Papers

Assistants from 2011 answered and waited, and agents from 2023 did a task and stopped. In September 2026 Meta's Muse showed an agent for one person, with accounts, devices, memory and a conversation that lasts, closed, in a vendor's cloud, in one country. Such an agent is expected to act on a person's accounts and devices, remember them across weeks, speak first when it is worth it, and answer for what it did. It is a kind of software, not a model, and until now had no open counterpart. This report defines the personal agent in five questions and three horizons. It reads how Muse is built from Meta's public record and a copy of its production prompt, each statement marked by its source. It then presents nanoMuse, the open-source counterpart under the GPL-3.0, one agent on every device a person owns, with hands on the phone's screen and the computer's. They share one conversation over a relay anyone can run; every action goes through a Sentinel, memory is files the person can read, and the model is their choice. Its size and cost are given as estimates. What is open, memory with provenance, an evaluation suite for the hands and an open model for them, is set out as a roadmap.

Large-model-based point-to-CAD generation holds immense potential for advancing industrial design and enhancing 3D modeling efficiency. However, most existing methods approach the problem as a general point-cloud encoding and token prediction task, neglecting the tokenization and supervision specifically for CAD-related primitives. As a result, these methods often struggle to accurately reconstruct the intricate primitive structures. To address this limitation, we propose PrimitiveCAD, a novel multi-stage paradigm for point-to-CAD reconstruction that enhances the geometric accuracy of generated CAD models while better preserving critical geometric features. First, we introduce a primitive-aware point cloud tokenization model, enabling the system to learn more robust geometric representations from CAD point clouds. Next, we perform supervised finetuning on a large language model (LLM) and introduce an operation alignment loss to align key CAD operation frequencies, thereby improving the preservation of global shape features. Finally, we incorporate reinforcement learning (RL) and introduce a feature-line alignment reward to further reduce stochasticity and enhance the fine-grained preservation of geometric features. Experiments on the DeepCAD and Fusion360 datasets show that our method achieves state-of-the-art performance in code validity, geometric accuracy, and geometric feature preservation.

In this contribution, we address the exact linearization of forward- and backward-flat discrete-time nonlinear systems with an arbitrary number of inputs. Building on structural properties of the flat parameterization, we show that forward-flat systems can be rendered static feedback linearizable by prolongations of suitably transformed inputs. Analogously, for backward-flat systems, we derive a constructive procedure showing that an exact linearization can be achieved by prelongations of suitably chosen functions of the system variables. In this way, we extend previously established results for two-input systems to the general m-input case. Finally, we illustrate the proposed linearization procedures by an academic example and a discrete-time model of a quadrotor.

Integrated sensing and communication (ISAC) is a key technology for next generation wireless networks, requiring highly sensitive, compact receiver front ends with multiband coverage. Rydberg atomic receivers (RARs) provide a promising architecture that satisfies these requirements owing to their high sensitivity and continuous frequency tuning capability. Unlike in classical ISAC systems utilizing orthogonal resource allocation, executing concurrent multi band tasks within a single atomic ensemble relies on shared energy state transitions. Owing to this shared structure, the communication and sensing tradeoff in RAR-based ISAC introduces an atomic-response tradeoff beyond the classical subspace and deterministic-random tradeoffs. This paper studies a multi-branch RAR (MB-RAR) architecture for joint communication and radar sensing. By solving the five level Lindblad master equation at the local oscillator (LO) biased operating point and linearizing the communication and sensing radio frequency (RF) signals as small perturbations, we obtain the double-sideband dynamic signal response of the MB-RAR. The model reveals two intrinsic constraints. First, increasing the coupling laser strength broadens the response bandwidth but can weaken the atomic response. Second, strengthening one branch's LO can suppress the other branch's conversion gain through the shared Rydberg state, even when the latter's LO bias remains fixed, which is termed the inter band gain inhibition (IBGI). Based on these mechanisms, we formulate the achievable Cramér–Rao bound (CRB)-rate region and develop a rate-constrained joint search over coupling strength and LO power allocation. Numerical results validate the proposed response model and show that joint optimization provides a more complete characterization of the achievable CRB-rate region.

