Papers

Construction environments evolve continuously, causing large geometric changes that degrade static mapping and registration performance. This necessitates active perception, where robots deliberately select sensing configurations to resolve the environment's current state. We present DeltaSeek, an initial framework toward active perception in evolving built environments. While our broader objective is a system that reasons about where, how, and when to observe, this paper addresses a critical prerequisite: how a robot's sensing embodiment constrains the observations it can acquire. We formalize an embodiment's permissible observation set and evaluate with a Husky A300 equipped with a UR5e on an IFC-derived benchmark under chassis-mounted and wrist-mounted RGB-D configurations, scoring observations by geometric visibility and effort by drivable distance. In a room-scale scene with eight controlled changes spanning four observability conditions, exhaustive evaluation over 240 permissible base poses and five arm postures shows that two changes admit no chassis viewpoint whatsoever, while the wrist camera resolves both. For changes observed by both embodiments, the median base travel is $6.0$~m for the wrist camera and $15.2$~m for the chassis camera. These results distinguish sensing limitations from acquisition costs, clarifying whether an observation is impossible or simply requires more travel.

Weak-to-strong generalization (W2SG) occurs when a student trained on a teacher's predictions outperforms that teacher. We study when this happens under fully converged, ridgeless two-stage learning, with no early stopping, no explicit regularization, and no assumption that the student is more expressive than the teacher. In two-stage linear regression, a teacher is fit from $n$ labeled examples and a student is trained solely on the teacher's predictions on $m$ fresh, unlabeled inputs. Although both stages share the same hypothesis class and the same training rule, we show that the student outperforms the teacher exactly when $m$ lies in an explicit intermediate range: too few pseudo-labels leave the student without enough signal, too many let it inherit the teacher's noise. Under power-law covariance, we derive this range in closed form as a function of the spectral decay and noise level, including regimes where the improving region splits into two disjoint intervals of $m$. We then study a random-feature model in which the student has strictly more features than the teacher, and identify two regimes, again given by explicit thresholds: one where improvement occurs only for $m$ in a bounded interval, and one where it occurs only once the student width $N_S$ exceeds an explicit threshold. Both regimes are governed by a single "active-bottleneck" principle: whichever of $m$ or $N_S$ is scarcer controls how much teacher error is filtered out, while increasing the other resource only reduces estimation noise. Together, these results show that finite data and finite width can themselves regularize a two-stage learner, with no explicit mechanism doing so.

Diffusion models are now widely used in Bayesian inverse problems in imaging as priors, where latent diffusion models are often used for larger scale problems to keep the computational complexity and model-size manageable. Unfortunately, the auto-encoder based compression results in loss of spatial detail. In addition, the optimization is converted to a non-linear problem. In this paper, we introduce a posterior sampling algorithm customized for the pyramidal/cascaded architecture, which relies on a coarse to fine hierarchical strategy to generate images in the pixel domain. We present CLIMB-Flow which alternates between three steps: an end-point estimation from the current coarse and noisy image, data-consistent update of the clean image, and re-noising it back to the level the network expects. Together these steps sample the posterior at that scale using an approximate Gibbs sampling from two conditional distributions. Experiments on ImageNet, CelebA, AFHQ and fastMRI span inpainting, deblurring, super-resolution and accelerated MRI, with PSNR gains of 1.37-7.66 dB over the strongest competing method on CelebA and pixel-domain reconstruction up to 512x512.

