixed-capacity table of high-scoring solutions to guide new proposals, edits of existing solutions, and remixes of several solutions. Performance was evaluated across six Gymnasium control tasks, each with ten generations and fifty candidate slots per generation. Each task was evaluated across ten independently initialized searches. Candidate-score medians rose between the first and final generations in all six tasks, and selected-winner search-score medians rose in five. In CartPole-v1, the selected winner reached the search-score ceiling in the first generation of all ten searches. All ten paired held-out endpoints improved in four of the six tasks. BipedalWalker-v3 and CarRacing-v3 improved without reaching their reporting targets. Six of ten Acrobot-v1 searches declined on held-out episodes, and the median paired change was negative. The results document improvement during program search and the transfer of selected-winner gains to new episodes in some tasks. Comparisons and ablations are needed to establish EliteTable's relative performance and the contributions of its components.

Privacy-preserving order matching enables traders to discover and execute compatible orders while limiting disclosure of sensitive trading information. Existing systems span financial markets and blockchain-based asset exchanges and employ diverse market structures, privacy mechanisms, and trust assumptions, making their privacy guarantees difficult to compare. We present a systematization of privacy-preserving order matching covering 22 protocol families. We organize the order lifecycle into four phases: order representation and submission, order discovery and collection, order-matching, and settlement, and characterize each phase by protected and exposed information, access roles, privacy mechanisms, and trust assumptions. Our analysis shows that privacy is not a single protocol-level property: protecting selected order attributes or individual phases does not necessarily prevent leakage elsewhere in the lifecycle. We identify recurring trade-offs among privacy, functionality, trust, and verifiability, highlighting the challenge of achieving end-to-end privacy across the complete order lifecycle.

Large-scale tensors arising in science and engineering are often too large to form or store explicitly. Their multilinear contractions, however, can frequently be evaluated efficiently, which makes randomized Frobenius-norm estimation a practical alternative to exact computation. In this paper, we study estimation of the squared Frobenius norm of a real order-$d$ tensor from such contractions using independent real, complex, or quaternion Gaussian probes. For a rank-one tensor, one normalized sample is a product of $d-1$ independent Gamma variables. This gives exact relative-variance factors $3^{d-1}-1$, $2^{d-1}-1$, and $(3/2)^{d-1}-1$, respectively. Under the same budget of underlying real Gaussian vectors, the quaternion probe has smaller rank-one variance than four independent real probes for $d\ge3$; a real-channel representation identifies the gain as modewise pooling. For orthogonal CP tensors, the exact relative variance equals the corresponding rank-one factor divided by the participation-ratio effective rank and the number of samples. For arbitrary real tensors, we prove sharp uniform moment bounds, attained at rank one, and obtain Chebyshev guarantees. A classical block-median construction then gives two-sided exponential confidence for the modified estimator. Numerical examples test the rank-one and effective-rank formulas. An implicit infinite-series kernel tensor illustrates contraction-based norm evaluation; EEG maps provide a qualitative spatial interpolation illustration.

We investigate the dynamics of charged test particles around a Letelier black hole, i.e. a black hole surrounded by a cloud of strings (CoS), in Kalb-Ramond (KR) gravity, immersed in an external, asymptotically uniform magnetic field aligned with the symmetry axis. The geometry depends on the CoS parameter and on the Lorentz-violating KR parameter only through a single combination of the two, so that the two effects can compensate each other and the Schwarzschild metric is recovered when both parameters are equal. Because the spacetime is not Ricci flat, the Wald potential is not a solution of the Maxwell equations; we show that the regular source-free magnetic test field is given in closed form by a Gauss hypergeometric function of the radial coordinate, which reduces to the Wald solution in the Schwarzschild limit. Using this exact field we derive the equations of motion, the effective potential, the circular orbits, the innermost stable circular orbit (ISCO), representative escaping, captured and curled trajectories, and the orbital, radial and vertical epicyclic frequencies in closed form; all frequency expressions are cross-checked against direct numerical integration of the full equations of motion. Within the relativistic precession model, we perform a Markov chain Monte Carlo (MCMC) analysis of the twin-peak high-frequency quasi-periodic oscillations (QPOs) of GRO J1655-40, XTE J1550-564 and GRS 1915+105 for four nested models, complemented by nested sampling to compute Bayesian evidences and to weight the two (Larmor and anti-Larmor) magnetic modes of the posterior. The Schwarzschild model is excluded for GRO J1655-40 and XTE J1550-564, with a difference in the Akaike information criterion of about 110.

