Search

Benchmarks are important tools for tracking the rapid advancements in large language model (LLM) capabilities. However, benchmarks are not keeping pace in difficulty: LLMs now achieve over 90\% accuracy on popular benchmarks like MMLU, limiting informed measurement of state-of-the-art LLM capabilities. In response, we introduce Humanity's Last Exam (HLE), a multi-modal benchmark at the frontier of human knowledge, designed to be the final closed-ended academic benchmark of its kind with broad subject coverage. HLE consists of 2,700 questions across dozens of subjects, including mathematics, humanities, and the natural sciences. HLE is developed globally by subject-matter experts and consists of multiple-choice and short-answer questions suitable for automated grading. Each question has a known solution that is unambiguous and easily verifiable, but cannot be quickly answered via internet retrieval. State-of-the-art LLMs demonstrate low accuracy and calibration on HLE, highlighting a significant gap between current LLM capabilities and the expert human frontier on closed-ended academic questions. To inform research and policymaking upon a clear understanding of model capabilities, we publicly release HLE at https://lastexam.ai., Comment: 27 pages, 6 figures

Multi-agent systems address issues of accessibility and scalability of artificial intelligence (AI) foundation models, which are often represented by large language models. We develop a framework - the "Society of HiveMind" (SOHM) - that orchestrates the interaction between multiple AI foundation models, imitating the observed behavior of animal swarms in nature by following modern evolutionary theories. On the one hand, we find that the SOHM provides a negligible benefit on tasks that mainly require real-world knowledge. On the other hand, we remark a significant improvement on tasks that require intensive logical reasoning, indicating that multi-agent systems are capable of increasing the reasoning capabilities of the collective compared to the individual agents. Our findings demonstrate the potential of combining a multitude of diverse AI foundation models to form an artificial swarm intelligence capable of self-improvement through interactions with a given environment., Comment: 11 pages (excl. appendix)

We calculate the effects of quantum fluctuations of a scalar field in the "ballpoint pen" cosmic string geometry. Using the approach to renormalization established previously for the energy density in two space dimensions, we extend those calculations to $3+1$ dimensions, nonzero scalar mass, and the full stress-energy tensor, including its contribution to the null energy condition for radial geodesics. The calculation demonstrates in detail the process of renormalization in curved spacetime, including the effects of the conformal anomaly. This calculation provides one of the few examples where quantum effects in curved spacetime can be explicitly calculated., Comment: 12 pages, 10 .eps figures

The Galactic Bulge Time Domain Survey (GBTDS) of the Roman Space Telescope will take high cadence data of the Galactic bulge. We investigate the asteroseismic potential of this survey for red giants. We simulate the detectability of global asteroseismic frequencies, $\nu_{\mathrm{max}}$ and $\Delta\nu$, by modify ing Kepler data to match nominal GBTDS observing strategies, considering different noise models, observing cadences, and detection algorithms. Our baseline case, using conservative assumptions, consistently leads to asteroseismic $\nu_{\mathrm{max}}$ detection probabilities above 80% for red clump and red giant branch stars brighter than 16th magnitude in Roman's F146 filter. We then inject these detection probabilities into a Galaxia model of the bulge to estimate asteroseismic yields. For our nominal case, we detect 290,000 stars in total, with 185,000 detections in the bulge. Different assumptions give bulge yields from 135,000 to 349,000 stars. For stars with measured $\nu_{\mathrm{max}}$, we find that we can recover $\Delta\nu$ in 21% to 42% of red clump stars, and 69% to 92% of RGB stars. Implications for survey strategy and asteroseismic population studies are discussed more., Comment: 16 pages, 16 figures, submitted to ApJ

State-of-the-art superconducting quantum processors containing tens to hundreds of qubits have demonstrated the building blocks for realizing fault-tolerant quantum computation. Nonetheless, a fundamental barrier to scaling further is the prevalence of fluctuating quantum two-level system (TLS) defects that can couple resonantly to qubits, causing excess decoherence and enhanced gate errors. Here we introduce a scalable architecture for site-specific and in-situ manipulation of TLS frequencies out of the spectral vicinity of our qubits. Our method is resource efficient, combining TLS frequency tuning and universal single qubit control into a single on-chip control line per qubit. We independently control each qubit's dissipative environment to dynamically improve both qubit coherence times and single qubit gate fidelities -- with a constant time overhead that does not scale with the device size. Over a period of 40 hours across 6 qubits, we demonstrate a $36\%$ improvement in average single qubit error rates and a $17\%$ improvement in average energy relaxation times. Critically, we realize a 4-fold suppression in the occurrence of TLS-induced performance outliers, and a complete reduction of simultaneous outlier events. These results mark a significant step toward overcoming the challenges that TLS defects pose to scaling superconducting quantum processors.