Flying a quadrotor through a cluttered environment requires not only planning a collision-free reference trajectory based on perceived obstacles, but the reference also needs to be dynamically feasible and within the actuation limits of the vehicle, so that the controller can track it precisely. Existing methods either optimize a smooth polynomial inside a convex corridor, which limits agility, or treat obstacles as soft costs traded against tracking performance. We propose a Nonlinear Model Predictive Planning (NMPP) that imposes perceived obstacles as hard geometric constraints and hands a full-state reference to an obstacle-blind SE(3) controller. Our planner achieves a 58-67 % lower position RMSE than a linear Model Predictive Control trajectory planner and a 41-70 % lower RMSE than a polynomial trajectory planner. It also completes all forest flights with up to 9.5 m/s speed without collisions, and achieves 86 % flight success rate under a more aggressive speed profile where a state-of-the-art planner has only 26 % success rate. The real-world deployment showed reliable execution flying up to 5.5 m/s in an unknown cluttered environment.

Gene expression is widely measured in biomedicine, yet clinical outcome prediction remains challenging due to high dimensionality, strong feature correlations, and limited labeled data. Large self-supervised transcriptomic foundation models often fail to outperform simple supervised baselines. Tabular foundation models offer an alternative through in-context learning, but are typically pretrained on generic synthetic data rather than transcriptomic structure. We ask whether transcriptomics-aware pretraining, rather than scale, is the missing ingredient. Towards this end, we introduce GeneICL, a 4.2M-parameter tabular foundation model combining a semi-synthetic pretraining prior built from measured bulk expression profiles with a parameter-efficient recurrent architecture. We further enable right-censored survival prediction via a training-free reduction to regression using Cox partial-likelihood residuals. We evaluate GeneICL on 80 clinical outcome-prediction tasks spanning classification, regression, and survival. Tabular foundation models consistently outperform self-supervised transcriptomic models, while GeneICL achieves the best overall rank among evaluated foundation models and tuned baselines. GeneICL does so with up to 387$\times$ fewer parameters, no gradient updates at inference, and predictions within seconds on a laptop CPU.

Accurate numerical simulation of nonlinear partial differential equation initial value problems typically requires high-dimensional full-order discretizations at substantial computational cost. While purely data-driven surrogates can speed up prediction, these "black box" approximations do not necessarily satisfy the governing equations during inference. In this paper we present a hybrid framework, which combines neural operators with projection-based reduced-order modeling for nonlinear PDE simulation. The key idea is to train a neural operator to map the initial condition to a time-dependent reduced subspace, and then evolve low-dimensional reduced coordinates by solving the governing dynamical equations projected on this learned moving trial subspace. In this way, the network predicts an adaptive reduced representation rather than the full trajectory directly, while the online solver preserves a physics-based reduced evolution. We test this method on two representative nonlinear PDEs: the viscous Burgers' equation and the Fisher–KPP reaction–diffusion equation. In both cases, the proposed approach achieves successful reduced-order simulation with substantial wall-clock speedup relative to the corresponding full-order solver while maintaining relative errors on the order of \(10^{-2}\) to \(10^{-1}\) across the tested resolutions. These results demonstrate that a low-dimensional learned time-dependent reduced subspace can be used to predict solutions to nonlinear PDEs while retaining the advantages of projection-based reduced dynamics. Overall, this method provides a promising framework for fast and physically grounded data-assisted simulation of nonlinear PDEs.

In September 2026 the U.S. Department of War released an infrared video recorded from a military aircraft over the Middle East, with a mission report describing six small spherical objects moving at an estimated 480 miles per hour. We measure the brightness of each object in every frame. All six brighten and dim steadily at 7.1-7.9 times per second, each at its own rate and out of step with the others. Tests with planted constant-brightness objects show that neither compression nor the sensor produces this flicker. The same analysis, applied to the Galileo Project's ground-based infrared cameras near Las Vegas, finds flying animals beating steadily at 3.4-10.0 times per second, within the range of bird wingbeats. A nearly stationary object at 60-80% of the aircraft's height would produce the reported ground-projected speed of 480 miles per hour. The six objects are best explained as a small flock of birds flying nearer to the aircraft than the crew assumed. The same analysis also finds 19 objects with similar beats, at 2.3-11.0 times per second, in nine other released military videos. Releasing the redacted sensor and platform data would provide information that could refute the bird hypothesis