Generative creative interfaces routinely report the user's position within a space of possibilities. Such displays imply that the quantities shown were measured from the artefact; in a growing class of systems they are instead predicted by a language model describing its own output, and the two are indistinguishable on screen. Most creative domains lack an agreed measurement against which such a self-report can be checked; audio admits one. In a measurement study of a working LLM-driven timbre interface, what a user can cause proves separable from what the interface reports. The first is exactly characterisable: the reachable set is a zonotope whose dimension increases by one per sound placed, so the space the model may act within is demonstrably the user's own. The second is inaccurate. The displayed position differs from measurement by approximately three times the width of that region, and the error has two independent sources: model-predicted coordinates, whose accuracy varies twofold between the two models evaluated, and the interpolation step, which arises from combining synthesiser parameters and is unaffected by model quality. End to end, 26 of 36 refinement instructions produced no perceptible change, and under the weaker model the facility performed no action while reporting convergence. Effective agency (what a user can cause) and epistemic agency (whether they can know it) are therefore separable and independently engineered: structurally enforced constraints survive model failure, whereas guarantees resting on the model's own report do not. The failure mode that remains is silence presented as success. We close with a short protocol for building and auditing such displays.

We give an explicit formula for the Lusztig parameters of determinantial modules over symmetric quiver Hecke algebras. The formula applies to every Bruhat pair and every fixed reduced expression of the upper Weyl group element. In finite types $A$, $D$, and $E$, we identify the parameters of the modules in Ménard's seed using Demazure weaves. We then factor prefix determinantial modules into these seed modules, with multiplicities given by Lusztig cycles. We also give a criterion for recognizing monomials in this fixed seed and an explicit formula recovering their multiplicities from Lusztig parameters.

Training-free feature reuse accelerates diffusion transformer inference by replacing full denoising computation at selected steps with cached approximations. Existing methods mainly improve the reconstruction of skipped features, but we find that once a stable first-order reconstructor is used, increasing reconstruction complexity brings limited gains, whereas the placement of exact-computation steps has a much larger effect on fidelity. More importantly, absolute shallow-probe variation is jointly affected by input-trajectory displacement and network response strength, and therefore does not directly reflect cache risk. We address this confounding by normalizing probe-output variation with the corresponding input displacement, yielding a directional gain per unit trajectory movement. Based on this signal, we propose CacheScope, which identifies high-risk reuse steps online and selectively falls back to full denoising computation while leaving the underlying cache reconstructor unchanged. CacheScope requires no additional training, gradients, auxiliary networks, or prompt-specific offline schedule optimization. Across FLUX.1-dev, Wan2.1-T2V-1.3B, and Open-Sora 1.2, CacheScope achieves up to 3.57x acceleration and reduces median LPIPS by up to 70.1% over the strongest probe-based baseline.

Large language models (LLMs) are increasingly being explored to automate operations research modeling, reduce modeling time and lower the expertise barrier to optimization based decision support. However, this process often introduces inconsistencies between generated artifacts and the original problem, which we refer to as optimization modeling hallucinations. Despite progress in general-purpose hallucination mitigation, optimization modeling poses a distinct repair challenge: restoring fidelity to the original problem through coordinated model-code updates. We introduce OPT-Kraken, a training-free hallucination repair framework. To the best of our knowledge, it is the first post-hoc framework for repairing hallucinations in LLM-based optimization modeling artifacts. It extracts problem requirements and instance data, localizes errors by checking problem-model fidelity, model-code consistency, and data grounding, and selects repair granularity by the affected scope. For local repair, category rules and instance evidence are compiled into contracts specifying permitted edits, expected effects, and protected content. Experiments on established optimization modeling benchmarks show positive gains across automatic modeling frameworks and fine-tuned models. On zero-shot artifacts from multiple base models, OPT-Kraken effectively repairs optimization modeling hallucinations while preserving initially correct artifacts. These results demonstrate effective and controlled repair for LLM-based optimization modeling artifacts.