One of the most prolific methods of studying exoplanet atmospheres is transmission spectroscopy, which measures the difference between the depth of an exoplanet's transit signal at various wavelengths and attempts to correlate the depth changes to potential features in the exoplanet's atmosphere. Here we present reconnaissance observations of 21 exoplanet atmospheres measured with the Exoplanet Transmission Spectroscopy Imager (ETSI), a recently deployed spectro-photometer on the McDonald Observatory Otto Struve 2.1 m telescope. ETSI measurements are mostly free of systematics through the use of a novel observing technique called common-path multi-band imaging (CMI), which has been shown to achieve photometric color precision on-par with space-based observations (300ppm or 0.03%). This work also describes the various statistical tests performed on the data to evaluate the efficacy of the CMI method and the ETSI instrument in combination. We find that none of the 8 comparisons of exoplanet atmospheres measured with ETSI and other observatories (including the Hubble Space Telescope) provide evidence that the spectra are statistically dissimilar. These results suggest that ETSI can provide initial transmission spectroscopy observations for a fraction of the observational and monetary overhead previously required to detect an exoplanet's atmosphere. Ultimately these reconnaissance observations increase the number of planets with transmission spectroscopy measurements by ~10% and provide an immediate prioritization of 21 exoplanets for future follow-up with more precious observatories, such as the James Webb Space Telescope. The reconnaissance spectra are available through the Filtergraph visualization portal at the URL: https://filtergraph.com/etsi/., Comment: 28 pages, 8 figures, 11 tables, article published in AJ. ETSI data is available on request and spectra can be found at the URL https://filtergraph.com/etsi/

The coexistence of multiple types of orders is a common thread in condensed matter physics and unconventional superconductors. The nature of superconducting orders may be unveiled by analyzing local perturbations such as vortices. For thin films, the vortex magnetic profile is characterized by the Pearl-length {\Lambda}, which is inversely proportional to the 2D superfluid density; hence, normally, also inversely proportional to the film thickness, d. Here we employ the scanning SQUID-on-tip microscopy to measure {\Lambda} in NbSe2 flakes with thicknesses ranging from N=3 to 53 layers. For N>10, we find the expected dependence {\Lambda}{\varpropto}1/d. However, six-layer films show a sharp increase of {\Lambda} deviating by a factor of three from the expected value. This value remains fixed for N=3 to 6. This unexpected behavior suggests the competition between two orders; one residing only on the first and last layers of the film while the other prevails in all layers., Comment: Main text: 10 pages, 4 figures. Supplementary information: 13 pages, 6 figures

Automated defect detection in industrial manufacturing is essential for maintaining product quality and minimizing production errors. In air disc brake manufacturing, ensuring the precision of laser-engraved nameplates is crucial for accurate product identification and quality control. Engraving errors, such as misprints or missing characters, can compromise both aesthetics and functionality, leading to material waste and production delays. This paper presents a proof of concept for an AI-driven computer vision system that inspects and verifies laser-engraved nameplates, detecting defects in logos and alphanumeric strings. The system integrates object detection using YOLOv7, optical character recognition (OCR) with Tesseract, and anomaly detection through a residual variational autoencoder (ResVAE) along with other computer vision methods to enable comprehensive inspections at multiple stages. Experimental results demonstrate the system's effectiveness, achieving 91.33% accuracy and 100% recall, ensuring that defective nameplates are consistently detected and addressed. This solution highlights the potential of AI-driven visual inspection to enhance quality control, reduce manual inspection efforts, and improve overall manufacturing efficiency.

We obtained high spectral resolution spectra ($\Delta v$ $\sim$ 2.5 km s$^{-1}$) for DG Tau A from 4800 \r{A} to 7500 \r{A} using Subaru High Dispersion Spectrograph (HDS) for the first time. The low-velocity components (LVCs, |$v$| < 100 km s$^{-1}$) were observed in the [O I] 5577, 6300, 6364 \r{A}, [S II] 6716, 6731 \r{A} lines. The offset position spectra observed in the LVCs show a "negative velocity gradient", supporting the presence of a wide-angled wind associated with the LVC emission. The offset position spectra observed in a component within the LVC velocity range between -16 km s$^{-1}$ to -41 km s$^{-1}$, namely, LVC-M, show a "negative velocity gradient'', supporting the presence of a wide-angled wind. With 12-70 au wind lengths measured using spectroastrometry, we estimate a lower limit to the wind mass-loss rate of $\sim$10$^{-8}$ M$_\odot$ yr$^{-1}$. In addition to the LVCs, we identify two high-velocity components (HVCs, |$v$| > 100 km s$^{-1}$) associated with the collimated jet in 26 lines ([N I], [N II], [O I], [O II], [O III], [S II], [Ca II], [Fe II], H$\alpha$, H$\beta$, He I). The one with a clear spatial offset from the star ($n_e$ $\sim$10$^4$ cm$^{-3}$, HVC1) is associated with an internal shock surface of the jet, while the other at the base ($n_e$ $\sim$10$^6$ cm$^{-3}$, HVC2) may be a stationary shock component. We find that the observed line profiles and the spatial scales of the LVC emission do not agree with the existing predictions for photoevaporative or magnetohydrodynamical (MHD) disk winds. These could be explained by the X-wind model, but synthetic observations are required for detailed comparisons., Comment: 28 pages, 14 Figures, 5 Tables, accepted by ApJ

