Search

Recent advancements in vision-language models (VLMs) offer potential for robot task planning, but challenges remain due to VLMs' tendency to generate incorrect action sequences. To address these limitations, we propose VeriGraph, a novel framework that integrates VLMs for robotic planning while verifying action feasibility. VeriGraph employs scene graphs as an intermediate representation, capturing key objects and spatial relationships to improve plan verification and refinement. The system generates a scene graph from input images and uses it to iteratively check and correct action sequences generated by an LLM-based task planner, ensuring constraints are respected and actions are executable. Our approach significantly enhances task completion rates across diverse manipulation scenarios, outperforming baseline methods by 58% for language-based tasks and 30% for image-based tasks.

Neural networks have shown remarkable performance in various tasks, yet they remain susceptible to subtle changes in their input or model parameters. One particularly impactful vulnerability arises through the Bit-Flip Attack (BFA), where flipping a small number of critical bits in a model's parameters can severely degrade its performance. A common technique for inducing bit flips in DRAM is the Row-Hammer attack, which exploits frequent uncached memory accesses to alter data. Identifying susceptible bits can be achieved through exhaustive search or progressive layer-by-layer analysis, especially in quantized networks. In this work, we introduce Impactful Bit-Flip Search (IBS), a novel method for efficiently pinpointing and flipping critical bits in full-precision networks. Additionally, we propose a Weight-Stealth technique that strategically modifies the model's parameters in a way that maintains the float values within the original distribution, thereby bypassing simple range checks often used in tamper detection.

Language models (LMs) are capable of acquiring elements of human-like syntactic knowledge. Targeted syntactic evaluation tests have been employed to measure how well they form generalizations about syntactic phenomena in high-resource languages such as English. However, we still lack a thorough understanding of LMs' capacity for syntactic generalizations in low-resource languages, which are responsible for much of the diversity of syntactic patterns worldwide. In this study, we develop targeted syntactic evaluation tests for three low-resource languages (Basque, Hindi, and Swahili) and use them to evaluate five families of open-access multilingual Transformer LMs. We find that some syntactic tasks prove relatively easy for LMs while others (agreement in sentences containing indirect objects in Basque, agreement across a prepositional phrase in Swahili) are challenging. We additionally uncover issues with publicly available Transformers, including a bias toward the habitual aspect in Hindi in multilingual BERT and underperformance compared to similar-sized models in XGLM-4.5B.

From CMB polarization data alone we reconstruct the CMB lensing power spectrum, comparable in overall constraining power to previous temperature-based reconstructions, and an unlensed E-mode power spectrum. The observations, taken in 2019 and 2020 with the South Pole Telescope (SPT) and the SPT-3G camera, cover 1500 deg$^2$ at 95, 150, and 220 GHz with arcminute resolution and roughly 4.9$\mu$K-arcmin coadded noise in polarization. The power spectrum estimates, together with systematic parameter estimates and a joint covariance matrix, follow from a Bayesian analysis using the Marginal Unbiased Score Expansion (MUSE) method. The E-mode spectrum at $\ell>2000$ and lensing spectrum at $L>350$ are the most precise to date. Assuming the $\Lambda$CDM model, and using only these SPT data and priors on $\tau$ and absolute calibration from Planck, we find $H_0=66.81\pm0.81$ km/s/Mpc, comparable in precision to the Planck determination and in 5.4$\sigma$ tension with the most precise $H_0$ inference derived via the distance ladder. We also find $S_8=0.850\pm0.017$, providing further independent evidence of a slight tension with low-redshift structure probes. The $\Lambda$CDM model provides a good simultaneous fit to the combined Planck, ACT, and SPT data, and thus passes a powerful test. Combining these CMB datasets with BAO observations, we find that the effective number of neutrino species, spatial curvature, and primordial helium fraction are consistent with standard model values, and that the 95% confidence upper limit on the neutrino mass sum is 0.075 eV. The SPT data are consistent with the somewhat weak preference for excess lensing power seen in Planck and ACT data relative to predictions of the $\Lambda$CDM model. We also detect at greater than 3$\sigma$ the influence of non-linear evolution in the CMB lensing power spectrum and discuss it in the context of the $S_8$ tension.(abridged), Comment: 28 pages, 21 figures + appendices

Active metasurfaces hold great promise for spatial light modulation, and electro-optic tuning using lithium niobate is particularly attractive due to its high transparency and well-established thin-film platform. In this work, we present a free-space transmissive light modulator based on a seemingly un-patterned thin-film lithium niobate-on-insulator platform, integrated with a transparent conductive oxide meta-grating fabricated through a single lithography process. Guided mode resonance in the near-infrared region is induced with high mode confinement within lithium niobate layer, which directly in contact with the electrodes. A notable resonance shift of 0.38 nm is observed for the fundamental mode under +-10 V bias, while a maximum modulation amplitude of 4.6% is achieved for another higher-order mode. Incident angle is further exploited as another tuning parameter to split and sensitively shift the resonances. These results demonstrate the potential of this design for applications in compact, scalable, and tunable spatial light modulation devices.