Continuous glucose monitoring (CGM) foundation models encode each day as one vector that mixes who the person is, what happened that day and how the sensor measured it, so measurement differences travel with the representation into new cohorts. We ask whether pretraining can place the measurement component that harms transfer in known coordinates, so that deployment drops them with a fixed projection and no target data. GlucoPRISM supervises a GlucoFM backbone with response contracts, rules for how each part of the representation should change when only the day or only the sensor changes: a 64-dimensional Trait band should change with neither, a 48-dimensional State band only across days, and a 16-dimensional Sensor band should track the device. Synthetic sensor views calibrated on paired Dexcom and Libre recordings, different-day pairs and glucose-summary targets supply this supervision without clinical labels. Across twelve cross-cohort directions, dropping the Sensor band from the same frozen encoder raises mean transfer AUROC from 68.75 to 73.57, against 69.3 for GlucoFM. The gain has two parts. Into CGMacros, the one target cohort recorded with two sensor types, filtering adds 11 to 19 points, and this holds when the generator is recalibrated on half of the participants and scored on the other half. On the other eight directions filtering is neutral and the encoder itself exceeds GlucoFM by 3.5 points. The device remains linearly decodable after filtering, so the filter removes a device-related component that harms transfer rather than all measurement information. The retained bands improve window AUROC on thirteen of fourteen public task–cohort cells (Holm-adjusted p=0.031) and correlation on fourteen of fifteen clinical endpoints in an external cohort of 9,303 participants.

Distribution operators need visibility of behind-the-meter photovoltaic (PV) generation, yet most residential customers are observed only through net smart meters. Deep-learning-based PV disaggregation can restore this visibility, but its operational value depends on whether the model serves known customers, new customers, future periods, or new regions. We quantify this dependence using four deployment regimes that separate household transfer from temporal extrapolation, applied to Ausgrid smart-meter data with cross-dataset adaptation to the Pecan Street data. Using a fixed training protocol, we benchmark representative deep sequence architectures for PV disaggregation under regime-matched evaluation. At a fixed 48-hour context length, daylight mean absolute error (MAE) of the best model rises from 0.104~kW on seen households in the future test period to 0.185~kW on unseen households over the same period, and under a fixed training contract the deployment regime explains far more of the error variation than model choice. Under target-domain normalization, fine-tuning on five target-region households reduces source-model transfer error by 28%–36%. Long input windows help known-customer extrapolation, while their new-customer benefits vary by model. These findings suggest regime-matched error budgets, limited target supervision for model adaptation in new regions, and use-case-dependent context selection for energy data analytics.

Machine learning is shown not to increase Eve's successful bit-guessing probability in the binary-classifier-based wire-resistance attack on the Kirchhoff Law-Johnson Noise (KLJN) key exchange when the resistor gap is narrow. Using time-domain simulations of a non-ideal KLJN loop with $R_L = 9$ k$\Omega$, $R_H = 10$ k$\Omega$ and wire resistances from 0 to 3200 $\Omega$, we generate 3000 secure bits per wire resistance and evaluate the mean-square noise voltages at both ends over 100-sample bit exchange periods (BEPs). This probability $p$ is evaluated for the sign rule, the nearest-line rule, logistic regression, support vector machine (SVM), $k$-nearest neighbors, random forest and gradient boosting. The sign rule gives $p = 0.711$ at 50 $\Omega$ and 0.682 at 3200 $\Omega$. The nearest-line rule, logistic regression and SVM stay close to it and the other classifiers perform worse. For Gaussian noise, the likelihood ratio between the HL and LH cases depends only on the difference of the two mean-square voltages, so the sign rule is Bayes-optimal. The resulting $p_{\max}$ reproduces the measured values and predicts that about 900 to 1200 samples per bit are needed for $p = 0.95$. This difference carries about 0.19 bit per secure bit, 1.5 times the hard-decision value.