We present a detailed analysis of the nature of migration of protoplanetary clumps formed via disc instability in self-consistent 3D hydrodynamical (HD) and magneto-hydrodynamical (MHD) simulations of self-gravitating discs. Motivated by the complex structure of protoplanetary clumps we do not introduce sink particles. We find that the orbital evolution of the clumps has a stochastic character but also exhibits recurrent properties over many orbits. Clump migration is governed by two sources of gravitational torques: a torque originating from a region about twice the Hill sphere around each clump's orbit, and the torque resulting from clump-clump interactions. Compared to non-magnetized companion runs, the latter are more frequent in MHD simulations, which give rise to more numerous clumps starting off at smaller masses, often below a Neptune mass. Clump-clump interactions can lead to temporary strong accelerations of migration in both directions, but integrated over time provide a lesser impact than disc-driven torques. They can also lead to clump mergers but do not cause ejections; a difference to previous works which adopted sink particles. The local "Hill torque" is responsible for the fast migration, inward or outward. Estimating the characteristic timescales of conventional migration in our regime, we find that the disc-driven migration timescales are in agreement with Type III migration. However, the dominant local torque is rapidly fluctuating, which reflects the turbulent nature of the flow. The resulting stochastic migration pattern is markedly different from Type III runaway migration and appears to be a distinctive feature of orbital dynamics in a fragmenting disc., Comment: 16 pages, 16 figures

We study and compare the expressive power of transformers, RNNs, and transformers with chain of thought tokens on a simple and natural class of problems we term Compositional Reasoning Questions (CRQ). This family captures problems like evaluating Boolean formulas and multi-step word problems. Assuming standard hardness assumptions from circuit complexity and communication complexity, we prove that none of these three architectures is capable of solving CRQs unless some hyperparameter (depth, embedding dimension, and number of chain of thought tokens, respectively) grows with the size of the input. We also provide a construction for each architecture that solves CRQs. For transformers, our construction uses depth that is logarithmic in the problem size. For RNNs, logarithmic embedding dimension is necessary and sufficient, so long as the inputs are provided in a certain order. (Otherwise, a linear dimension is necessary). For transformers with chain of thought, our construction uses $n$ CoT tokens. These results show that, while CRQs are inherently hard, there are several different ways for language models to overcome this hardness. Even for a single class of problems, each architecture has strengths and weaknesses, and none is strictly better than the others.

Trigonometric derivatives are fundamental in both mathematics and physics, yet their proper application, particularly the distinction between radians and degrees, poses a significant challenge for college students globally. This study identifies a widespread "blind spot" in understanding trigonometric derivatives and their implications for physical systems, highlighting a critical gap in physics education. A multinational survey of 769 college students, primarily undergraduate and graduate STEM majors, from Israel, the United States, China, and India assessed their ability to differentiate between radians and degrees in mathematical and physical contexts, focusing on harmonic motion. Results reveal that only 26.3\% of students correctly identified that the well-known expressions for trigonometric derivatives hold exclusively in radians, while 70.7\% incorrectly assumed both radians and degrees are valid. Notably, students demonstrated improved recognition of radians in physical contexts (59.0\% correct responses) compared to mathematical ones, suggesting that students rely on familiar physical equations as cognitive reference points when applying mathematical concepts. These misunderstandings appear worldwide, suggesting a universal challenge. The findings highlight the need for curriculum reforms to better connect mathematical formalism with physical application.

The coexistence of topological bands around the Fermi level ($E_F$) and superconductivity provides a fundamental platform for exploring their interplay. However, few materials inherently display both properties. In this study, we demonstrate the coexistence of topological surface states at the $E_F$ and superconductivity in NiTe$_2$ single crystals, a material hitherto not recognized as superconducting. Quasiparticle interference measurements performed via scanning tunneling microscopy suggest the presence of topological surface states at the $E_F$, which is further corroborated by density functional theory simulations. Experimental evidence for superconductivity is provided via electronic transport measurements and specific heat capacity analyses. Our results suggest that NiTe$_2$ represents a promising platform for investigating the rich interplay between topological states and superconductivity.

Test-time inference has emerged as a powerful paradigm for enabling language models to ``think'' longer and more carefully about complex challenges, much like skilled human experts. While reinforcement learning (RL) can drive self-improvement in language models on verifiable tasks, some models exhibit substantial gains while others quickly plateau. For instance, we find that Qwen-2.5-3B far exceeds Llama-3.2-3B under identical RL training for the game of Countdown. This discrepancy raises a critical question: what intrinsic properties enable effective self-improvement? We introduce a framework to investigate this question by analyzing four key cognitive behaviors -- verification, backtracking, subgoal setting, and backward chaining -- that both expert human problem solvers and successful language models employ. Our study reveals that Qwen naturally exhibits these reasoning behaviors, whereas Llama initially lacks them. In systematic experimentation with controlled behavioral datasets, we find that priming Llama with examples containing these reasoning behaviors enables substantial improvements during RL, matching or exceeding Qwen's performance. Importantly, the presence of reasoning behaviors, rather than correctness of answers, proves to be the critical factor -- models primed with incorrect solutions containing proper reasoning patterns achieve comparable performance to those trained on correct solutions. Finally, leveraging continued pretraining with OpenWebMath data, filtered to amplify reasoning behaviors, enables the Llama model to match Qwen's self-improvement trajectory. Our findings establish a fundamental relationship between initial reasoning behaviors and the capacity for improvement, explaining why some language models effectively utilize additional computation while others plateau.