Adversarial attacks have long posed a fundamental threat to machine learning systems. As multimodal large language models (MLLMs) rapidly evolve and become widely deployed, assessing their vulnerability to such attacks is essential for their safe use. In this work, we investigate whether a single adversarial image can consistently mislead diverse frontier MLLMs in black-box settings. We propose O-Attack, a highly transferable black-box attack framework. This framework builds on our insight that surrogate models contain a broad, high-level, cross-modally aligned semantic space. This space extends beyond final-layer outputs and provides multiple semantically consistent representations that remain underexploited by existing attacks. Within this space, O-Attack anchors aligned representations, progressively broadens semantic conditions, and optimizes perturbations through semantic consensus to promote consistent target alignment. By fully exploiting this space with the same surrogate models as M-Attack, O-Attack raises attack success rates on GPT-5.4 (29.1% to 77.2%), Claude-4.6 (42.8% to 81.6%), and Gemini-3.1 (38.2% to 80.9%). Extensive experiments across 24 MLLMs show that O-Attack outperforms six state-of-the-art methods in black-box transferability, with consistent effectiveness across prompts and improved efficiency and imperceptibility. This work exposes the practical safety risks posed by black-box adversarial attacks against frontier MLLMs, underscoring the need for more rigorous robustness evaluation and more effective defenses.

World-action models can jointly predict future visual observations and robot actions. However, discrepancies may exist between their visual predictions and the consequences implied by generated actions. We observe that WAMs can often generate visually plausible task-completion outcomes before producing action sequences that reliably achieve them. Consequently, we treat the WAM-generated visual prediction as a goal-conditioned visual proposal rather than a directly executable plan. We use a frozen action-conditioned world model to predict action-conditioned consequences and construct feedback based on consistency between the two future predictions and alignment with the terminal goal. Leveraging this feedback, we employ Flow Policy Optimization (FPO) to optimize the action head of the WAM. This framework avoids online robot interaction and additional training of task-specific reward models. Across four real-world UR5 manipulation tasks, our method increases the mean success rate from 43.4% to 75.1%, compared with 61.4% for $\pi_{0.5}$. These results show that cross-model prediction discrepancy can provide useful feedback for improving robot policies under the evaluated manipulation tasks. Website: https://imagine-to-achieve.github.io/

Tor is a widely used anonymity network that provides network privacy by routing client communications through a sequence of relays to their destinations. An adversary observing a single Tor relay cannot readily link clients to their destinations, as doing so requires identifying the other relays along the client circuit. Identifying these relays from traffic observed at the monitored relay alone is challenging: each relay carries traffic for many circuits simultaneously, and client circuits typically last only about 10 min, limiting observations of a given circuit. In contrast, Tor onion services use introduction circuits that remain fixed for 18 to 24 h. We develop an intersection attack that exploits this design, allowing an adaptive adversary observing one Tor relay at a time to identify the network location of an onion service. We evaluate the attack on the live Tor network against a self-operated onion service carrying genuine third-party traffic. Across nine end-to-end experiments, the attack reconstructs the complete Tor circuit in every run, with an estimated median time of 2.2 h. To mitigate this attack, we propose a mechanism that periodically rebuilds internal introduction circuits to limit evidence accumulation without changing publicly advertised information.

Guessing Random Additive Noise Decoding-Aided Macrosymbol (GRAND-AM) provides a coding-based solution for non-orthogonal multiple access (NOMA) systems, enabling joint multiuser detection and error correction. GRAND-AM outperforms prior methods in symbol error rates while providing hard-detection output. We introduce soft-output GRAND-AM (SOGRAND-AM), a macrosymbol-level joint multiuser detection and decoding framework which generates calibrated bitwise a posteriori error probabilities. SOGRAND-AM extends GRAND-AM by enabling soft-decision outer forward error correction decoders, transparently integrating with existing decoders. SOGRAND-AM achieves approximately 2 dB improvement in bit error rate at the output of the outer soft-input decoder compared to hard input with GRAND-AM.

For cocompact Fuchsian groups $\Gamma \leq \hbox{PSL}(2, \mathbb{R})$ we improve Selberg's bound for the error term of the counting function in the hyperbolic lattice counting problem, achieving an essentially optimal upper bound for the error term \begin{equation*} E(X;z, w) = O(X^{1/2+\varepsilon}) \end{equation*} for almost every pair of points $z,w$. We extend this pointwise result for cocompact lattices acting on the $n$-dimensional real hyperbolic space $\mathbb{H}^n$. Moreover, in the $2$-dimensional case we study the second moment of the error term of the counting function. We improve the upper bounds of Chamizo and Cherubini for almost all pairs $z,w$, but we disprove a conjecture of Phillips and Rudnick for the case of cocompact Fuchsian groups.