The detection of satellite thermal emission at millimeter wavelengths is presented using data from the 3rd-Generation receiver on the South Pole Telescope (SPT-3G). This represents the first reported detection of thermal emission from artificial satellites at millimeter wavelengths. Satellite thermal emission is shown to be detectable at high signal-to-noise on timescales as short as a few tens of milliseconds. An algorithm for downloading orbital information and tracking known satellites given observer constraints and time-ordered observatory pointing is described. Consequences for cosmological surveys and short-duration transient searches are discussed, revealing that the integrated thermal emission from all large satellites does not contribute significantly to the SPT-3G survey intensity map. Measured satellite positions are found to be discrepant from their two-line element (TLE) derived ephemerides up to several arcminutes which may present a difficulty in cross-checking or masking satellites from short-duration transient searches.

One of the most fundamental results in classical cryptography is that the existence of Pseudo-Random Generators (PRG) that expands $k$ bits of randomness to $k+1$ bits that are pseudo-random implies the existence of PRG that expand $k$ bits of randomness to $k+f(k)$ bits for any $f(k)=poly(k)$. It appears that cryptography in the quantum realm sometimes works differently than in the classical case. Pseudo-random quantum states (PRS) are a key primitive in quantum cryptography, that demonstrates this point. There are several open questions in quantum cryptography about PRS, one of them is - can we expand quantum pseudo-randomness in a black-box way with the same key length? Although this is known to be possible in the classical case, the answer in the quantum realm is more complex. This work conjectures that some PRS generators can be expanded, and provides a proof for such expansion for some specific examples. In addition, this work demonstrates the relationship between the key length required to expand the PRS, the efficiency of the circuit to create it and the length of the resulting expansion.

Intraoperative 2D quantitative angiography (QA) for intracranial aneurysms (IAs) has accuracy challenges due to the variability of hand injections. Despite the success of singular value decomposition (SVD) algorithms in reducing biases in computed tomography perfusion (CTP), their application in 2D QA has not been extensively explored. This study seeks to bridge this gap by investigating the potential of SVD-based deconvolution methods in 2D QA, particularly in addressing the variability of injection durations. The study included three internal carotid aneurysm (ICA) cases. Virtual angiograms were generated using Computational Fluid Dynamics (CFD) for three physiologically relevant inlet velocities to simulate contrast media injection durations. Time-density curves (TDCs) were produced for both the inlet and aneurysm dome. Various SVD variants, including standard SVD (sSVD) with and without classical Tikhonov regularization, block-circulant SVD (bSVD), and oscillation index SVD (oSVD), were applied to virtual angiograms. The method was applied on virtual angiograms to recover the aneurysmal dome impulse response function (IRF) and extract flow related parameters such as Peak Height PHIRF, Area Under the Curve AUCIRF, and Mean transit time MTT. Furthermore, we found that SVD can effectively reduce QA parameter variability across various injection durations, enhancing the potential of QA analysis parameters in neurovascular disease diagnosis and treatment. Implementing SVD-based deconvolution techniques in QA analysis can enhance the precision and reliability of neurovascular diagnostics by effectively reducing the impact of injection duration on hemodynamic parameters.

Training a policy that can generalize to unknown objects is a long standing challenge within the field of robotics. The performance of a policy often drops significantly in situations where an object in the scene was not seen during training. To solve this problem, we present NeRF-Aug, a novel method that is capable of teaching a policy to interact with objects that are not present in the dataset. This approach differs from existing approaches by leveraging the speed and photorealism of a neural radiance field for augmentation. NeRF- Aug both creates more photorealistic data and runs 3.83 times faster than existing methods. We demonstrate the effectiveness of our method on 4 tasks with 11 novel objects that have no expert demonstration data. We achieve an average 69.1% success rate increase over existing methods. See video results at https://nerf-aug.github.io.