This report presents the findings of a NIAC Phase I feasibility study for the Artemis-enabled Stellar Imager (AeSI), a proposed high-resolution, UV/Optical interferometer designed for deployment on the lunar surface. Its primary science goal is to image the surfaces and interiors of stars with unprecedented detail, revealing new details about their magnetic processes and dynamic evolution and enabling the creation of a truly predictive solar/stellar dynamo model. This capability will transform our understanding of stellar physics and has broad applicability across astrophysics, from resolving the cores of Active Galactic Nuclei (AGN) to studying supernovae, planetary nebulae, and the late stages of stellar evolution. By leveraging the stable vacuum environment of the Moon and the infrastructure being established for the Artemis Program, AeSI presents a compelling case for a lunar-based interferometer. In this study, the AeSI Team, working with the NASA Goddard Space Flight Center's Integrated Design Center (IDC), has firmly established the feasibility of building and operating a reconfigurable, dispersed aperture telescope (i.e., an interferometer) on the lunar surface. The collaboration produced a credible Baseline design featuring 15 primary mirrors arranged in an elliptical array with a 1 km major axis, with the potential to expand to 30 mirrors and larger array sizes through staged deployments. Additionally, this study identified numerous opportunities for optimization and the necessary trade studies to refine the design further. These will be pursued in follow-up investigations, such as a NIAC Phase II study, to advance the concept toward implementation., Comment: 128 pages, 59 figures

Collective neutrino flavor oscillations are of primary importance in understanding the dynamic evolution of core-collapse supernovae and subsequent terrestrial detection, but also among the most challenging aspects of numerical simulations. This situation is complicated by the quantum many-body nature of the problem due to neutrino-neutrino interactions which demands a quantum treatment. An additional complication is the presence of three flavors, which often is approximated by the electron flavor and a heavy lepton flavor. In this work, we provide both qubit and qutrit encodings for all three flavors, and develop optimized quantum circuits for the time evolution and analyze the Trotter error. We conclude our study with a hardware experiment of a system of two neutrinos with superconducting hardware: the IBM Torino device for qubits and AQT device for qutrits. We find that error mitigation greatly helps in obtaining a signal consistent with simulations. While hardware results are comparable at this stage, we expect the qutrit setup to be more convenient for large-scale simulations since it does not suffer from probability leakage into nonphsycial qubit space, unlike the qubit setup.

This paper presents performance measurements of a new readout electronics system based on silicon photomultipliers for the PHENIX electromagnetic calorimeter. Installation of the lead-scintillator Shashlik style calorimeter into the SeaQuest/SpinQuest spectrometer has been proposed to broaden the experiment's dark sector search program, an upgrade known as DarkQuest. The calorimeter and electronics system were subjected to testing and calibration at the Fermilab Test Beam Facility. Detailed studies of the energy response and resolution, as well as particle identification capabilities, were performed. The background rate in the actual experimental environment was also examined. The system is found to be well-suited for a dark sector search program on the Fermilab 120 GeV proton beamline., Comment: Prepared for submission to Nuclear Instrumentation and Methods

Orbital Fulde-Ferrell-Larkin-Ovchinnikov (FFLO) states are analyzed within mean-field theory, where orbital FF state is layer-polarized with inversion symmetry broken, while orbital LO state is Josephson vortex array with translation symmetry reduced. Phase diagrams of orbital FFLO states are obtained, and properties such as induced orders, superconducting diode effects, Fraunhofer pattern and topological defects are studied for the probe of FF versus LO states., Comment: 4 pages (maintext)+1 page (references)+2 pages (appendix), 3 figures

Motion planning for locomotion systems typically requires translating high-level rigid-body tasks into low-level joint trajectories-a process that is straightforward for car-like robots with fixed, unbounded actuation inputs but more challenging for systems like snake robots, where the mapping depends on the current configuration and is constrained by joint limits. In this paper, we focus on generating continuous families of optimal gaits-collections of gaits parameterized by step size or steering rate-to enhance controllability and maneuverability. We uncover the underlying geometric structure of these optimal gait families and propose methods for constructing them using both global and local search strategies, where the local method and the global method compensate each other. The global search approach is robust to nonsmooth behavior, albeit yielding reduced-order solutions, while the local search provides higher accuracy but can be unstable near nonsmooth regions. To demonstrate our framework, we generate optimal gait families for viscous and perfect-fluid three-link swimmers. This work lays a foundation for integrating low-level joint controllers with higher-level motion planners in complex locomotion systems., Comment: 17 pages, submitted to IEEE Transactions on Robotics