Optimization layers enable the incorporation of structured constraints and decision problems into learning systems. Training such systems requires differentiating through the embedded optimization problem, which can be challenging for general conic programs. We introduce dOPT, a solver-agnostic framework that, rather than differentiating the full conic formulation, reduces it at a computed primal-dual solution to an equality-constrained quadratic program that preserves the reference solution and its first-order sensitivity. The reduction captures the local first- and second-order conic geometry relevant to differentiation and remains well defined at singular configurations. Computing solution derivatives then requires a single symmetric linear solve, independently of the forward solver. We derive explicit reductions for convex NLPs, QPs, SOCPs, and SDPs. Numerical experiments validate the computed gradients and show favorable backward-pass scalability, with substantial speedups over existing differentiable conic optimization methods as problem size increases.

The SLGP package provides nonparametric estimation of fields of probability densities indexed by covariates, from unevenly sized samples with few or no replicates. A Spatial Logistic Gaussian Process applies a logistic density transformation to a finite-rank Gaussian process over a joint index-response domain, yielding a prior over conditional densities that adapts in location, shape, and modality. The contribution of this paper is the implementation: a finite-rank Random-Fourier-Feature representation that keeps the transformed process numerically tame, a log-concave likelihood that makes the fit a convex problem, and a family of quadrature shortcuts for the normalising integrals that turn a per-observation cost into a per-grid cost. Three estimation modes are available: MAP, Laplace, and full MCMC. We demonstrate the package throughout on the classical Fiji-Tonga earthquake catalogue, estimating how the distribution of hypocentre depths deforms across the region, and on a discrete-response variant.

Following a similar path as the recent construction of a complex structure on $S^6$ we construct another such family, disjoint from the known ones.

We quantize the charged sector of the electroweak theory in an external electromagnetic plane-wave background after spontaneous symmetry breaking, which we refer to as the electroweak Volkov-Proca theory, and analyze its global $U(1)$ symmetry and continuous residual translation subgroup together with the associated conserved charges and momentum components that label the states. We further derive the completeness and orthonormality relations for the Volkov-Proca states in Ritus form, and construct the off-shell and on-shell propagators, including their representation in terms of the scalar Volkov propagator. Finally, we derive the Källén-Lehmann spectral representation and the LSZ reduction formula for the Volkov-Proca states.

We study average consensus over slowly time-varying networks and its application to decentralized optimization. We introduce WAVE, a windowed Chebyshev method that limits the accumulated effect of network variation by restarting the recurrence after finite windows. If $\chi$ bounds the network condition number and successive communication operators satisfy $\|L_{k+1}-L_k\|\leq\beta$, WAVE reaches $\epsilon$-consensus in $O((\sqrt{\chi}+\min\lbrace\beta\chi^2,\chi\rbrace)\ln(e/\epsilon))$ communication rounds. This rate matches the fixed-network $\sqrt{\chi}$ dependence when $\beta=O(\chi^{-3/2})$ and smoothly interpolates across the full range of network variation. For piecewise-constant networks where each change is detected when it occurs and consecutive changes are at least $\tau$ rounds apart, WAVE reaches $\epsilon$-consensus in $O((\sqrt{\chi}+\chi/\tau)\ln(e/\epsilon))$ communication rounds. Finally, we use WAVE as the consensus step in an accelerated decentralized optimization method for $\alpha$-smooth convex local objectives with a $\mu$-strongly convex average. For sufficiently slow network variation, every agent obtains an $\epsilon$-solution to the global optimization problem after $\widetilde O(\sqrt{\alpha/\mu}\,\sqrt{\chi})$ communication rounds, matching the fixed-network dependence.