Large language model agents are accelerating scientific automation, yet verified executions rarely become persistent program-level improvements, and existing evaluations do not examine this process across sequential tasks in both the natural and social sciences. We formalize ScienceClaw as fixed-parameter program self-evolution that unifies task solving, scientific verification, and program updates. ScienceClaw-Eval spans 23 disciplines and measures scientific correctness, evolutionary gain, retention, cross-dataset transfer, and evolution cost through sequential streams and independent reset evaluation. Our framework repairs executable workflows through multi-turn interaction, converts re-execution-verified failure–success trajectories into linked Skill and Operator candidates, and retains an update only when source-task replay reproduces the repair and independent scientific tasks improve. Code is available at https://github.com/beita6969/ScienceClaw.

Let $G$ be a finite nonabelian group of odd order, let $p$ be the smallest prime divisor of $|G|$, and let $V$ be a finite faithful completely reducible $G$-module, possibly of mixed characteristic. Suppose that $M$ is the largest orbit size in the action of $G$ on $V$. It is known that $|G:G'|\le M$, and that equality holds only for abelian groups if $|G|$ is odd. We prove that $|G:G'|\le M/p$, unless $p=3$ and $G$ is the direct product of an abelian group and a group isomorphic to $\Gamma(2^3)$ or to $\Gamma(2^3)\times C_3$ acting on $V$ in a specific way, in which case $|G:G'|=3M/7$. In particular, $|G:G'|\le 3M/7$ for every nonabelian group of odd order. If $|V|$ is odd, then $|G:G'|=M/p$ holds if and only if $|G'|=p$; more precisely, nilpotent groups then have regular orbits, whereas non-nilpotent groups satisfy $|G:G'|\le 3M/13$, and even $|G:G'|<M/(2p)$ if $p\ge5$. The bound $3M/13$ is best possible.

Counterfactual inference in Gaussian-process structural causal models (GP-SCMs) has been developed primarily for continuous endogenous variables, limiting applicability to causal graphs that contain discrete child nodes with continuous parents. We introduce a unified probabilistic framework for counterfactual inference with heterogeneous variable types by pairing GP predictors with explicit exogenous noise mechanisms. For discrete outcomes, we derive exact conditional noise-abduction procedures using a uniform threshold for binary variables, a Gumbel-max race for nominal categories, and a latent Gaussian cut-point model for ordinal ones. In each case, we propagate abducted noise through interventions while accounting for posterior uncertainty in the GP latent functions, and prove that the resulting mechanisms reproduce the fitted model's observational and interventional distributions. On synthetic SCMs with known ground-truth counterfactuals, we evaluate estimation accuracy, consistency, and robustness to coupling misspecification. A key finding is that applying a categorical coupling to ordinal data inflates counterfactual error roughly threefold even when observational fit remains comparable, and that this error does not diminish with more data. As the training set grows, the fitted structural equation converges to the truth while the counterfactual error flattens onto a floor. In the reverse direction, forcing a false order onto nominal data instead degrades the fitted equation itself. The choice of coupling must therefore be justified on structural grounds rather than read off the fit.

This work presents a framework for accelerating transformer-based language models (LMs) on resource-constrained IoT devices. The framework targets compact LMs: BERT-Tiny (B-Ty), MobileBERT (M-Bt), MiniLM (M-Lm), Electra (E-Lt) and DeBERTa (D-Bt) – selected for their architectural diversity and use in edge inference scenarios. The proposed flow derives lightweight instruction set extensions tailored to the non-obvious computational patterns of these models. In addition, a custom instruction is introduced to accelerate the address generation stage of batch matrix multiplication, achieving a performance improvement of 15.39–21.74% with modest ASIC overheads of 6.79% in area and 2.33% in power. To further enhance performance without incurring additional processor core hardware cost, an optional compiler-directed loop unrolling strategy is employed, trading increased code size for overall reduced execution time. Evaluation on the Synopsys trv32p3f RISC-V core demonstrates inference speedups of up to 2.19x, and FPGA implementation on the AMD Zynq UltraScale+ ZCU102 shows a 32.94% area overhead at 75 MHz, whereas the ASIC implementation using the TSMC 28 nm library incurs a 21.08% area overhead while operating at 250 MHz.