Training large language models to predict beyond their training context lengths has drawn much attention in recent years, yet the principles driving such behavior of length generalization remain underexplored. We propose a new theoretical framework to study length generalization for the next-token prediction task, as performed by decoder-only transformers. Conceptually, we show that length generalization occurs as long as each predicted token depends on a small (fixed) number of previous tokens. We formalize such tasks via a notion we call $k$-sparse planted correlation distributions, and show that an idealized model of transformers which generalize attention heads successfully length-generalize on such tasks. As a bonus, our theoretical model justifies certain techniques to modify positional embeddings which have been introduced to improve length generalization, such as position coupling. We support our theoretical results with experiments on synthetic tasks and natural language, which confirm that a key factor driving length generalization is a ``sparse'' dependency structure of each token on the previous ones. Inspired by our theory, we introduce Predictive Position Coupling, which trains the transformer to predict the position IDs used in a positional coupling approach. Predictive Position Coupling thereby allows us to broaden the array of tasks to which position coupling can successfully be applied to achieve length generalization.

Device error rates on current quantum computers have improved enough to where demonstrations of error correction below break-even are now possible. Still, the circuits required for quantum error correction introduce significant overhead and sometimes inject more errors than they correct. In this work, we introduce adaptive syndrome extraction as a scheme to improve code performance and reduce the quantum error correction cycle time by measuring only the stabilizer generators that are likely to provide useful syndrome information. We provide a concrete example of the scheme through the [[4,2,2]] code concatenated with a hypergraph product code and a syndrome extraction cycle that uses quantum error detection to modify the syndrome extraction circuits in real time. Compared to non-concatenated codes and non-adaptive syndrome extraction, we find that the adaptive scheme achieves over an order of magnitude lower logical error rates while requiring fewer CNOT gates and physical qubits. Furthermore, we show how to achieve fault-tolerant universal logical computation with [[4,2,2]]-concatenated hypergraph product codes., Comment: 26 pages, 15 figures

Microcalorimeter instruments aboard future X-ray observatories will require an anti-coincidence (anti-co) detector to veto charged particle events and reduce the non-X-ray background. We have developed a large-format, TES-based prototype anti-coincidence detector that is particularly suitable for use with spatially-extended (~ 10 cm^2}) TES microcalorimeter arrays, as would be used for a future large field-of-view X-ray missions. This prototype was developed in the context of the Line Emission Mapper (LEM) probe concept, which required a ~ 14 cm^2 anti-co detector with > 95% live time and a low-energy threshold below 20 keV. Our anti-co design employs parallel networks of quasiparticle-trap-assisted electrothermal feedback TESs (QETs) to detect the athermal phonon signal produced in the detector substrate by incident charged particles. We developed multiple prototype anti-co designs featuring 12 channels and up to 6300 QETs. Here we focus on a design utilizing tungsten TESs and present characterization results. Broad energy range measurements have been performed (4.1 keV -- 5.5 MeV). Based on noise and responsivity measurements, the implied low-energy threshold is < 1 keV and a live time fraction of > 96% can be achieved up to 5.5 MeV. We also find evidence of mm-scale-or-better spatial resolution and discuss the potential utility of this for future missions. Finally, we discuss the early development of a soild-state physics model of the anti-co towards understanding phonon propagation and quasiparticle production in the detector., Comment: 26 pages, 16 figures

We consider a superconductor under external perturbation, which forces Cooper pairs to develop with a finite total momentum $q$. The condensation energy of such a state decreases with $q$ and vanishes at a critical $q_c$. We analyze how superconducting order evolves at $ q \approx q_c$. In 3D, the result is well-known: the pairing susceptibility diverges at $q = q_c +0$, and the gap amplitude $\Delta (q)$ gradually increases as $q$ decreases below $q_c$ and reaches its largest value $\Delta_0$ at $q=0$. In 2D, we find different behavior. Namely, for a parabolic dispersion, the pairing susceptibility also diverges at $q = q_c +0$, but at $q = q_c -0$, the gap amplitude jumps to the maximal $\Delta_0$ and remains equal to it for all $q 1$ the transition becomes second-order, but the gap evolution is rather sharp, whereas for $\alpha<1$ it becomes first-order, but $\Delta (q)$ is non-monotonic. This is similar, but not identical, to the behavior of magnetization near a Stoner transition in 2D., Comment: 26 pages + 9 figures

We study a system of forced viscous shallow water equations with nontrivial bathymetry in two spatial dimensions. We develop a well-posedness theory for small but arbitrary forcing data, as well as for a fixed data profile but large amplitude. In the latter case, solutions may actually fail to exist for large amplitude, but in this case we prove that one of three physically meaningful breakdown scenarios occurs. Through the use of implicit function theorem techniques and a priori estimates, we construct both spatially periodic and solitary (non-periodic but spatially localized) solutions. The solitary case is substantially more complicated, requiring a delicate analysis in weighted Sobolev spaces. To the best of our knowledge, these results constitute the first general construction of stationary wave solutions, large or otherwise, to the viscous shallow water equations and the first general analysis of large solitary wave solutions to any viscous free boundary fluid model., Comment: 64 pages, 2 figures