We compare three definitions of $\Xi_c$–$\Xi_c'$ mixing within a constituent-quark model: a spin-component probability angle, a common-source overlap angle, and an SU(3)-breaking mass-operator angle. The probability angle is $3.11^\circ$, while the two operator-based definitions give $2.48$–$2.58^\circ$; a second parameter set gives $2.38$–$2.49^\circ$. The operator-based angles are closer to the lattice determinations $1.2(1)^\circ$ and $0.810(50)(200)^\circ$, but remain larger. We then couple conventional $qsQ$ states to virtual $qsQg$ configurations, where $q=u,d$ and $Q=c,b$. An orbital-$S$-wave chromomagnetic transition containing both $j_{Qg}=1/2$ and $3/2$ raises the source and mass-operator charm angles to $2.67$–$3.16^\circ$ when normalized to reference short-range mass shifts. The hybrid probability stays below $1.7\%$, and the corresponding bottom angles are $0.42$–$0.48^\circ$. Eliminating the hybrid sector gives a rank-one mass correction precisely when all weighted transitions select the same retained-state combination. The mode-resolved calculation separates the flavor-breaking response that changes the mixing angle from the response that generates a second transition direction. The accompanying effective-current relation connects this mass analysis to radiative and semileptonic amplitudes through their current matrix elements.

As autonomous AI agents become more integrated into sensitive domains, it is crucial that single and multi-agent AI systems are secured. This paper investigates how prompt injection attacks can exploit both direct user inputs and inter?agent communication to leak sensitive information and trigger cross-agent infections. While our tests mostly highlight key vulnerabilities in single-agent systems, we also extend these attacks to multi-agent frameworks, investigating how weaknesses can propagate across agents. Our findings emphasize the need for more robust defenses as AI agents become more commonly embedded in real-world workflows.

Machine-learned conformer generators are scored on GEOM-DRUGS by proximity to a reference ensemble that standard benchmark metrics implicitly treat as ground truth. We test whether that finite set is complete on the GFN2-xTB surface used to construct it. Across five public generators, settled, low-energy structures remain outside the $0.75$\,Å reference radius after removing reflections, incorrect stereochemistry, and post-relaxation connectivity changes ($2.3$–$7.0\%$ of submitted conformers, depending on architecture). In a targeted DFT audit, $107$ of $113$ chemically screened novel/reference pairs remain outside that radius after full $\omega$B97X-D3BJ re-optimization. A threshold-free complementary test needs no radius and no energy window at all: on $15.0\%$ $[8.0, 22.0]$ of molecules some generator has already found a chemically intact structure lying below the entire released ensemble. A complementary CREST search reproduces the minimum released in $65$ of $69$ molecules, and in $11$ of them it reaches more than $0.5$\,kcal/mol below the complete released ensemble. Augmenting the references with conformers proposed by held-out architectures raises absolute coverage-precision for every generator tested, by $1.48$–$2.99$ points. Together, these results show that GEOM-DRUGS precision is sensitive not only to generator quality but also to incomplete reference support.