Quantifying the global reach of planetarium dome shows presents significant challenges due to the lack of standardized viewership tracking mechanisms across diverse planetarium venues. We present an analysis of the global impact of dome shows, presenting data regarding four documentary films from a single visualization lab. Specifically, we designed and administered a viewership survey of four long-running shows that contained cinematic scientific visualizations. Reported survey data shows that between 1.2 - 2.6 million people have viewed these four films across the 68 responding planetariums (mean: 1.9 million). When we include estimates and extrapolate for the 315 planetariums that licensed these shows, we arrive at an estimate of 16.5 - 24.1 million people having seen these films (mean: 20.3 million).

Gate-based quantum computers are an innovative tool for experimentally studying the core principles of quantum mechanics. This work presents the first observation of quantum anomalous heat flow between two qubits and investigates the role of mid-circuit measurements in this context. Using mid-circuit measurements, we designed quantum circuits that violate the semi-classical heat flow bound, witnessing negativities in the underlying Kirkwood-Dirac quasiprobability distribution, which indicates the presence of quantum correlations between the subsystems. Mid-circuit measurements, crucial for probing qubits during the experiment, enabled these observations but also introduced disturbances, such as energy leakage, leading to deviations from theoretical predictions. We modeled these noise effects, providing insight into the limitations of current mid-circuit measurement techniques., Comment: 9 pages, 9 figures

We give an different proof of our result computing the stable homology of dihedral group Hurwitz spaces. This proof employs more elementary methods, instead of higher algebra., Comment: 32 pages. Comments welcome!

We compute the average number of surjections from class groups of quadratic function fields over $\mathbb F_q(t)$ onto finite odd order groups $H$, once $q$ is sufficiently large. These yield the first known moments of these class groups, as predicted by the Cohen--Lenstra heuristics, apart from the case $H = \mathbb Z/3\mathbb Z$. The key input to this result is a topological one, where we compute the stable rational homology groups of Hurwitz spaces associated to non-splitting conjugacy classes., Comment: 55 pages. Comments welcome!

Online paging is a fundamental problem in the field of online algorithms, in which one maintains a cache of $k$ slots as requests for fetching pages arrive online. In the weighted variant of this problem, each page has its own fetching cost; a substantial line of work on this problem culminated in an (optimal) $O(\log k)$-competitive randomized algorithm, due to Bansal, Buchbinder and Naor (FOCS'07). Existing work for weighted paging assumes that page weights are known in advance, which is not always the case in practice. For example, in multi-level caching architectures, the expected cost of fetching a memory block is a function of its probability of being in a mid-level cache rather than the main memory. This complex property cannot be predicted in advance; over time, however, one may glean information about page weights through sampling their fetching cost multiple times. We present the first algorithm for online weighted paging that does not know page weights in advance, but rather learns from weight samples. In terms of techniques, this requires providing (integral) samples to a fractional solver, requiring a delicate interface between this solver and the randomized rounding scheme; we believe that our work can inspire online algorithms to other problems that involve cost sampling.

GPT-4o is an autoregressive omni model that accepts as input any combination of text, audio, image, and video, and generates any combination of text, audio, and image outputs. It's trained end-to-end across text, vision, and audio, meaning all inputs and outputs are processed by the same neural network. GPT-4o can respond to audio inputs in as little as 232 milliseconds, with an average of 320 milliseconds, which is similar to human response time in conversation. It matches GPT-4 Turbo performance on text in English and code, with significant improvement on text in non-English languages, while also being much faster and 50\% cheaper in the API. GPT-4o is especially better at vision and audio understanding compared to existing models. In line with our commitment to building AI safely and consistent with our voluntary commitments to the White House, we are sharing the GPT-4o System Card, which includes our Preparedness Framework evaluations. In this System Card, we provide a detailed look at GPT-4o's capabilities, limitations, and safety evaluations across multiple categories, focusing on speech-to-speech while also evaluating text and image capabilities, and measures we've implemented to ensure the model is safe and aligned. We also include third-party assessments on dangerous capabilities, as well as discussion of potential societal impacts of GPT-4o's text and vision capabilities.