Existing regulatory frameworks for biomedical AI include robustness as a key component but lack detailed implementational guidance. The recent rise of biomedical foundation models creates new hurdles in testing and certification given their broad capabilities and susceptibility to complex distribution shifts. To balance test feasibility and effectiveness, we suggest a priority-based, task-oriented approach to tailor robustness evaluation objectives to a predefined specification. We urge concrete policies to adopt a granular categorization of robustness concepts in the specification. Our approach promotes the standardization of risk assessment and monitoring, which guides technical developments and mitigation efforts., Comment: under review, comments welcome

Deep subwavelength RF imaging with atomic Rydberg sensors has overcome fundamental limitations of traditional antennas and enabled ultra-wideband detection of omni-directional time varying fields all in a compact form factor. However, in most applications, Rydberg sensors require the use of a secondary strong RF reference field to serve as a phase reference. Here, we demonstrate a new type of Rydberg sensor for angle-of-arrival (AoA) sensing which utilizes subwavelength imaging of standing wave fields. By placing a metallic plate within the Rydberg cell, we can determine the AoA independent of the strength of incoming RF field and without requiring a secondary strong RF phase reference field. We perform precision AoA measurements with a robotic antenna positioning system for 4.2, 5.0, and 5.7 GHz signals and demonstrate a 1.7 deg polar angular resolution from 0 deg to 60 deg AoA and 4.1 deg over all possible angles.

Let $Y$ be a genus $2$ curve over $\mathbb Q$. We provide a method to systematically search for possible candidates of prime $\ell\geq 3$ and a genus $1$ curve $X$ for which there exists genus $3$ curve $Z$ over $\mathbb Q$ whose Jacobian is up to quadratic twist $(\ell, \ell, \ell)$-isogenous to the product of Jacobians of $X$ and $Y$, building off of the work by Hanselman, Schiavone, and Sijsling for $\ell=2$. We find several such pairs $(X,Y)$ for prime $\ell$ up to $13$. We also improve their numerical gluing algorithm, allowing us to successfully glue genus $1$ and genus $2$ curves along their $13$-torsion., Comment: 21 pages

We prove that any triple of complete flags in $\mathbb R^d$ admits a common generating set of size $\lfloor 5d/3\rfloor$ and that this bound is sharp. This result extends the classical linear-algebraic fact -- a consequence of the Bruhat decomposition of $\text{GL}_d(\mathbb R)$ -- that any pair of complete flags in $\mathbb R^d$ admits a common generating set of size $d$. We also deduce an analogue for $m$-tuples of flags with $m>3$., Comment: 16 pages, 6 figures, comments are welcome

Hawking radiation and black hole thermodynamics are well understood in the frameworks of quantum field theory and general relativity, with contemporary extensions in string theory, AdS/CFT, and loop quantum gravity. However, an open question remains: Can Hawking radiation be consistently described using statistical mechanics? This challenge arises due to the ambiguous quantum nature of Hawking particles and constraints imposed by information conservation. This study develops a heuristic statistical model using a worldine Lagrangian formalism, treating Hawking particles as timelike with an effective mass and following an arbitrary statistical distribution. The flow of information is modeled as a transfer of microstates from the event horizon to the radiation background within a closed ensemble. From this framework, the Hawking particle mass $m_H=\hbar/(2\pi r_S)$ is recovered, and the worldline Lagrangian yields the black hole's thermal energy, $E=\hbar/(8\pi GM)$. Beyond preserving conventional insights, reinforcing the statistical nature of black hole evaporation, this approach introduces novel perspectives: modeling the black hole as a structured energy well containing microstates, quantifying the evaporation rate of individual microstates, and determining the total number of microstates from the horizon's surface area. This study leaves the door open for a stochastic mechanical analysis, which is introduced in the Discussion section., Comment: (Version 3: in style of PLB preprint): 5 pages, 1 figure; typo errors were addressed

Achieving a delicate balance between fostering trust in law enforcement and protecting the rights of both officers and civilians continues to emerge as a pressing research and product challenge in the world today. In the pursuit of fairness and transparency, this study presents an innovative AI-driven system designed to generate police report drafts from complex, noisy, and multi-role dialogue data. Our approach intelligently extracts key elements of law enforcement interactions and includes them in the draft, producing structured narratives that are not only high in quality but also reinforce accountability and procedural clarity. This framework holds the potential to transform the reporting process, ensuring greater oversight, consistency, and fairness in future policing practices. A demonstration video of our system can be accessed at https://drive.google.com/file/d/1kBrsGGR8e3B5xPSblrchRGj-Y-kpCHNO/view?usp=sharing