Large language models (LLMs) are increasingly used in molecular discovery and structure elucidation, where decisions depend on physical observations. Existing evaluations primarily focus on end-to-end molecular reconstruction, where correct predictions do not necessarily indicate reliance on the supplied properties. We introduce MolP2SBench, a benchmark for molecular property-to-structure (P2S) reasoning in LLMs. Each task presents a masked molecular scaffold with given properties and asks the model to select the missing structure from plausible candidates. We shuffle or replace properties to test reliance on correct property-structure pairing and responsiveness to counterfactual property changes. A channel-wise leaderboard uses Property-Grounded Recovery (PGR), giving equal weight to source and counterfactual recovery. Evaluating seven models, we find that LLMs can use physical properties for structure prediction, but performance varies substantially across property types. Higher accuracy with a property does not necessarily imply greater responsiveness to it: a model may benefit more from $^1$H NMR than from $^{13}$C NMR, yet be less likely to change its prediction when the $^1$H data is replaced. LLM performance also need not reflect the relative difficulty of these properties for human chemists. Although UV-Vis spectra provide less direct structural information than NMR spectra, Astra achieves higher accuracy with UV-Vis data than with NMR peaks. MolP2SBench supports studying how LLMs extract and use physical-property information, revealing patterns that may differ from conventional chemical intuition and providing a foundation for more reliably property-grounded models for molecular and quantum-chemical inference. Code and dataset: https://github.com/DUAL-Xiao/MolP2SBench.

While learning dexterous manipulation from a single human video offers a promising alternative to costly robot demonstrations, many recent methods predominantly imitate demonstrated motions. Such strict motion matching often limits generalization to initial object poses, goal poses, and grasps not shown in the video. Alternatively, discovering a policy via reinforcement learning (RL) allows for broad generalization, but without prior guidance, it struggles with high-dimensional exploration in complex, multi-stage tasks. To address these coupled generalization and exploration challenges, we present Dex-One2Many, a real-to-sim-to-real framework that learns a generalizable dexterous manipulation policy from a single human video. Our key insight is to abstract the video into sequential scene graphs that guide RL, enabling efficient exploration while preserving broad generalizability. The graphs serve as generative constraints for sampling diverse reset states and provide dense rewards for each stage. Because the graphs constrain relations rather than exact poses, these reset states cover object poses and grasps beyond the video, while initializing each stage from them with dense rewards keeps exploration short and guided. Trained entirely in simulation, Dex-One2Many transfers zero-shot to a real multi-fingered hand. Across five tool-use and manipulation tasks, Dex-One2Many exceeds baselines by 6.5% in seen configurations, while its robust generalization widens this gap to 71% in unseen scenarios.

Instruction-guided image editors have become highly capable, yet improving them further still depends on human-edited training pairs or external reward models. Such supervision is costly to obtain and can reward plausible failures: a realistic output may leave the requested change undone or alter content that should be preserved. In this work, we strive to improve a pretrained image editor using only its own generations, without human-edited targets or an external training-time reward model. To this end, we propose a self-evolving framework, named Rubric-CEPR, that verifies the editor's own samples with its internal representations through a rubric-augmented Contrastive Edit-Preservation Reward (CEPR). A Planner proposes structured edit instructions from unlabeled images, the Editor samples multiple candidate edits, and a frozen Critic scores each candidate with decomposed rubric checks for edit realization, removal of the old state, and content preservation, using features already exposed by the editor. Non-compensatory gates reject infeasible candidates, and the best verified candidate is distilled into the editor through lightweight adapter training. On Qwen-Image-Edit, Rubric-CEPR improves ImgEdit from 4.36 to 4.60 (+5.5%), with a +24.9% gain on object isolation, and transfers to GEdit-Bench and Complex-Edit. The same procedure also improves Step1X-Edit by +7.8% on ImgEdit. We hope our approach will serve as a solid baseline for image editors that improve themselves from their own verified samples. Our code is publicly available at https://github.com/riteshthawkar/Rubric-CEPR

We present DreamTrue, a multi-view, cross-embodiment robot world model for action-faithful and physically plausible video prediction. Training such a model on existing robot datasets faces two obstacles: imprecise calibration can impair action following, while limited coverage of unsuccessful interactions can bias predictions toward successful outcomes. To improve action following across embodiments, we render action trajectories into image-space conditions and introduce offline geometric calibration to align these conditions with the target videos. To broaden interaction coverage, we introduce counterfactual post-training, modifying recorded action trajectories and generating future videos under a wider range of actions and contact configurations. To provide feedback on these predictions without paired ground-truth futures, we construct a human-annotated video dataset covering robot, object, and interaction defects and use it to train an embodied video reward model. Its scores guide reinforcement-learning post-training toward more physically plausible interaction outcomes. On AgiBot, DreamTrue attains state-of-the-art action following, while reducing the human-assessed interaction defect rate from from 48.12% to 6.25%. Notably, our model ranks first in the world model track of the AgiBot World Challenge 2026. The project page can be found at https://brave-eai.github.io/DreamTrue.