Uniaxial ferroelectrics have long served as benchmark systems for time-resolved studies of polarization evolution due to their simple 180-domain structure. However, recent experiments in these materials have revealed complex topological textures with enigmatic features, including the unexpected charge-free head-to-head and tail-to-tail domain walls which still wait for their consistent explanation. Here, we use a universal three-dimensional phase-field model to investigate the kinetics of domain structure development in triglycine sulfate, accounting for the monoclinic symmetry of its parent and ferroelectric phases. We show that isolated saddle-like polarization structures, interconnected saddle cascades, saddle lines and specific monkey saddles can form only when transverse degrees of freedom for spontaneous polarization are allowed in the polar plane of a uniaxial ferroelectric. The abundance and diversity of saddle-like structures at the early stages of domain ordering drives a far more effective compensation of bound charges at head-to-head and tail-to-tail domain walls than those predicted by single-component order parameter models. As domains coarsen and residual bound charges migrate toward the surfaces of the sample, the topological landscape simplifies, leaving behind only isolated saddle points. The present model successfully reproduces exotic structures observed experimentally in triglycine sulfate and lead germanate and reveals a domain topology that extends beyond the conventional Ising picture of uniaxial ferroelectrics. Although the mechanisms of bound-charge compensation are likely to be material-specific, similar topological structures may emerge in other uniaxial ferroelectrics. The general formulation of our phase-field model opens the way for future studies of such phenomena across a broader class of ferroelectric materials of arbitrary symmetry.

We show that if an open smooth $6$-manifold $M^6$ is the interior of a compact, contractible manifold with boundary $X^6$, such that $\pi_3(X,\partial X)=0$, and $M$ supports a complete metric with uniform positive BiRicci curvature and Ricci curvature bounded below, then $M$ is diffeomorphic to $\mathbb{R}^6$. Additionally, we show that if an open smooth $n$-manifold $M^n$, $n\in\{6,7\}$, is the interior of a compact, contractible manifold with boundary $X^n$, such that $\pi_3(X,\partial X)=0$, and $M$ supports a complete metric with positive BiRicci curvature with at most $C$-quadratic decay at infinity for some $C>C_n$ and nonnegative Ricci curvature, then $M$ is diffeomorphic to $\mathbb{R}^n$. Lastly, for each $n\geq6$, we construct a compact, contractible Newman $n$-manifold, $N^n$, based on a presentation of the Higman group such that $\pi_i(N,\partial N)=0$, $3\leq i \leq n-3$, and $\mathrm{int}(N)$ admits a complete metric with uniform positive scalar curvature and Ricci curvature bounded below.

Motivated by observations in atmospheric modelling, we investigate the effect of “stochastic inertia”: in certain dynamical systems, small noise induces trajectories to stay longer in a dynamically critical subset of state space than in the purely deterministic situation. Under mild general assumptions on the stochastic differential equation, we derive an integral formula along deterministic trajectories whose sign provides a sufficient condition for the occurrence of stochastic inertia. Excluding situations with open sets of infinite escape times and assuming uniformly elliptic diffusion, we additionally identify this integral formula as the zero noise derivative of the average escape time. The proofs make crucial use of Dynkin's formula and suitable adaptations of Freidlin–Wentzell theory. We present an algorithmic approach to compute the integral formula and, by examples of saddle points and relaxation oscillators, demonstrate the following heuristic: stochastic inertia appears in the presence of an unstable direction that predominantly leads the escape dynamics.