This paper focuses on creating synthetic data to improve the quality of image captions. Existing works typically have two shortcomings. First, they caption images from scratch, ignoring existing alt-text metadata, and second, lack transparency if the captioners' training data (e.g. GPT) is unknown. In this paper, we study a principled approach Altogether based on the key idea to edit and re-align existing alt-texts associated with the images. To generate training data, we perform human annotation where annotators start with the existing alt-text and re-align it to the image content in multiple rounds, consequently constructing captions with rich visual concepts. This differs from prior work that carries out human annotation as a one-time description task solely based on images and annotator knowledge. We train a captioner on this data that generalizes the process of re-aligning alt-texts at scale. Our results show our Altogether approach leads to richer image captions that also improve text-to-image generation and zero-shot image classification tasks., Comment: accepted by EMNLP 2024; MetaCLIPv2

We report results of a search for nuclear recoils induced by weakly interacting massive particle (WIMP) dark matter using the LUX-ZEPLIN (LZ) two-phase xenon time projection chamber. This analysis uses a total exposure of $4.2\pm0.1$ tonne-years from 280 live days of LZ operation, of which $3.3\pm0.1$ tonne-years and 220 live days are new. A technique to actively tag background electronic recoils from $^{214}$Pb $\beta$ decays is featured for the first time. Enhanced electron-ion recombination is observed in two-neutrino double electron capture decays of $^{124}$Xe, representing a noteworthy new background. After removal of artificial signal-like events injected into the data set to mitigate analyzer bias, we find no evidence for an excess over expected backgrounds. World-leading constraints are placed on spin-independent (SI) and spin-dependent WIMP-nucleon cross sections for masses $\geq$9 GeV/$c^2$. The strongest SI exclusion set is $2.1\times10^{-48}$ cm$^{2}$ at the 90% confidence level at a mass of 36 GeV/$c^2$, and the best SI median sensitivity achieved is $5.0\times10^{-48}$ cm$^{2}$ for a mass of 40 GeV/$c^2$., Comment: 9 pages, 7 figures. See https://www.hepdata.net/record/155182 for a data release related to this paper

In any scalar-tensor theory of gravity exhibiting a screening mechanism, the fifth force mediated by the scalar field is dynamically suppressed at sub-Solar system scales, allowing it to pass existing tests of gravity. As a result, a major research effort has been carried out over the past decades to 'outsmart' screened scalars in this game of hide-and-seek. While most of such tests rely on fifth force effects, one should keep in mind that the latter are by no means the only physical feature of scalar-tensor gravity. In particular, this article investigates the possibility of testing screened scalar-tensor models by means of gravitational redshift measurements performed with atomic clocks. Upon deriving the expression for the redshift in this framework, we propose a thought experiment for testing the chameleon model by clock comparisons, which guides us towards more realistic experimental setups, in the laboratory and in space. We find that the current level of precision exhibited by leading-edge technology is still too low to yield competitive constraints on the chameleon model., Comment: 22 pages, 8 figures

Scaling inference compute in large language models (LLMs) through repeated sampling consistently increases the coverage (fraction of problems solved) as the number of samples increases. We conjecture that this observed improvement is partially due to the answer distribution of standard evaluation benchmarks, which is skewed towards a relatively small set of common answers. To test this conjecture, we define a baseline that enumerates answers according to their prevalence in the training set. Experiments spanning two domains -- mathematical reasoning and factual knowledge -- reveal that this baseline outperforms repeated model sampling for some LLMs, while the coverage for others is on par with that of a mixture strategy that obtains $k$ answers by using only $10$ model samples and similarly guessing the remaining $k-10$ attempts via enumeration. Our baseline enables a more accurate measurement of how much repeated sampling improves coverage in such settings beyond prompt-agnostic guessing.

Precise action localization in untrimmed video is vital for fields such as professional sports and minimally invasive surgery, where the delineation of particular motions in recordings can dramatically enhance analysis. But in many cases, large scale datasets with video-label pairs for localization are unavailable, limiting the opportunity to fine-tune video-understanding models. Recent developments in large vision-language models (LVLM) address this need with impressive zero-shot capabilities in a variety of video understanding tasks. However, the adaptation of image-based LVLMs, with their powerful visual question answering capabilities, to action localization in long-form video is still relatively unexplored. To this end, we introduce a true ZEro-shot Action Localization method (ZEAL). Specifically, we leverage the built-in action knowledge of a large language model (LLM) to inflate actions into highly-detailed descriptions of the archetypal start and end of the action. These descriptions serve as queries to LVLM for generating frame-level confidence scores which can be aggregated to produce localization outputs. The simplicity and flexibility of our method lends it amenable to more capable LVLMs as they are developed, and we demonstrate remarkable results in zero-shot action localization on a challenging benchmark, without any training.