Text-conditioned motion generators produce trackable whole-body motion, but they have no notion of scene-dependent safety: the same action may target an object or a person. Existing safeguards either inspect the prompt, require labeled motion data, or enforce geometric constraints; therefore, they do not directly account for how scene context changes a motion's meaning. We introduce contextual safety filtering (CSF), a training-free filter that grounds natural-language safety rules in safe and unsafe reference trajectories produced by the generator. For each active rule, safe and unsafe reference trajectories define an affine safety value that a safe reference tracking CBF-QP enforces. Across four pretrained generators with different architectures, CSF activates the intended rules in all explicit and scene-triggered unsafe cases and reduces the danger-event rate by up to 90%, while preserving 88-100% of benign motions. We demonstrate the complete system on a real-world Unitree G1, where it successfully prevents unsafe motions in a variety of scenarios, including interactions with humans and objects.

METR's 50% time horizon measures the human completion time of software tasks that an AI solves with 50% probability, allowing AI capabilities to be expressed in interpretable units. On 228 tasks and 26 AIs, we recompute the time horizons using splines and item-response theory to relax the assumption that the AI difficulty of a task depends linearly on the log of human time. Our fitted spline can be interpreted as a function that converts human time to AI difficulty; it is nearly flat in a region from 2–30 min but close to linear elsewhere. Hence, a time-horizon jump from 3 min to 30 min is much easier than one from 30 min to 5 hours despite the same multiplier of $10 \times$. Overall, we contribute time-horizon point estimates that perform better under a cross-validated suite of proper scoring rules, as well as diagnostic plots for assessing time horizons' construct validity. We suggest that time horizons be interpreted together with the diagnostic plots, especially as new time-horizon-based benchmarks are proposed or existing ones grow to include longer tasks.

General-purpose robots must perform a wide range of tasks from agile locomotion to dexterous manipulation. While sim-to-real reinforcement learning (RL) has proven to be a useful tool for this goal, current RL pipelines depend on engineering-heavy, per-task structural priors such as shaped rewards and demonstrations. Recent work has shown that diverse simulator resets, combined with massively parallel simulation, can alleviate much of this engineering burden on several manipulation problems. However, we find that naively scaling this paradigm to more precise or dynamic problems remains non-trivial. While simulator resets can help with exploration, uniformly sampling over this distribution wastes a growing fraction of learning experience on task configurations the policy has already mastered or cannot yet attempt. This makes it challenging to see the expected benefits of scaling parallel environments for RL, since much of the learning signal in a batch is wasted during learning. To mitigate this, we introduce Success Guided Sampling (SGS), a simple adaptive sampler that concentrates RL training on task configurations around the frontier of the policy's capabilities. Doing so allows large-scale simulated RL to make the most out of the experience in a batch, enabling much more effective scaling to large-scale parallel simulation. Across experiments using up to $2^{20}$ (over one million) parallel environments, SGS enables RL to solve challenging multi-terrain quadruped locomotion and contact-rich assembly tasks that prior methods fail to solve. Finally, we distill the learned manipulation policies into RGB-based policies and demonstrate zero-shot transfer to several challenging assembly tasks on real hardware. Project website: https://sgs-rl.github.io/.