Modern video models can generate realistic videos from real appearance references and proxy renders that specify scene structure, viewpoint changes, and motion. Evaluating this render-to-real capability requires a real target video depicting the same scene evolution, paired with an editable, geometrically registered 3D replica. Such data has traditionally required substantial manual modeling, calibration, and animation effort. We introduce RenderBench, a benchmark of 12 reconstructed real-world scenes spanning large-scale indoor environments and egocentric viewpoints, with both static and dynamic settings. Our construction pipeline combines visual geometry, neural reconstruction, and assisted 3D authoring. Each scene is decomposed into static objects and dynamic actors, registered to the capture cameras, and accepted only after multi-view geometric and temporal validation. Each evaluation unit contains appearance reference images, a held-out real target video, an editable digital twin, a matched proxy render, and renderer-native scene annotations. We evaluate transfer models against paired real target videos, retain PAI-Bench-C-compatible structural projections, and use scene annotations to localize failures by object, visibility, articulation, and motion. The first release retains 12 of 14 registered samples (85.7%), comprising 1,496 paired real-proxy frames. All released scenes pass file-integrity and environment-edit audits, while proxy diagnostics yield a depth si-RMSE of 0.2170 and instance mIoU of 0.3673. RenderBench provides paired real observations and editable scene state for assessing both appearance fidelity and preservation of geometry and dynamics.

Full-duplex speech models are trained to converse with a person, but they are increasingly made to converse with each other, in self-play data generation, agent societies, and model-based evaluation. In that loop no human absorbs a timing error: each model's turn-taking is the other's input. We ask what timing the loop settles into. Two PersonaPlex-7B instances exchange audio tokens on a shared clock in unscripted conversation, and one floor-transfer rule is applied to them and to Switchboard. Their timing is coupled: re-pairing speakers across conversations destroys it. But the floor changes hands late, at a median of 400-560 ms against 137 ms for humans, and the last 120 ms of the partner's turn, where human projection places a tenth of its transfers, holds 1% of theirs. Delaying one direction of the channel shifts the response one-for-one and leaves the run-up to it empty, consistent with a reactive wait after the perceived end rather than the turn-end projection human timing requires.

We solve a longstanding problem in Bernoulli percolation: we derive sharp Ornstein–Zernike asymptotics for the probability that two distant points are connected by a finite cluster in the whole supercritical regime. Our proof has two main ingredients. First, the renewal structure from Fridbergh and Hammond (2024) yields an a priori bound on cluster volume. Second, we adapt the coarse-graining and irreducible-cluster framework of Companion, Ioffe, Velenik (2008) to finite supercritical connections. The main distinction from the subcritical analysis is that our argument does not require a separation of masses between irreducible and unrestricted finite connections. The volume bound and a modified coarse-graining construction instead provide stretched-exponential tails for the effective random-walk increments, which suffice to obtain the sharp asymptotics.

Visualization research has long examined embellishment and animation, while the design of rich interaction-contingent feedback remains under-articulated as a systematic design space. We introduce JuicyVIS, a theory-informed operationalization of juicy feedback for interactive visualization that translates ideas from game studies into a visualization-centered framework. JuicyVIS introduces three dimensions, interaction type, feedback timing, and feedback intensity, which we explore through 26 controlled prototypes, grounded in established visualization interaction categories and prior work on juicy feedback in games. We evaluate the juicy design space through three exploratory mixed-methods online studies. Results show that juicy feedback was consistently associated with higher engagement and aesthetic experience, while the clearest gains came from post-interaction feedback, not, for example, by simply increasing intensity. We discuss how juicy feedback can be thought of as a design material in visualizations, shaping how visualization interactions feel, how action consequences are perceived, and how users remain engaged with data representations.

Endpoint accuracy on abstract-reasoning benchmarks does not reveal whether a language model has acquired a transferable rule or fit distribution-specific regularities. We study this distinction in small language models on the ARC-TGI benchmark, which organizes abstract grid transformations into controllable task families and supports resampling, spatial shifts, and cross-benchmark transfer. Across more than 1,000 runs, we profile decoder-only, encoder–decoder, and mixture-of-experts model families under supervised fine-tuning. We examine the efficiency and stability of skill acquisition, robustness beyond the training distribution, interactions with model family and task formulation, and layer-wise attention signatures that accompany behavioral differences. Substantial in-distribution accuracy is attainable, but acquisition is sensitive to optimization and unevenly distributed across task families. Performance deteriorates sharply outside the training distribution, including when the rule is retained but grid scale changes. Greater training-set depth and breadth yield uneven gains, while the effect of additional in-context examples depends on model family. Executable-rule induction also yields correct solutions not observed under direct grid generation. On selected tasks, attention diagnostics show distinct concentration and context-dependence profiles, but do not establish general causal mechanisms. Overall, abstract-reasoning scores are conditional on the model, adaptation regime, evaluation distribution, and response format.