We present the Quantum Accelerator (QuAcc), an open-source workflow library for atomistic simulations with an emphasis on quantum-mechanical calculations. QuAcc provides predefined workflow recipes spanning first-principles electronic-structure methods, semiempirical and tight-binding approaches, classical potentials, and foundation machine-learned interatomic potentials (MLIPs). A central design feature of QuAcc is its separation of domain-specific scientific logic from the workflow engine used to orchestrate and execute calculations. Workflows are written as ordinary Python functions and can be executed with multiple supported workflow engines without modifying the underlying source code, lowering the barrier to developing and contributing new workflows. QuAcc also streamlines the evaluation of foundation MLIPs by providing a unified platform for generating ab initio reference calculations consistent with the model of interest, mitigating methodological drift when assessing model performance. Together, these features make QuAcc a flexible and accessible framework for atomistic simulation workflows that have become central to the current era of machine learning and artificial intelligence.

An agent harness controls how a language model accesses information, uses tools, preserves memory, and checks its work. Improving this software is costly when each evaluation requires a long interaction with an environment. We introduce Vestrum, a framework that turns failures in execution traces into scoped harness changes without training the task model. Its organizing overhypothesis is that tasks of a shared kind may exhibit recurring failures whose remedies transfer within that kind. Vestrum expresses failures as recognizable classes, proposes changes across verification, retrieval, decomposition, and knowledge synthesis, and screens their scope before evaluating them as a bundle. A persistent lessons file informs subsequent proposals. Across five settings and two baseline harnesses, the frozen harnesses improve held-out performance: UltraHorizon rises from 47.6 to 59.8 over GAM, Terminal-Bench 4 Hard from 63.7% to 70.3% of checks passed over Claude Code on eight held-out tasks at 1.03x test cost, and cell-type annotation agreement from 67.5% to 77.8% on held-out sections of one slide, alongside gains on LoCoMo and AMA-Bench. Across our searches, verification grounded in evidence helped both intermediate steps and final answers, at lower cost at intermediate steps, while critics asked to rebuild finished answers broke more than they repaired. On the three memory benchmarks, Vestrum also scores above the evaluated GEPA configurations in every paired evaluation.

High-energy QCD evolution at NLO suffers from large collinear logarithms that make the perturbative expansion of the Balitsky-Kovchegov (BK) equation unstable. We develop a systematic scheme-transformation framework using the Doi-Peliti formalism to relate projectile ($k^+$) and target ($k^-$) factorization schemes. The transformation of high-energy operators and the evolution kernel generates NLO terms that cancel the problematic double collinear logarithm. The resulting NLO BK equation is free of this instability, with only single collinear and double anti-collinear enhancements remaining. Numerical solutions show stable evolution up to $Y=10$ for both Golec-Biernat-Wusthoff and McLerran-Venugopalan initial conditions, providing a consistent path toward combining stable NLO evolution with rotated NLO impact factors.

We improve our earlier omega bounds for the error term in the Piltz divisor problem over number fields. Our proof uses a weighted extension of a counting theorem of Balasubramanian and Ramachandra and an extension of Lamzouri's kernel to apply the long resonance method. We also discuss applications of our results to lattice points in norm-form regions, invariant sublattices, abelian covers of curves, spectral counting, and representation growth of Heisenberg groups.