World models offer a promising alternative to physics-based simulators, yet remain far from practical deployment. We ask how far scaling ego-centric human video takes them, using a dataset of 30,000 hours spanning over 1,000 scene types and 14,000 contributors. Rather than relying on opaque downstream metrics, we directly evaluate agent and object-interaction fidelity on a challenging out-of-distribution benchmark. Increasing training data by 100x improves both, but unevenly: the agent is modeled well, while object fidelity remains far lower and improves slowly. We show that the agent gains need not come from data, and a careful visual conditioning design saturates fidelity with a fraction of it, which lets us measure object fidelity on its own and discover its saturation point. We then introduce a supervision scheme that shifts capacity from scene appearance toward object dynamics, improving object fidelity though a substantial gap remains. Finally, our conclusions transfer to downstream humanoid modeling. Overall, our results suggest that scaling ego-centric data brings agent modeling close to its limit while leaving its effects on the world far behind, and that closing this gap will depend on how models are trained, not only on how much data they see.

In 2026, cybersecurity evaluations involving OpenAI, Anthropic, and Google agents reached real systems outside their authorized test scope. The paths were different. OpenAI agents exploited research infrastructure, coordinated across runs, and compromised parts of Hugging Face's production environment. Anthropic reported cases in which a misconfigured third-party environment exposed real systems to agents pursuing simulated cyber tasks. In a separately reported evaluation, Google's Gemini accessed three real organizations through an unintended internet route; Google stated that the model stopped in all three instances. Taken together, the cases show why an evaluation cannot rely on an assumed boundary. That boundary must be verified while the agent is operating. This comparative instrumental case study develops a Proactive Agent Security Assurance Cycle (PASAC) and a five-layer Boundary Assurance Stack. The framework combines risk-tiered task design, executable scope contracts, pre-run validation, least-capability access, independent egress enforcement, credential restrictions, cross-run monitoring, automatic stop conditions, and evidence-based reauthorization. A leading-indicator model, nine design propositions, and seven falsifiable hypotheses turn these lessons into a testable research program. Because the public Gemini record is limited to attributed statements and journalism, its detailed causal mechanism remains provisional. The central conclusion is straightforward: proactive agent security requires continuous assurance across the full execution system, not confidence in any single sandbox or safeguard.

Variability is a crucial probe of active galactic nuclei (AGN) disks, but the impact of the structure and internal physics of the disk on this variability is poorly understood. We perform 3D multi-frequency radiation magnetohydrodynamic simulations of the $1300-5940~\rm{\AA}$ emitting region of an AGN disk. This turbulent disk simulation happens to stochastically have different disk thickness above and below the midplane. Light curves emitted from the thin top of the disk are broadly consistent with traditional variability models, where the X-ray irradiation emitted from the corona is the main driver of variability in UV-optical light curves. Conversely, the UV-optical light curves emitted from the thicker bottom of the disk are not strongly correlated with the X-ray light curve and have disk continuum reverberation mapping lags that are $3-5\times$ longer than the light travel time. While these features are inconsistent with traditional variability models, they are in good agreement with numerous recent observations. Our simulations allow us to interpret these recent observations with a new model for AGN disks, where the thicker disk leads to greater X-ray absorption, less X-ray reprocessing at larger disk radii, and a stronger signal from intrinsic variability driven by fluctuations in the AGN disk. In our simulations, we can measure the inflow timescale of these fluctuations as a long lag between light curves emitted from different disk radii. Understanding how disk structure impacts AGN variability and reverberation mapping can allow us to predict the thickness of AGN disks using UV-optical AGN light curves.

Recent 3D world models generate photorealistic, explorable scenes that remain frozen in time. OuroWorld is a mask-free framework that turns any static 3D Gaussian Splatting scene into a 3D cinemagraph: a dynamic scene with vivid, diverse motion looping seamlessly from any viewpoint. A vision-language model infers plausible dynamics and guides a video model to synthesize a reference video, which we lift and complete into multi-view videos. To learn from this imperfect supervision, we propose Inconsistency-Robust Periodic 4DGS: a Fourier-series deformation field guarantees looping by construction, while a Grounded Drift Field anchored at the reference view absorbs cross-view inconsistency. Unlike prior Eulerian methods limited to fluid-like motion, we capture general deformation, object motion, and illumination change. We introduce a ground-truth-free evaluation covering vividness, naturalness, loop seam coherence, and scene quality. On 39 reconstructed and generated scenes, OuroWorld outperforms all baselines and wins 70.8%-99.0% of user-study comparisons. Project page: https://ouroworld.userwei.com