We first prove a rigidity result for uniformly rotating vortex patch solutions to Euler equations in bounded simply-connected planar domains. Namely, if a bounded simply-connected domain with real-analytic boundary contains a nontrivial vortex patch that remains strictly inside the domain while undergoing a uniform rigid rotation, then the ambient domain must be a disc centered at the rotation center. The proof combines the rotating-frame formulation with harmonic unique continuation and the generalized Schwarz's reflection principle across analytic boundary arcs. These tools imply that, unless the outer domain is rotationally invariant, the velocity must vanish in an open region surrounding the patch, contradicting the nonzero circulation generated by the patch. In the same spirit, we show that only circular domains possess a circular point vortex orbit. As a second result, we construct quasi-periodic vortex patch solutions to the Euler equations in domains that are small perturbations of the unit disc $\mathbb{D}$. The amplitudes of the solutions are linked to the size of the domain deformation, allowing the classical Rankine vortices $b\cdot\mathbb{D}$, $b\in(0,1)$, to be regarded as equilibrium states. Our analysis is perturbative with respect to the construction developed in the undeformed case, and the resulting quasi-periodic solutions exist for most values of the inner radius $b$. This provides the first construction of quasi-periodic vortex patch solutions to the Euler equations in non-radial domains. Moreover, combined with the rigidity result established above, this shows that, in the periodic case, these solutions cannot arise from a purely rigid rotation.

Speculative decoding accelerates inference for a large language model (LLM), referred to as the target model, by first using a smaller model, referred to as the draft model, to generate candidate tokens and then verifying them with the target model for acceptance or rejection. Prior studies primarily focused on the efficiency-utility trade-off of speculative decoding, e.g., lossy speculative decoding, leaving its security implications largely unexplored. In this work, we bridge this gap by providing the first systematic study of the security implications of speculative decoding. Through a large-scale measurement study, we reveal a pronounced security-utility asymmetry: across a wide range of lossy speculative decoding methods, improvements in inference efficiency come at a disproportionately high cost to security, with attack success rates for jailbreak and prompt injection attacks increasing much faster than utility degrades. We then propose SecureSD, a new theory-guided speculative decoding method that enhances security while maintaining efficiency and utility. Specifically, our theoretical analysis reveals that security degradation primarily originates from the early tokens generated by the draft model. Motivated by this insight, SecureSD applies a stricter verification criterion to draft-model tokens at early decoding positions. Extensive experiments on both security and utility benchmarks demonstrate that SecureSD significantly improves security while preserving efficiency and utility compared to existing speculative decoding methods.

Off-policy evaluation (OPE) for contextual bandit policies becomes challenging when action-level importance weighting incurs excessive variance. Doubly robust (DR) estimation remains unbiased under common support but retains these high-variance action-level weights. A prior estimator, Off-policy evaluation with Conjunct Effect Model (OffCEM), replaces them with more stable cluster-level weights, at the cost of relying on local correctness of the reward model. In this paper, we show that, under the assumptions required by DR and OffCEM, there exists an unbiased family of estimators that interpolates between OffCEM and DR. Building on this result, we propose the Variance Optimal-CEM (VOCEM) estimator, which selects the interpolation coefficient to minimize variance. We derive the population-optimal coefficient in closed form and show that the resulting estimator has variance no larger than either endpoint, OffCEM or DR. Experiments in controlled synthetic settings and on two large-action benchmarks show that VOCEM improves upon both endpoints in all 23 evaluated conditions, exhibiting greater stability and empirical robustness.