A unitary signing of a $d$-regular graph is a Hermitian adjacency matrix $M$ whose nonzero entries lie in $\{\pm1,\pm i\}$ and satisfies $M^2=dI$. Motivated by the work of Alon and Zheng on orthogonal and unitary signings of cube-like graphs, we introduce the $\Theta$-property for a generating set $S\subseteq\mathbb Z_2^n$: whenever three pairwise disjoint subsets of $S$ have the same sum, at least two of them have even size. We prove that every zero-free generating set with the $\Theta$-property gives a cube-like graph $Q_S$ admitting a unitary signing, and we give an explicit local formula for such a signing. For Sidon sets, the $\Theta$-property is also necessary, yielding a characterization of the Sidon cube-like graphs that admit unitary signings. In this setting there are exactly $2^{|S|-n}$ switching-equivalence classes of unitary signings, and we characterize when a unitary signing can be chosen to be orthogonal. The $\Theta$-property admits a linear-algebraic description in terms of the dependency space $\mathcal D$ of $S$: \[ |D_1\cap D_2| \equiv |D_1||D_2| \pmod2 \qquad (D_1,D_2\in\mathcal D). \] Equivalently, the map \[ D\longmapsto \binom{|D|}{2}\pmod2 \] is linear on $\mathcal D$. For the corresponding extremal set problem, in which zero is permitted, this formulation yields the sharp bound $|S|\leq 2n+1$ for generating sets with the $\Theta$-property. We give elementary constructions attaining this bound for every $n\geq3$. Under the additional Sidon condition, the same extremal value is attained for every $n\geq10$ using binary self-dual codes. The exact Sidon maximum is also determined for $3\leq n\leq9$.

Visual anomaly detectors identify deviations from known-normal data, but their anomaly signals may mix evidence of actual anomalies with benign visual variation. We investigate whether automated interpretability can augment visual anomaly detectors by identifying and intervening on different components of this signal. We decompose PatchCore nearest-normal residuals into sparse features using Sparse Autoencoders (SAEs), and provide high-activation and contrastive non-active examples to a Multimodal LLM, which describes each feature and labels it as anomaly, distractor, or uncertain. These labels guide interventions in the SAE hidden representation, where distractor features are suppressed and anomaly features amplified. The edited representation is then used to reconstruct patch embeddings, which are rescored with PatchCore. Across 40 categories from four benchmarks, applying both interventions jointly improves macro-average image-level AUROC from 0.8724 to 0.8857 on source data and from 0.8066 to 0.8210 under synthetic corruptions. On three additional RobustAD categories with real acquisition shifts, the same interventions improve AUROC from 0.8745 to 0.9056 on source data and from 0.6069 to 0.6599 under real acquisition shifts. Finally, individual feature interventions across all 43 categories show that the MLLM labels are aligned in aggregate with how features differently affect normal and anomalous images.

JWST has discovered an overabundance of UV-luminous galaxies compared to extrapolations from pre-JWST galaxy formation models. Distinguishing between proposed theoretical explanations requires constraining the halo masses of these systems, which can be probed through clustering. However, at $z>10$, low galaxy number densities render traditional clustering measurements challenging. We introduce a novel formalism that uses information about the local environments ($\lesssim$2 arcmin) of individual UV-bright galaxies at $z>10$ to test the galaxy-halo connection implied by theoretical models for the UVLF excess. We create mock JWST observations using (500 cMpc)$^3$ 21cmFAST simulations to forward-model the spatial distribution of fainter photometric candidates around UV-bright galaxies. We find that the distance to the nearest neighbor is sensitive to the stochasticity of the UV luminosity-halo mass relation, and can distinguish between models where UV luminosity is boosted uniformly in galaxies versus models with high stochasticity in UV luminosity at fixed halo mass. As a proof-of-concept, we apply our framework to observations around the two brightest confirmed galaxies at $z>10$, GN-z11 and JADES-GS-z14-0, and find that the lack of neighbors within $<0.5$ arcmin of these sources disfavors models where UV luminosity is boosted uniformly, suggesting that stochasticity is important for explaining their high UV luminosities. We forecast that distinguishing between these models at high significance requires deep imaging surveying $<1.1$ arcmin around $7$-$15$ bright galaxies ($M_{\rm UV}\lesssim-21$) at $z =10$-$12$, dropping to $\lesssim5$ at $z\sim14$, capable of detecting their neighbors down to $M_{\rm UV,lim}\approx-18$. These observations are within reach of JWST, Roman and Euclid surveys, providing a path to constraining the galaxy-halo connection at our redshift frontier.