With the increasing adoption of large language models (LLMs) in education, concerns about inherent biases in these models have gained prominence. We evaluate LLMs for bias in the personalized educational setting, specifically focusing on the models' roles as "teachers". We reveal significant biases in how models generate and select educational content tailored to different demographic groups, including race, ethnicity, sex, gender, disability status, income, and national origin. We introduce and apply two bias score metrics--Mean Absolute Bias (MAB) and Maximum Difference Bias (MDB)--to analyze 9 open and closed state-of-the-art LLMs. Our experiments, which utilize over 17,000 educational explanations across multiple difficulty levels and topics, uncover that models perpetuate both typical and inverted harmful stereotypes., Comment: 46 Pages, 55 Figures, dataset release pending publication

We develop a new longitudinal count data regression model that accounts for zero-inflation and spatio-temporal correlation across responses. This project is motivated by an analysis of Iowa Fluoride Study (IFS) data, a longitudinal cohort study with data on caries (cavity) experience scores measured for each tooth across five time points. To that end, we use a hurdle model for zero-inflation with two parts: the presence model indicating whether a count is non-zero through logistic regression and the severity model that considers the non-zero counts through a shifted Negative Binomial distribution allowing overdispersion. To incorporate dependence across measurement occasion and teeth, these marginal models are embedded within a Gaussian copula that introduces spatio-temporal correlations. A distinct advantage of this formulation is that it allows us to determine covariate effects with population-level (marginal) interpretations in contrast to mixed model choices. Standard Bayesian sampling from such a model is infeasible, so we use approximate Bayesian computing for inference. This approach is applied to the IFS data to gain insight into the risk factors for dental caries and the correlation structure across teeth and time.

Large language models (LLMs) frequently make errors when handling even simple numerical problems, such as comparing two small numbers. A natural hypothesis is that these errors stem from how LLMs represent numbers, and specifically, whether their representations of numbers capture their numeric values. We tackle this question from the observation that LLM errors on numerical tasks are often distributed across \textit{the digits} of the answer rather than normally around \textit{its numeric value}. Through a series of probing experiments and causal interventions, we show that LLMs internally represent numbers with individual circular representations per-digit in base 10. This digit-wise representation, as opposed to a value representation, sheds light on the error patterns of models on tasks involving numerical reasoning and could serve as a basis for future studies on analyzing numerical mechanisms in LLMs.

Generative AI's rapid advancement sparks interest in its cognitive abilities, especially given its capacity for tasks like language understanding and code generation. This study explores how several recent GenAI models perform on the Clock Drawing Test (CDT), a neuropsychological assessment of visuospatial planning and organization. While models create clock-like drawings, they struggle with accurate time representation, showing deficits similar to mild-severe cognitive impairment (Wechsler, 2009). Errors include numerical sequencing issues, incorrect clock times, and irrelevant additions, despite accurate rendering of clock features. Only GPT 4 Turbo and Gemini Pro 1.5 produced the correct time, scoring like healthy individuals (4/4). A follow-up clock-reading test revealed only Sonnet 3.5 succeeded, suggesting drawing deficits stem from difficulty with numerical concepts. These findings may reflect weaknesses in visual-spatial understanding, working memory, or calculation, highlighting strengths in learned knowledge but weaknesses in reasoning. Comparing human and machine performance is crucial for understanding AI's cognitive capabilities and guiding development toward human-like cognitive functions.

Empowerment has the potential to help agents learn large skillsets, but is not yet a scalable solution for training general-purpose agents. Recent empowerment methods learn diverse skillsets by maximizing the mutual information between skills and states; however, these approaches require a model of the transition dynamics, which can be challenging to learn in realistic settings with high-dimensional and stochastic observations. We present Latent-Predictive Empowerment (LPE), an algorithm that can compute empowerment in a more practical manner. LPE learns large skillsets by maximizing an objective that is a principled replacement for the mutual information between skills and states and that only requires a simpler latent-predictive model rather than a full simulator of the environment. We show empirically in a variety of settings--including ones with high-dimensional observations and highly stochastic transition dynamics--that our empowerment objective (i) learns similar-sized skillsets as the leading empowerment algorithm that assumes access to a model of the transition dynamics and (ii) outperforms other model-based approaches to empowerment.