It is conjectured that in the Lorentz mirror model on $\Z^2$ all trajectories are finite. Quas showed that, almost surely, the two halves of any infinite trajectory, cut at an edge, meet at infinitely many vertices, each of which is pivotal: changing its state can make the trajectory finite. We construct plane graphs on which infinite trajectories exist, and all of them have this property. Given $p\in(0,1)$, we replace each edge of a random one-ended subtree of $\Z^2$ by a number of parallel edges, depending on $p$, that is at most logarithmic in the height of the corresponding descendant tree. The medial graph of the resulting plane multigraph is a factor of the tree, and for the uniform spanning tree it has finite vertex intensity. In the mirror model on this medial graph, with probability $p$ for mirrors along primal edges and any probability $r\in(0,1-p]$ for mirrors along dual edges, infinite trajectories exist, and the two halves of each of them meet infinitely often. The construction rests on Bernoulli percolation: every random one-ended locally finite tree can be decorated in this way such that bond percolation with parameter $p$ is critical and percolates, and every vertex of the infinite cluster has infinitely many pivotal edges. We characterize this property for bond and site percolation and for mirrors.

Video generators and video-based world models can synthesize plausible visual trajectories, but long-horizon procedural tasks require generation to adapt to what has actually been produced. A model must determine the next action from its generated state, execute that action, and recognize when the task is complete. Open-loop generation cannot adapt to execution outcomes, while existing closed-loop systems often rely on pretrained executors or indirect verification. This leaves a gap between deciding an action and successfully realizing it. We formulate procedural video generation as closed-loop task execution in visual world space and introduce WorldGuide. Given only an initial image and a task goal, WorldGuide predicts an atomic action, generates its corresponding video clip, and uses the generated result to select the next action or terminate. The Planner and Executor are trained on the same step-level procedural demonstrations: the Planner learns to predict the next atomic action or task completion from visual progress, while the Executor is directly trained to realize the predicted actions. Hierarchical visual memory maintains state across long-horizon execution with bounded history token cost. Due to the lack of step-level action-video supervision for joint planner-executor training, we introduce WorldGuide Bench: approximately 59K step-annotated videos across 245 tasks and 27 procedural categories. WorldGuide achieves a 33.33% Task Success on WorldGuide-Bench, compared with 29.90% for the strong recent video model MiniMax-H3, even though MiniMax-H3 receives reference action plans, and achieves 47.69% on VideoCraft-Bench compared with 32.73% for MiniMax-H3 under goal-only conditioning. These results demonstrate the importance of coupling planning with learned execution for goal-directed procedural video generation.

Multimodal large language models (MLLMs) are rapidly evolving toward continuous audio–visual reasoning, creating an urgent need for evaluations that expose their capability limits. Audio–visual captioning is an ideal diagnostic task, yet current benchmarks face a coupled trade-off: whole-caption scores provide coverage without localization, local probes provide localization without coverage, and unconstrained LLM judges introduce instability. We introduce OmniCapBench (Omni-Video Caption Benchmark), a benchmark that reframes audio–visual caption evaluation as a deep-structured diagnostic framework. OmniCapBench shifts the prediction target from free-form text to sets of atomic, verifiable evaluation units across three tracks: entity references, visual shots, and audio events, enabling reliable scoring with deterministic constraint checks and localized LLM-based semantic comparisons. With 786 densely annotated videos, OmniCapBench effectively distinguishes MLLM perception errors, including temporal grounding failures, identity drift, cross-modal misalignment, and hallucinated descriptions. Evaluating frontier MLLMs reveals strong local perception but weak long-horizon audio–visual reasoning, particularly in identity drift and cross-modal misalignment, providing a fine-grained roadmap for omnimodal development.