Search

In "Embers of Autoregression" (McCoy et al., 2023), we showed that several large language models (LLMs) have some important limitations that are attributable to their origins in next-word prediction. Here we investigate whether these issues persist with o1, a new system from OpenAI that differs from previous LLMs in that it is optimized for reasoning. We find that o1 substantially outperforms previous LLMs in many cases, with particularly large improvements on rare variants of common tasks (e.g., forming acronyms from the second letter of each word in a list, rather than the first letter). Despite these quantitative improvements, however, o1 still displays the same qualitative trends that we observed in previous systems. Specifically, o1 -- like previous LLMs -- is sensitive to the probability of examples and tasks, performing better and requiring fewer "thinking tokens" in high-probability settings than in low-probability ones. These results show that optimizing a language model for reasoning can mitigate but might not fully overcome the language model's probability sensitivity., Comment: 6 pages; updated to fix typo in Fig 4 caption

Tin oxide aerogels were synthesized using an epoxide-assisted technique and characterized with Fourier transform infrared, X-ray diffraction, and UV-Vis to study the effects of post-synthesis annealing and peroxide treatment. While bulk tin oxide exhibits an optical bandgap of $3.6$ eV, its aerogel form often displays a larger apparent bandgap around $4.6$ eV due to defects. Our study reveals that annealing induces a partial phase change from SnO$_2$ to SnO, but is ineffective in removing defects. Conversely, peroxide passivation effectively lowers the bandgap and disorder levels, suggesting that dangling bonds are the primary cause of the increased bandgap in tin oxide aerogels. These findings offer insights for optimizing the optical properties of tin oxide aerogels for applications like solar cells.

Terrain awareness is an essential milestone to enable truly autonomous off-road navigation. Accurately predicting terrain characteristics allows optimizing a vehicle's path against potential hazards. Recent methods use deep neural networks to predict traversability-related terrain properties in a self-supervised manner, relying on proprioception as a training signal. However, onboard cameras are inherently limited by their point-of-view relative to the ground, suffering from occlusions and vanishing pixel density with distance. This paper introduces a novel approach for self-supervised terrain characterization using an aerial perspective from a hovering drone. We capture terrain-aligned images while sampling the environment with a ground vehicle, effectively training a simple predictor for vibrations, bumpiness, and energy consumption. Our dataset includes 2.8 km of off-road data collected in forest environment, comprising 13 484 ground-based images and 12 935 aerial images. Our findings show that drone imagery improves terrain property prediction by 21.37 % on the whole dataset and 37.35 % in high vegetation, compared to ground robot images. We conduct ablation studies to identify the main causes of these performance improvements. We also demonstrate the real-world applicability of our approach by scouting an unseen area with a drone, planning and executing an optimized path on the ground., Comment: 7 pages, 5 figures, submitted to ICRA 2025

The second moment method is a linear acceleration technique which couples the transport equation to a diffusion equation with transport-dependent additive closures. The resulting low-order diffusion equation can be discretized independent of the transport discretization, unlike diffusion synthetic acceleration, and is symmetric positive definite, unlike quasi-diffusion. While this method has been shown to be comparable to quasi-diffusion in iterative performance for fixed source and time-dependent problems, it is largely unexplored as an eigenvalue problem acceleration scheme due to thought that the resulting inhomogeneous source makes the problem ill-posed. Recently, a preliminary feasibility study was performed on the second moment method for eigenvalue problems. The results suggested comparable performance to quasi-diffusion and more robust performance than diffusion synthetic acceleration. This work extends the initial study to more realistic reactor problems using state-of-the-art discretization techniques. Results in this paper show that the second moment method is more computationally efficient than its alternatives on complex reactor problems with unstructured meshes., Comment: 29 pages, 5 tables, 7 figures

Magnetic skyrmion crystals with zero net skyrmion charge and zero topological Hall response are interesting candidate phases which can occur at a vanishing magnetic field in centrosymmetric systems. We study a minimal bilayer model of skyrmion crystals having opposite chirality and topological charge in the two layers, and show that it can host nearly flat electronic bands with quasi-uniform Berry curvature and quantum metric. Using Hartree-Fock theory, we show that weak to moderate short-range electron interactions induce two distinct types of symmetry breaking patterns depending on the band dispersion: an intra-unit-cell charge density modulation from Chern band mixing or a layer-imbalanced phase with a nonzero ferroelectric polarization. Both phases break inversion symmetry leading to a spontaneous and large net topological Hall effect, with the phase diagram tunable by external electric fields. Our results may be relevant to centrosymmetric skyrmion materials such as Gd$_2$PdSi$_3$ and Gd$_3$Ru$_4$Al$_{12}$ as well as artificially engineered heterostructures. We also discuss its relation to recent work on twisted transition metal dichalcogenide bilayers.