Large language models (LLMs) acquire beliefs about gender from training data and can therefore generate text with stereotypical gender attitudes. Prior studies have demonstrated model generations favor one gender or exhibit stereotypes about gender, but have not investigated the complex dynamics that can influence model reasoning and decision-making involving gender. We study gender equity within LLMs through a decision-making lens with a new dataset, DeMET Prompts, containing scenarios related to intimate, romantic relationships. We explore nine relationship configurations through name pairs across three name lists (men, women, neutral). We investigate equity in the context of gender roles through numerous lenses: typical and gender-neutral names, with and without model safety enhancements, same and mixed-gender relationships, and egalitarian versus traditional scenarios across various topics. While all models exhibit the same biases (women favored, then those with gender-neutral names, and lastly men), safety guardrails reduce bias. In addition, models tend to circumvent traditional male dominance stereotypes and side with 'traditionally female' individuals more often, suggesting relationships are viewed as a female domain by the models., Comment: EMNLP Findings 2024

Traditionally, model-based reinforcement learning (MBRL) methods exploit neural networks as flexible function approximators to represent $\textit{a priori}$ unknown environment dynamics. However, training data are typically scarce in practice, and these black-box models often fail to generalize. Modeling architectures that leverage known physics can substantially reduce the complexity of system-identification, but break down in the face of complex phenomena such as contact. We introduce a novel framework for learning semi-structured dynamics models for contact-rich systems which seamlessly integrates structured first principles modeling techniques with black-box auto-regressive models. Specifically, we develop an ensemble of probabilistic models to estimate external forces, conditioned on historical observations and actions, and integrate these predictions using known Lagrangian dynamics. With this semi-structured approach, we can make accurate long-horizon predictions with substantially less data than prior methods. We leverage this capability and propose Semi-Structured Reinforcement Learning ($\texttt{SSRL}$) a simple model-based learning framework which pushes the sample complexity boundary for real-world learning. We validate our approach on a real-world Unitree Go1 quadruped robot, learning dynamic gaits -- from scratch -- on both hard and soft surfaces with just a few minutes of real-world data. Video and code are available at: https://sites.google.com/utexas.edu/ssrl, Comment: v2: corrected typos in eqs (1) and (3); add CoRL footnote

In this paper, we establish the global convergence of the actor-critic algorithm with a significantly improved sample complexity of $O(\epsilon^{-3})$, advancing beyond the existing local convergence results. Previous works provide local convergence guarantees with a sample complexity of $O(\epsilon^{-2})$ for bounding the squared gradient of the return, which translates to a global sample complexity of $O(\epsilon^{-4})$ using the gradient domination lemma. In contrast to traditional methods that employ decreasing step sizes for both the actor and critic, we demonstrate that a constant step size for the critic is sufficient to ensure convergence in expectation. This key insight reveals that using a decreasing step size for the actor alone is sufficient to handle the noise for both the actor and critic. Our findings provide theoretical support for the practical success of many algorithms that rely on constant step sizes.

Recent advancements in LLM-based web agents have introduced novel architectures and benchmarks showcasing progress in autonomous web navigation and interaction. However, most existing benchmarks prioritize effectiveness and accuracy, overlooking crucial factors like safety and trustworthiness which are essential for deploying web agents in enterprise settings. The risks of unsafe web agent behavior, such as accidentally deleting user accounts or performing unintended actions in critical business operations, pose significant barriers to widespread adoption. In this paper, we present ST-WebAgentBench, a new online benchmark specifically designed to evaluate the safety and trustworthiness of web agents in enterprise contexts. This benchmark is grounded in a detailed framework that defines safe and trustworthy (ST) agent behavior, outlines how ST policies should be structured and introduces the Completion under Policies metric to assess agent performance. Our evaluation reveals that current SOTA agents struggle with policy adherence and cannot yet be relied upon for critical business applications. Additionally, we propose architectural principles aimed at improving policy awareness and compliance in web agents. We open-source this benchmark and invite the community to contribute, with the goal of fostering a new generation of safer, more trustworthy AI agents. All code, data, environment reproduction resources, and video demonstrations are available at https://sites.google.com/view/st-webagentbench/home.

There is increasing interest in tracking the capabilities of general intelligence foundation models. This study benchmarks leading large language models and vision language models against human performance on the Wechsler Adult Intelligence Scale (WAIS-IV), a comprehensive, population-normed assessment of underlying human cognition and intellectual abilities, with a focus on the domains of VerbalComprehension (VCI), Working Memory (WMI), and Perceptual Reasoning (PRI). Most models demonstrated exceptional capabilities in the storage, retrieval, and manipulation of tokens such as arbitrary sequences of letters and numbers, with performance on the Working Memory Index (WMI) greater or equal to the 99.5th percentile when compared to human population normative ability. Performance on the Verbal Comprehension Index (VCI) which measures retrieval of acquired information, and linguistic understanding about the meaning of words and their relationships to each other, also demonstrated consistent performance at or above the 98th percentile. Despite these broad strengths, we observed consistently poor performance on the Perceptual Reasoning Index (PRI; range 0.1-10th percentile) from multimodal models indicating profound inability to interpret and reason on visual information. Smaller and older model versions consistently performed worse, indicating that training data, parameter count and advances in tuning are resulting in significant advances in cognitive ability.