On-policy distillation (OPD) bridges teacher supervision and student behavior, but different teacher-student tokenizers introduce misalignment in both input tokenization (#1) and output logits (#2). Existing approaches address the former by matching same-text spans or converting tokens to bytes, often losing fine-grained token information or disrupting the native-token paradigm, while for the latter, strategies such as ranking, padding, or key-token selection retain only shared logit dimensions, resulting in much distribution loss. In this paper, we propose Dual-Vocabulary Language Model (DVLM), which replaces the teacher's LM head with a new student-vocabulary projection head and obtains full-dimensional student logits (for #2). To support student tokens (for #1), it takes a Parallel-Tokenized Sequence (PTS) as input, which concatenates the original teacher-tokenized sequence and a re-tokenized sequence formed by independently converting each student token into a teacher-token group. To avoid inference inconsistency with the original teacher tokens, the Hybrid-Prefix Attention (HPA) further restricts re-tokenized groups to their corresponding teacher prefix and uses its last state as the aggregation of the original student-token representation for projection into the student vocabulary space. Similarly, via the combined use of PTS and HPA, the DVLM teacher can provide distribution-aligned supervision with the student's input-tokenization and output-logit during OPD. Experimental results demonstrate that our DVLM teacher has a similar converged loss as the original teacher model and enables student models to improve performance across six reasoning tasks.

The quantum data processing inequality asserts that quantum relative entropy is monotonically non-increasing under quantum channels. Here, we promote this fundamental inequality to an exact equality via boundary and surface Poisson integral representations across a complex strip. The relative-entropy loss is resolved into two independent, nonnegative physical mechanisms: an operational Petz recovery component governing state reconstructibility, and a boundary term measuring the mismatch of relative modular dynamics. We further express the loss as the measured relative entropy between the original state and its averaged Petz recovery state plus a nonnegative variational remainder. For the Belavkin–Staszewski (BS) loss, analogous integral representations yield a decomposition into negative log Uhlmann fidelity with an averaged reconstruction plus a nonnegative variational remainder. Furthermore, we establish recovery bounds at any prescribed modular parameter, including the canonical unrotated Petz map, with optimal square-root scaling in the relative entropy loss. The prefactors depend on distinct-eigenvalue counts or logarithmic spectral quantities, improving upon the inverse-power dependence on small eigenvalues in previous bounds.

We characterize Bregman Douglas-Rachford splitting (BDRS) for unregularized discrete optimal transport and develop an anytime primal-dual certificate in linear memory. We first establish that BDRS coincides with warm-started Inexact Proximal point method for exact Optimal Transport (IPOT) using a single inner Sinkhorn iteration. By eliminating the primal transport plan from the updates, we derive an equivalent dual formulation that reveals BDRS as annealed Sinkhorn under an implicit inverse-linear temperature schedule, with an additional log-scaling momentum term and a cooler kernel. While this explains the role of the temperature parameter in BDRS as an initial temperature, it also reduces the solver's memory requirement from quadratic to linear. Utilizing this annealing perspective, we introduce overrelaxed BDRS, which combines annealing and overrelaxed scaling within a single recursion. We derive a primal-dual certificate for both methods that can be evaluated in linear memory without transport plan construction, thus providing a computable stopping rule. On the DOTmark benchmark, combining momentum with the cooler kernel produces substantially smaller optimality gaps than annealed Sinkhorn under the same schedule. For pixel-level color transfer between $1024\times1024$ images, BDRS attains a lower repaired transport cost than MDOT-TNT with a 9$\times$ speed up, reaching a relative duality gap of $1.59\%$ in 24 minutes. We further demonstrate a color transfer with $4238\times2365$ images, yielding 10 million pixels per image and approximately one hundred trillion implicit transport entries, reaching a best relative duality gap of $2.41\%$ and $2.80\%$ within 35 hours in each direction on a single NVIDIA L40S GPU.