In this paper we consider matrices whose entries are rational functions of several parameters over a Moore-Penrose field, that is, a field with an involutory automorphism over which every matrix has Moore-Penrose inverse. We prove that over any such field the Penrose conditions can be solved with linear algebra alone, even though they form a polynomial system of degree two. We then bound the degrees of the numerator and of the denominator of the pseudoinverse in terms of the degree of the entries and of the rank, and not of the dimensions of the matrix. Specialization is the thread: the closed form is computed once over the field of rational functions, and the question is at which parameter values it still returns the pseudoinverse of the specialized matrix. We determine the values where it does not, from the matrix alone and before computing the pseudoinverse. They are the zeros of a polynomial built from the maximal minors, and they are the values at which the rank decreases. This turns previous sufficient conditions into an exact characterization. The theory also gives symbolic algorithms for the real, the complex and the parametric case, which we implement in Maple and test on 972 timed replications. We also apply them to a Leontief economic model, where the excluded value is the point at which the economy stops being viable.

Financial reports repeat values across periods, metrics and accounting lines, allowing an LLM-generated calculation to be numerically correct while citing the wrong financial role. We evaluate what probabilistic evidence verification adds beyond number matching using Jev as a source-support verifier for GPT-4.1-mini calculation traces. A signed-number-at-pointer baseline explains most recovery over exact quotation checks. To isolate the remaining role-recognition problem, we hold operands and arithmetic fixed, move citations between same-number cells, and retain controls that express equivalent facts. These contrasts reveal both wrong-role citations that pass and valid alternative citations that are withheld. Explicit column labels improve selected wrong-role decisions while also lowering support for some equivalent evidence. A constructed follow-up on 36 new source pages, labeled by a non-author reviewer, extends this evaluation and exposes the same tradeoff between detecting role errors and retaining valid citations. The contribution is a controlled evaluation that identifies what a probabilistic financial verifier distinguishes when numerical matching is held fixed. For LLM-based financial assistants, it makes numerical correctness, cited-role support and acceptance outcomes separately assessable.

Current large video-language models (LVLMs) still face challenges when dealing with long videos, mainly because frames are often processed independently, making it difficult to capture temporal dependencies across events. Although retrieval-augmented approaches have been introduced to provide additional context, most of them operate at the frame or snippet level, which limits their ability to model how events evolve over time and relate to each other. In this paper, we propose Event Chain Retrieval-Augmented Generation (EC-RAG), a training-free framework that organizes video content into an explicit event chain before question answering. Instead of retrieving isolated frames or text segments, EC-RAG first partitions the video into semantically coherent segments, represents each segment using multi-modal signals, and then links them into a structured chain that preserves temporal order and captures inter-event relationships. Given a query, the system identifies relevant events within this chain and gathers supporting evidence from the associated modalities. Our approach offers several practical advantages: (i) event-level abstraction that better reflects how video content is naturally structured, enabling more reliable localization compared to frame-level retrieval; (ii) structured multi-modal fusion that aggregates speech, text, and visual cues at the event level, allowing complementary information to be more effectively utilized during reasoning; and (iii) plug-and-play compatibility with existing LVLM backbones, requiring no additional training or reliance on proprietary models. Experiments on Video-MME, MLVU, and LongVideoBench show that this event-centric design consistently outperforms frame-level retrieval baselines, highlighting the importance of modeling temporal structure for long-video understanding.

We discuss 2-dimensional $QCD$ with a fundamental quark of mass $m$ in the 't Hooft limit ($N \to \infty, \, g^2 \to 0, \, g^2N =$ const). The spectrum of this theory includes a chain of meson states. The masses of these states can be evaluated by solving the proper integral Bethe-Salpeter equation. An alternative simple procedure consists in solving the Schrödinger equation for the relativistic Hamiltonian $$ \hat H\ =\ 2\sqrt{\hat p^2 + m^2} + \sigma |x| $$ with the string tension $\sigma = g^2N/4$. We compare the results obtained by these two methods. In the leading semiclassical approximation, they coincide. We show, however, that the preasymptotic WKB corrections to the masses calculated in these two ways are different. For the ground state, they play an essential role. But already for $n=3$, the relative difference between the two predictions does not exceed 2.5% and rapidly falls off as $n$ increases.