Atomic beams offer measurement capabilities for hot vapors but often require complex, bulky systems, limiting miniaturization. This study explores hot vapors in thin cells (L=30,5 microns) as a simplified analogy to atomic beams. Coherence is maintained only for atoms moving parallel to the cell walls, as wall collisions randomize spin. By filtering non-parallel atoms, we mimic atomic beam behavior and observe the Faraday-Ramsey effect. Rubidium atoms are optically pumped with linearly polarized resonance radiation, evolving in a perpendicular magnetic field and probed via Faraday rotation. We believe that thin cell technology is paving the way for the miniaturization of atomic beam-based sensors., Comment: This PDF submission, intended for consideration by Optica Quantum, combines the main text and Supplementary Information (SI). The document includes 10 pages and 7 figures of the main text, followed by references and then 6 pages with 4 figures of the SI. This arrangement facilitates easy navigation and aligns with journal requirements

Error correction codes (ECC) are crucial for ensuring reliable information transmission in communication systems. Choukroun & Wolf (2022b) recently introduced the Error Correction Code Transformer (ECCT), which has demonstrated promising performance across various transmission channels and families of codes. However, its high computational and memory demands limit its practical applications compared to traditional decoding algorithms. Achieving effective quantization of the ECCT presents significant challenges due to its inherently small architecture, since existing, very low-precision quantization techniques often lead to performance degradation in compact neural networks. In this paper, we introduce a novel acceleration method for transformer-based decoders. We first propose a ternary weight quantization method specifically designed for the ECCT, inducing a decoder with multiplication-free linear layers. We present an optimized self-attention mechanism to reduce computational complexity via codeaware multi-heads processing. Finally, we provide positional encoding via the Tanner graph eigendecomposition, enabling a richer representation of the graph connectivity. The approach not only matches or surpasses ECCT's performance but also significantly reduces energy consumption, memory footprint, and computational complexity. Our method brings transformer-based error correction closer to practical implementation in resource-constrained environments, achieving a 90% compression ratio and reducing arithmetic operation energy consumption by at least 224 times on modern hardware.

Solution-based soft matter self-assembly (SA) promises unique materials properties from approaches including additive manufacturing/three-dimensional (3D) printing. We report direct ink writing derived, hierarchically porous transition metal nitride superconductors (SCs) and precursor oxides, structure-directed by Pluronics-family block copolymer (BCP) SA and heat treated in various environments. SCs with periodic lattices on three length scales show record nanoconfinement-induced upper critical field enhancements correlated with BCP molar mass, attaining values of 50 T for NbN and 8.1 T for non-optimized TiN samples, the first mapping of a tailorable SC property onto a macromolecular parameter. They reach surface areas above 120 m$^2$/g, the highest reported for compound SCs to date. Embedded printing enables the first BCP directed mesoporous non-self-supporting helical SCs. Results suggest that additive manufacturing may open pathways to mesoporous SCs with not only a variety of macroscopic form factors but enhanced properties from intrinsic, SA-derived mesostructures with substantial academic and technological promise.

Tensor networks are a compressed format for multi-dimensional data. One-dimensional tensor networks -- often referred to as tensor trains (TT) or matrix product states (MPS) -- are increasingly being used as a numerical ansatz for continuum functions by "quantizing" the inputs into discrete binary digits. Here we demonstrate the power of more general tree tensor networks for this purpose. We provide direct constructions of a number of elementary functions as generic tree tensor networks and interpolative constructions for more complicated functions via a generalization of the tensor cross interpolation algorithm. For a range of multi-dimensional functions we show how more structured tree tensor networks offer a significantly more efficient ansatz than the commonly used tensor train. We demonstrate an application of our methods to solving multi-dimensional, non-linear Fredholm equations, providing a rigorous bound on the rank of the solution which, in turn, guarantees exponentially scaling accuracy with the size of the tree tensor network for certain problems., Comment: Main text: 11 Pages and 6 Figures