The computational complexity of deep learning algorithms has given rise to significant speed and memory challenges for the execution hardware. In energy-limited portable devices, highly efficient processing platforms are indispensable for reproducing the prowess afforded by much bulkier processing platforms. In this work, we present a low-power Leaky Integrate-and-Fire (LIF) neuron design fabricated in TSMC's 28 nm CMOS technology as proof of concept to build an energy-efficient mixed-signal Neuromorphic System-on-Chip (NeuroSoC). The fabricated neuron consumes 1.61 fJ/spike and occupies an active area of 34 $\mu m^{2}$, leading to a maximum spiking frequency of 300 kHz at 250 mV power supply. These performances are used in a software model to emulate the dynamics of a Spiking Neural Network (SNN). Employing supervised backpropagation and a surrogate gradient technique, the resulting accuracy on the MNIST dataset, using 4-bit post-training quantization stands at 82.5\%. The approach underscores the potential of such ASIC implementation of quantized SNNs to deliver high-performance, energy-efficient solutions to various embedded machine-learning applications.

Artificial Intelligence (AI) is revolutionizing various fields, including public health surveillance. In Africa, where health systems frequently encounter challenges such as limited resources, inadequate infrastructure, failed health information systems and a shortage of skilled health professionals, AI offers a transformative opportunity. This paper investigates the applications of AI in public health surveillance across the continent, presenting successful case studies and examining the benefits, opportunities, and challenges of implementing AI technologies in African healthcare settings. Our paper highlights AI's potential to enhance disease monitoring and health outcomes, and support effective public health interventions. The findings presented in the paper demonstrate that AI can significantly improve the accuracy and timeliness of disease detection and prediction, optimize resource allocation, and facilitate targeted public health strategies. Additionally, our paper identified key barriers to the widespread adoption of AI in African public health systems and proposed actionable recommendations to overcome these challenges.

We solve a model of electrons with Hubbard-$U$ Coulomb repulsion and a random Yukawa coupling to a two-dimensional bosonic bath, using an extended dynamical mean field theory scheme. Our model exhibits a quantum critical point, at which the repulsive component of the electron interactions strongly enhances the effects of the quantum critical bosonic fluctuations on the electrons, leading to a breakdown of Fermi liquid physics and the formation of a strange metal with `Planckian' ($\mathcal{O}(k_B T/\hbar)$) quasiparticle decay rates at low temperatures $T\rightarrow 0$. Furthermore, the eventual Mott transition that occurs as the repulsion is increased seemingly bounds the maximum decay rate in the strange metal. Our results provide insight into low-temperature strange metallicity observed in proximity to a Mott transition, as is observed, for instance, in recent experiments on certain moir\'{e} materials., Comment: 4 pages, 4 figures

We present a two-phase field model and a hybrid particle-phase field model to simulate dilute colloidal sedimentation and flotation near a liquid-gas interface (or fluid-fluid interface in general). Both models are coupled to the incompressible Stokes equation, which is solved numerically using a combination of sine and regular Fourier transforms to account for the no-slip boundary conditions at the boundaries. The continuum two-phase field model allows us to analytically solve the equilibrium interfacial profile using a perturbative approach, demonstrating excellent agreement with numerical simulations. Notably, we show that strong coupling to particle dynamics can significantly alter the liquid-gas interface, thereby modifying the liquid-gas interfacial tension. In particular, we show that the renormalized surface tension is monotonically decreasing with increasing colloidal particle concentration and decreasing buoyant mass.