Humans naturally interpret numbers non-literally, effortlessly combining context, world knowledge, and speaker intent. We investigate whether large language models (LLMs) interpret numbers similarly, focusing on hyperbole and pragmatic halo effects. Through systematic comparison with human data and computational models of pragmatic reasoning, we find that LLMs diverge from human interpretation in striking ways. By decomposing pragmatic reasoning into testable components, grounded in the Rational Speech Act framework, we pinpoint where LLM processing diverges from human cognition -- not in prior knowledge, but in reasoning with it. This insight leads us to develop a targeted solution -- chain-of-thought prompting inspired by an RSA model makes LLMs' interpretations more human-like. Our work demonstrates how computational cognitive models can both diagnose AI-human differences and guide development of more human-like language understanding capabilities., Comment: 12 pages, 10 figures

We demonstrate the imaging of localized surface electric (E) field effects on the atomic spectrum in a vapor cell used in Rydberg atom-based sensors. These surface E-fields can result from an induced electric charge distribution on the surface. Induced surface charge distributions can dramatically perturb the atomic spectrum, hence degrading the ability to perform electrometry. These effects become pronounced near the walls of the vapor cell, posing challenges for vapor cell miniaturization. Using a fluorescence imaging technique, we investigate the effects of surface charge on the atomic spectrum generated with electromagnetically induced transparency (EIT). Our results reveal that visible light (480 nm and 511 nm), i.e., the coupling laser used in two-photon Rydberg EIT schemes, generates localized patches of charge or dipoles where this light interacts with the glass walls of the vapor cell, while a three-photon Rydberg EIT scheme using only near-infrared wavelength lasers shows no measurable field induction. Additionally, imaging in a vacuum chamber where a glass plate is placed between large electrodes confirms that the induced charge is positive. We further validate these findings by studying the photoelectric effect with broadband light during EIT and impedance measurements. These results demonstrate the power of the fluorescence imaging technique to study localized E-field distributions in vapor cells and to target the photoelectric effect of the alkali-exposed glass of vapor cells as a major disruptor in Rydberg atom-based sensors.

Realizing the vision of using AI agents to automate critical IT tasks depends on the ability to measure and understand effectiveness of proposed solutions. We introduce ITBench, a framework that offers a systematic methodology for benchmarking AI agents to address real-world IT automation tasks. Our initial release targets three key areas: Site Reliability Engineering (SRE), Compliance and Security Operations (CISO), and Financial Operations (FinOps). The design enables AI researchers to understand the challenges and opportunities of AI agents for IT automation with push-button workflows and interpretable metrics. ITBench includes an initial set of 94 real-world scenarios, which can be easily extended by community contributions. Our results show that agents powered by state-of-the-art models resolve only 13.8% of SRE scenarios, 25.2% of CISO scenarios, and 0% of FinOps scenarios. We expect ITBench to be a key enabler of AI-driven IT automation that is correct, safe, and fast.

The privacy vulnerabilities of the federated learning (FL) paradigm, primarily caused by gradient leakage, have prompted the development of various defensive measures. Nonetheless, these solutions have predominantly been crafted for and assessed in the context of synchronous FL systems, with minimal focus on asynchronous FL. This gap arises in part due to the unique challenges posed by the asynchronous setting, such as the lack of coordinated updates, increased variability in client participation, and the potential for more severe privacy risks. These concerns have stymied the adoption of asynchronous FL. In this work, we first demonstrate the privacy vulnerabilities of asynchronous FL through a novel data reconstruction attack that exploits gradient updates to recover sensitive client data. To address these vulnerabilities, we propose a privacy-preserving framework that combines a gradient obfuscation mechanism with Trusted Execution Environments (TEEs) for secure asynchronous FL aggregation at the network edge. To overcome the limitations of conventional enclave attestation, we introduce a novel data-centric attestation mechanism based on Multi-Authority Attribute-Based Encryption. This mechanism enables clients to implicitly verify TEE-based aggregation services, effectively handle on-demand client participation, and scale seamlessly with an increasing number of asynchronous connections. Our gradient obfuscation mechanism reduces the structural similarity index of data reconstruction by 85% and increases reconstruction error by 400%, while our framework improves attestation efficiency by lowering average latency by up to 1500% compared to RA-TLS, without additional overhead.

We demonstrate that applying an eventual decay to the learning rate (LR) in empirical risk minimization (ERM), where the mean-squared-error loss is minimized using standard gradient descent (GD) for training a two-layer neural network with Lipschitz activation functions, ensures that the resulting network exhibits a high degree of Lipschitz regularity, that is, a small Lipschitz constant. Moreover, we show that this decay does not hinder the convergence rate of the empirical risk, now measured with the Huber loss, toward a critical point of the non-convex empirical risk. From these findings, we derive generalization bounds for two-layer neural networks trained with GD and a decaying LR with a sub-linear dependence on its number of trainable parameters, suggesting that the statistical behaviour of these networks is independent of overparameterization. We validate our theoretical results with a series of toy numerical experiments, where surprisingly, we observe that networks trained with constant step size GD exhibit similar learning and regularity properties to those trained with a decaying LR. This suggests that neural networks trained with standard GD may already be highly regular learners., Comment: 26 pages, 8 figures