Quantization has established itself as the primary approach for decreasing the computational and storage expenses associated with Large Language Models (LLMs) inference. The majority of current research emphasizes quantizing weights and activations to enable low-bit general-matrix-multiply (GEMM) operations, with the remaining non-linear operations executed at higher precision. In our study, we discovered that following the application of these techniques, the primary bottleneck in LLMs inference lies in the softmax layer. The softmax operation comprises three phases: exponent calculation, accumulation, and normalization, Our work focuses on optimizing the first two phases. We propose an analytical approach to determine the optimal clipping value for the input to the softmax function, enabling sub-4-bit quantization for LLMs inference. This method accelerates the calculations of both $e^x$ and $\sum(e^x)$ with minimal to no accuracy degradation. For example, in LLaMA1-30B, we achieve baseline performance with 2-bit quantization on the well-known "Physical Interaction: Question Answering" (PIQA) dataset evaluation. This ultra-low bit quantization allows, for the first time, an acceleration of approximately 4x in the accumulation phase. The combination of accelerating both $e^x$ and $\sum(e^x)$ results in a 36.9% acceleration in the softmax operation.

Self-service machines are a form of pseudo-automation; rather than actually automate tasks, they offset them to unpaid customers. Typically implemented for customer convenience and to reduce labor costs, self-service is often criticized for worsening customer service and increasing loss and theft for retailers. Though millions of frontline service workers continue to interact with these technologies on a day-to-day basis, little is known about how these machines change the nature of frontline labor. Through interviews with current and former cashiers who work with self-checkout technologies, we investigate how technology that offsets labor from an employee to a customer can reconfigure frontline work. We find three changes to cashiering tasks as a result of self-checkout: (1) Working at self-checkout involved parallel demands from multiple customers, (2) self-checkout work was more problem-oriented (including monitoring and policing customers), and (3) traditional checkout began to become more demanding as easier transactions were filtered to self-checkout. As their interactions with customers became more focused on problem solving and rule enforcement, cashiers were often positioned as adversaries to customers at self-checkout. To cope with perceived adversarialism, cashiers engaged in a form of relational patchwork, using techniques like scapegoating the self-checkout machine and providing excessive customer service in order to maintain positive customer interactions in the face of potential conflict. Our findings highlight how even under pseudo-automation, workers must engage in relational work to manage and mend negative human-to-human interactions so that machines can be properly implemented in context., Comment: Forthcoming in the Proceedings of the 2025 Conference on Computer Supported Cooperative Work and Social Computing

A wide family of two dimensional (2D) systems, including stripe-phase superconductors, sliding Luttinger liquids, and anisotropic 2D materials, can be modeled by an array of coupled one-dimensional (1D) electron channels or nanowire arrays. Here we report experiments in arrays of conducting nanowires with gate and field tunable interwire coupling, that are programmed at the LaAlO$_3$/SrTiO$_3$ interface. We find a magnetically-tuned metal-to-insulator transition in which the transverse resistance of the nanowire array increases by up to four orders of magnitude. To explain this behavior, we develop a minimal model of a coupled two-wire system which agrees well with observed phenomena. These nanowire arrays can serve as a model systems to understand the origin of exotic behavior in correlated materials via analog quantum simulation.

Coupled quasi-one-dimensional (quasi-1D) electron systems host rich emergent physics that cannot be accounted for by understanding isolated 1D electron systems alone. Open questions remain about how transport in these arrays can be manipulated by the application of external electric and magnetic fields. In this theoretical study, we consider a pair of coupled nanowires with non-interacting electrons. We find that a metal-insulator transition is induced by an out-of-plane magnetic field and a transverse potential bias on an array of such coupled wires. We demonstrate the existence of distinct conductance features and highlight the crucial role played by the field dependence of the interwire potential barrier on transport properties. These predictions agree well with transport experiments performed on coupled nanowires sketched on an LaAlO3/SrTiO3 interface. Since our model makes minimal assumptions, we expect our predictions to hold for a wide class of coupled 1D systems.

Social dynamics in close human interactions pose significant challenges for Human Mesh Estimation (HME), particularly due to the complexity of physical contacts and the scarcity of training data. Addressing these challenges, we introduce a novel data generation method that utilizes Large Vision Language Models (LVLMs) to annotate contact maps which guide test-time optimization to produce paired image and pseudo-ground truth meshes. This methodology not only alleviates the annotation burden but also enables the assembly of a comprehensive dataset specifically tailored for close interactions in HME. Our Ask Pose Unite (APU) dataset, comprising over 6.2k human mesh pairs in contact covering diverse interaction types, is curated from images depicting naturalistic person-to-person scenes. We empirically show that using our dataset to train a diffusion-based contact prior, used as guidance during optimization, improves mesh estimation on unseen interactions. Our work addresses longstanding challenges of data scarcity for close interactions in HME enhancing the field's capabilities of handling complex interaction scenarios., Comment: Project webpage: https://laubravo.github.io/apu_website/