The coroutine has broadly diffused in the practice of concurrent programming in the form of generators and asynchronous functions as well as processes communicating through pipes. We wanted to use coroutines in Python to create single-threaded Unix-style pipelines. Unfortunately, available solutions in Python are cumbersome to use. The JavaScript push-stream protocol appeared to be a good alternative. However, its specification is incomplete and ambiguous. We have used TLA$^{+}$ and the TLC model checker to re-derive the protocol and obtain protocol-specific verification tools. In this paper, we present a formal specification of a push-stream protocol that 1) seamlessly combines synchronous and asynchronous modules, encapsulating the choice within each module; 2) provides flow control without using bounded buffers; 3) gracefully and unambiguously terminates; 4) does not require dynamic allocation of objects on the heap. In addition to completely describing expected behaviours, our specification improves on the original design by 1) allowing the input and output of intermediate pipeline modules to terminate independently and 2) explicitly reporting when a module is pending on the execution environment, to avoid incorrect resuming. We specify the protocol as a sequence of refinement steps and derive by equivalence a specification of what an abstract module may do. We then refine the latter into a module checker that can verify concrete module specifications for conformity. We have verified the key properties of all specifications and the validity of refinement steps with TLC. In supplemental material, we provide all TLA$^{+}$ specifications, show that the protocol is sufficiently expressive to implement a superset of all original JavaScript modules, as well as a performance comparison with Python alternatives and Unix pipes.

Repeated runs of the same coding agent are known to give different benchmark scores. We ask what that variation means for a team running an agent on its own task, by intensive replication on one machine-learning task: an agent improves the training code of an XGBoost classifier for airline delays, and a holdout it never sees scores the result. Across 584 runs, we compare six agents on six open-weight model endpoints, run six agent-model pairings 52 times each under fixed settings, and repeat three of them on a larger model from the same family. Identical runs of one pairing varied more than the pairings differed from one another, so comparisons of a few runs ranked them unreliably; resolving the agent differences we observed would take tens to more than a hundred runs of each. Runs on the larger model scored clearly higher, but by less than one run-to-run standard deviation, and the gap was more than twice as large with one agent as with the others. Fewer than one run in twenty broke the task's data rules, but those runs held the highest scores. Rejecting those runs first and keeping the best compliant result among a few attempts reliably improved the delivered model, even though a few runs could not rank the agents. On flights from a later year, the delivered models kept only a third of their gain over the starting code. At list prices, cost differed more than twentyfold between two agents on the same model, mostly through the prompt cache. Agents and models should be evaluated as pairings, over repeated attempts, with compliance reported beside quality. Data, code and every delivered program: https://github.com/earino/identical-runs-different-results

Motorized two-wheelers (MTW) dominate Indian roads but remain underrepresented in driver behavior research. This study presents the first large-scale analysis of MTW driver gaze behavior in naturalistic, heterogeneous urban traffic, using the myEye2Wheeler dataset. A semantic segmentation pipeline (YOLOv11 + SAM2) was used to extract object-level gaze metrics under two attention modes: direct gaze (foveal overlap) and central vision (parafoveal monitoring). Results reveal a functional division: central vision supports broad monitoring, while direct gaze enables brief, selective sampling. Novice riders exhibit road-anchored scanning, returning to the road between object fixations, while experienced riders form longer chains of attention across multiple objects. The findings suggest that experience primarily refines temporal rhythm rather than altering allocation strategy and reduces object-class effects in gaze patterns. These findings offer new insight into MTW attention structures and inform future work on behavior modeling and safety systems.