Threshold cryptography is a powerful and well-known technique with many applications to systems relying on distributed trust. It has recently emerged also as a solution to challenges in blockchain: frontrunning prevention, managing wallet keys, and generating randomness. This work presents Thetacrypt, a versatile library for integrating many threshold schemes into one codebase. It offers a way to easily build distributed systems using threshold cryptography and is agnostic to their implementation language. The architecture of Thetacrypt supports diverse protocols uniformly. The library currently includes six cryptographic schemes that span ciphers, signatures, and randomness generation. The library additionally contains a flexible adapter to an underlying networking layer that provides peer-to-peer communication and a total-order broadcast channel; the latter can be implemented by distributed ledgers, for instance. Thetacrypt serves as a controlled testbed for evaluating the performance of multiple threshold-cryptographic schemes under consistent conditions, showing how the traditional micro benchmarking approach neglects the distributed nature of the protocols and its relevance when considering system performance.

Plasmonic response in metals, defined as the ability to support subwavelength confinement of surface plasmon modes, is typically limited to a narrow frequency range below the metals' plasma frequency. This places severe limitations on the operational wavelengths of plasmonic materials and devices. However, when the volume of a metal film is massively decreased, highly confined quasi-two-dimensional surface plasmon modes can be supported out to wavelengths well beyond the plasma wavelength. While this has, thus far, been achieved using ultra-thin (nm-scale) metals, such films are quite difficult to realize, and suffer from even higher losses than bulk plasmonic films. To extend the plasmonic response to the infrared, here we introduce the concept of metaplasmonics, representing a novel plasmonic modality with a host of appealing properties. By fabricating and characterizing a series of metaplasmonic nanoribbons, we demonstrate large confinement, high quality factors, and large near-field enhancements across a broad wavelength range, extending well beyond the limited bandwidth of traditional plasmonic materials. We demonstrate $35\times$ plasmon wavelength reduction, and our numerical simulations suggest that further wavelength reduction, up to a factor of 150, is achievable using our approach. The demonstration of the metaplasmonics paradigm offers a promising path to fill the near- and mid-infrared technological gap for high quality plasmonic materials, and provides a new material system to study the effects of extreme plasmonic confinement for applications in nonlinear and quantum plasmonics.

In the classical caching problem, when a requested page is not present in the cache (i.e., a "miss"), it is assumed to travel from the backing store into the cache "before" the next request arrives. However, in many real-life applications, such as content delivery networks, this assumption is unrealistic. The "delayed-hits" model for caching, introduced by Atre, Sherry, Wang, and Berger, accounts for the latency between a missed cache request and the corresponding arrival from the backing store. This theoretical model has two parameters: the "delay" $Z$, representing the ratio between the retrieval delay and the inter-request delay in an application, and the "cache size" $k$, as in classical caching. Classical caching corresponds to $Z=1$, whereas larger values of $Z$ model applications where retrieving missed requests is expensive. Despite the practical relevance of the delayed-hits model, its theoretical underpinnings are still poorly understood. We present the first tight theoretical guarantee for optimizing delayed-hits caching: The "Least Recently Used" algorithm, a natural, deterministic, online algorithm widely used in practice, is $O(Zk)$-competitive, meaning it incurs at most $O(Zk)$ times more latency than the (offline) optimal schedule. Our result extends to any so-called "marking" algorithm., Comment: Accepted at INFOCOM2025

We investigate the inverse tsunami wave problem within the framework of the 1D nonlinear shallow water equations (SWE). Specifically, we focus on determining the initial displacement $\eta_0(x)$ and velocity $u_0(x)$ of the wave, given the known motion of the shoreline $R(t)$ (the wet/dry free boundary). We demonstrate that for power-shaped inclined bathymetries, this problem admits a complete solution for any $\eta_0$ and $u_0$, provided the wave does not break. In particular, we show that the knowledge of $R(t)$ enables the unique recovery of both $\eta_0(x$) and $u_0(x)$ in terms of the Abel transform. It is important to note that, in contrast to the direct problem (also known as the tsunami wave run-up problem), where $R(t)$ can be computed exactly only for $u_0(x)=0$, our algorithm can recover $\eta_0$ and $u_0$ exactly for any non-zero $u_0$. This highlights an interesting asymmetry between the direct and inverse problems. Our results extend the work presented in \cite{Rybkin23,Rybkin24}, where the inverse problem was solved for $u_0(x)=0$. As in previous work, our approach utilizes the Carrier-Greenspan transformation, which linearizes the SWE for inclined bathymetries. Extensive numerical experiments confirm the efficiency of our algorithms.

The arXiv preprint 2411.15164v1 \cite{Li:2024ltx} discusses how a weak Schwarzschild spacetime modifies Feynman diagram calculations and the governing equations of scalar and spinor particles. This short review offers an extended analysis on concerning points involving massless particles and limitations of weak-field approximation. Written for the Universit\"at Potsdam lecture \textit{New Developments in Astrophysics, Winter Semester 2024/25}., Comment: 3 pages, 1 figure; comment on arXiv:2411.15164