Mesh-agnostic facial animation retargeting transfers expressions across meshes with different structures, but preserving facial motion without surface artifacts remains challenging. To address this, we present PDB, Point-Based Deformation Blending for facial animation retargeting. PDB predicts a compact set of deformed control points from a source neutral-expression pair and blending weights from the target neutral mesh. The weights are computed once per target and reused across frames, while the control points vary with each source expression. ReLU enforces non-negative weights and permits exact zeros, followed by row-wise normalization. The target mesh is reconstructed directly by multiplying the weights and control points, without a predefined cage, precomputed coordinates, a learned per-element deformation decoder, or a global reconstruction solve. Trained only with self-retargeting reconstruction supervision, PDB supports cross-identity transfer without paired cross-identity training expressions. Experiments demonstrate accurate retargeting, fast inference, and localized support in the learned weights. Joint evaluation of expression accuracy and local surface preservation shows reduced surface artifacts relative to the evaluated dense displacement method while retaining the intended motion. Perceptual evaluations further support expression fidelity and visual quality in both self- and cross-retargeting.

Parity-violating initial conditions are uniquely compelling signatures of inflationary dynamics beyond the single-field slow-roll paradigm, and their effects on the distribution of late-time observables can be easily separated from that of gravitational nonlinearities and galaxy formation physics. In this work, we propose using the momentum field, $\mathbf{p} = (1+\delta)\mathbf{v}$, and, in particular, the density-density-momentum bispectrum as a novel probe of primordial parity violation. This class of bispectra greatly complements traditional searches using four-point statistics of the density field, furnishing new observables that can be sensitive to parity-violating physics. At the same time, its lower dimensionality substantially reduces the complexity associated with analysing high-dimensional data vectors and their covariance. We show that the sensitivity of the density-density-momentum bispectrum to parity violation effects persists even if we can only observe the projection of the momentum field along the line-of-sight. We perform a Fisher forecast for the parity-violating amplitude of the ultra-squeezed template in the cross bispectrum of kinetic Sunyaev-Zel'dovich (kSZ) anisotropies and galaxy clustering, and we find that our approach give competitive constraints when compared to existing methods. Our method can be easily co-opted to existing pipelines to study parity violation and, as the galaxy-galaxy-kSZ bispectrum is a key cosmological probe, take advantage of the significant existing efforts dedicated to mitigating systematics and covariance modelling.

An agent that can say an action is wrong and then take it anyway is a different failure from one that does not know better, and evaluations of stated values cannot see it. We ask how often open models act against their own moral judgment under pressure, and what decides and reduces it. We build a pre-registered panel of 248 scenarios across five kinds of pressure. Each scenario is posed to one model in two separate calls, as the agent choosing among fixed options and in the third person asking which is right; the model's own judgment is the reference, and the scenario's construction fixes which option counts as the violation. Every scenario has a twin with the pressure removed, and every model gets a positive control that bounds what the instrument can see. On OLMo-3-7B-Instruct, the model takes the action it judged wrong on one in five of the scenarios screened on its own actions, more often than with the pressure removed. Whether the gap appears depends on the post-training recipe: OLMo-3 and Meta's Llama-3.1-8B-Instruct carry it; Tulu 3, Qwen2.5-7B-Instruct and OpenAI's GPT-OSS-20B, read without reasoning, show none within stated detection bars. Meta's recipe and Ai2's Tulu 3 share the same Llama-3.1 weights; only Meta's carries the gap. On both models that carry it, reasoning about the stakes before acting moves the choice back toward the model's own judgment. What that deliberation tracks is the stated rule: where following a written rule harms an absent third party, it still moves OLMo-3 and GPT-OSS-20B toward the rule, not detectably diminished, while Llama-3.1's brake shrinks to about a seventh. Whether a model acts on its own moral judgment is therefore not settled by pretraining, and deliberation moves the action toward the stated rule, which protects people only when the rule does. The gap is a measurable target for post-training recipes, not a fixed property of the model.