Autoregressive text-to-speech (TTS) systems synthesize natural speech but, once trained, offer little control over speaking rate. We show that speaking rate can be steered at inference time, without retraining, by clamping a single decoder block's activation along a discovered speed axis. A decoder-block analysis recovers the rate axis, a neutral operating point, and a per-step intensity scale; at inference, the activation's projection onto this axis is set to a fixed scalar. Learning this direction from synthetically time-stretched and time-compressed speech yields rate control that largely preserves speaker identity, generalizes across model architectures, and maintains high naturalness in objective and human evaluations. Unlike standard additive steering, which breaks at the slow extreme, clamping remains stable on all three systems tested; at moderate targets, the better rule depends on the model. Finally, we show that rate information is decodable across layers but causally steerable only within a mid-depth window, and demonstrate the effectiveness of our approach on the public Seed-TTS-Eval benchmark.

The disruption of stars by supermassive black holes (SMBHs) as tidal disruption events (TDEs) can produce transient flares across the electromagnetic spectrum. Such flares can be used to probe SMBH properties, accretion physics and stellar dynamics in galactic nuclei. However, it can be difficult to distinguish TDE flares and those arising from other accretion phenomena around SMBHs, such as active galactic nuclei disk instabilities. Repeated flares from the same location can be produced by either such instabilities or by repeated partial disruptions of the same star, with the former expected to be at random intervals and the latter roughly periodic. Repeating nuclear transients can thus be used to distinguish between full and partial TDEs, as well as between TDEs and other accretion phenomena. Here we review observations of such repeating nuclear flares at various wavelengths. As time domain surveys become wider, deeper and faster, and as they accumulate longer baselines of data, more repeating phenomena at various time scales are being revealed. While these events show the potential for constraining several open questions regarding SMBH accretion phenomena, the small number of cases so far, the limited observational coverage for many of them, and the overall diversity of emission and repetition properties observed make it difficult to robustly associate most cases with physical scenarios at this stage. However, in the near future, as more data accumulate, the nature of various classes of repeating nuclear transients should become more clear, and with it the nature of many of the one-time flares as well.

We prove that any proper Shimura variety admitting a limpid integral model in the sense of Madapusi Pera becomes a perfectoid space at infinite level at $p$. Our method provides a new geometric approach that circumvents the need for a universal abelian scheme, making it applicable beyond Shimura varieties of Hodge type. Inspired by Scholze's proof of the perfectoidness of the Siegel modular variety, we construct analogues of the canonical subgroup and the anticanonical tower within a strict neighborhood of the ordinary locus. To achieve this we utilize the geometry of the universal $G$-aperture over the limpid integral model. Because every Shimura variety admits a limpid integral model for sufficiently large primes, our result implies that all proper Shimura varieties are globally perfectoid at infinite level for big enough primes.

How well can general-purpose multimodal models turn visual understanding and reasoning into embodied manipulation via executable code? We introduce CodeActionBench, a benchmark of 25 manipulation tasks that evaluates this capability through agentic Code-as-Policy. Without task-specific fine-tuning, demonstrations, external specialist perception or grasp modules, privileged scene state, or predefined task policies, agents should select visual evidence, form task-relevant 3D estimates, construct manipulation targets, and iteratively execute and revise their policies. A shared robot API provides RGB observations, calibrated geometric operations, robot feedback, and bounded motion, leaving task-dependent decisions to the evaluated agent. Fixed task instances, resource budgets, and a hidden physical-outcome verifier support controlled comparisons across models and harness configurations. Extensive evaluations across nine configurations and 675 attempts achieve success rates ranging from 2.7% to 73.3%. The strongest configuration, GPT-6 Astra with Codex CLI, solves 22 of 25 tasks at least once in three attempts, demonstrating the best performance while still leaving substantial room for improvement. Trajectory analyses reveal difficulties in spatial alignment, object retention, and completion judgment, including task failures despite successfully completed motions. CodeActionBench provides a controlled testbed for measuring how general-purpose models translate their capabilities into manipulation behavior and for examining typical failure scenarios in that process.