A magnitude-least-squares radiofrequency pulse design algorithm is reported which uses interleaved exact and stochastically-generated inexact updates to escape local minima and find low-cost solutions. Inexact updates are performed using a small randomly selected minibatch of the available B1+ measurements to update RF pulse weights, which perturbs the sequence of alternating projections. Applications to RF shimming, parallel transmit spokes RF pulse design, and spectral-spatial RF pulse design are considered. Numerical and simulation studies characterized the optimal minibatch size, which was found to consistently produce lower power and lower RMSE solutions across subjects, coil geometries, B1+ resolutions and orientations. The method was validated in-vivo at 7 Tesla and produced improvements in image quality in a slice-by-slice RF-shimmed imaging sequence. Compared to conventional methods, the pulse design method can more robustly design RF pulses that correct for B1+ inhomogeneities at ultra-high field strengths, and enable pulse designs to be completed with increased computational efficiency, Comment: 14 pages (including initial blank page), 15 figures (including 2 tables)

We provide practical simulation methods for scalar field theories on a quantum computer that yield improved asymptotics as well as concrete gate estimates for the simulation and physical qubit estimates using the surface code. We achieve these improvements through two optimizations. First, we consider a different approach for estimating the elements of the S-matrix. This approach is appropriate in general for 1+1D and for certain low-energy elastic collisions in higher dimensions. Second, we implement our approach using a series of different fault-tolerant simulation algorithms for Hamiltonians formulated both in the field occupation basis and field amplitude basis. Our algorithms are based on either second-order Trotterization or qubitization. The cost of Trotterization in occupation basis scales as $\widetilde{O}(\lambda N^7 |\Omega|^3/(M^{5/2} \epsilon^{3/2})$ where $\lambda$ is the coupling strength, $N$ is the occupation cutoff $|\Omega|$ is the volume of the spatial lattice, $M$ is the mass of the particles and $\epsilon$ is the uncertainty in the energy calculation used for the $S$-matrix determination. Qubitization in the field basis scales as $\widetilde{O}(|\Omega|^2 (k^2 \Lambda +kM^2)/\epsilon)$ where $k$ is the cutoff in the field and $\Lambda$ is a scaled coupling constant. We find in both cases that the bounds suggest physically meaningful simulations can be performed using on the order of $4\times 10^6$ physical qubits and $10^{12}$ $T$-gates which corresponds to roughly one day on a superconducting quantum computer with surface code and a cycle time of 100 ns, placing simulation of scalar field theory within striking distance of the gate counts for the best available chemistry simulation results., Comment: main text, 50 pages, supplementary 64 pages

Typical schemes for automated red-teaming large language models (LLMs) focus on discovering prompts that trigger a frozen language model (the defender) to generate toxic text. This often results in the prompting model (the adversary) producing text that is unintelligible and unlikely to arise. Here, we propose a reinforcement learning formulation of the LLM red-teaming task which allows us to discover prompts that both (1) trigger toxic outputs from a frozen defender and (2) have low perplexity as scored by the defender. We argue these cases are most pertinent in a red-teaming setting because of their likelihood to arise during normal use of the defender model. We solve this formulation through a novel online and weakly supervised variant of Identity Preference Optimization (IPO) on GPT-2 and GPT-2 XL defenders. We demonstrate that our policy is capable of generating likely prompts that also trigger toxicity. Finally, we qualitatively analyze learned strategies, trade-offs of likelihood and toxicity, and discuss implications. Source code is available for this project at: https://github.com/sisl/ASTPrompter/., Comment: 9 pages, 2 tables, 2 figures

Hot Jupiter atmospheres are possibly subject to a thermoresistive instability. Such an instability may develop as the ohmic heating increases the electrical conductivity in a positive feedback loop, which ultimately leads to a runaway of the atmospheric temperature. We extend our previous axisymmetric one-dimensional radial model, by representing the temperature and magnetic diffusivity as a first order Fourier expansion in longitude. This allows us to predict the hot spot offset during the unfolding of the thermoresistive instability and following Alfv\'enic oscillations. We show a representative simulation undergoing the thermoresistive instability, in which the peak flux offset varies between approximately $\pm 60^{\circ}$ on timescales of a few days with potentially observable brightness variations. Therefore, this thermoresistive instability could be an observable feature of hot Jupiters, given the right timing of observation and transit and the right planetary parameters., Comment: 12 pages, 8 figures, 1 table

We propose a possible nonlocal approximation of the Willmore functional, in the sense of Gamma-convergence, based on the first variation ot the fractional Allen-Cahn energies, and we prove the corresponding $\Gamma$-limsup estimate. Our analysis is based on the expansion of the fractional Laplacian in Fermi coordinates and fine estimates on the decay of higher order derivatives of the one-dimensional nonlocal optimal profile. This result is the nonlocal counterpart of that obtained by Bellettini and Paolini, where they proposed a phase-field approximation of the Willmore functional based on the first variation of the (local) Allen-Cahn energies.

An ultra-light axion with CP violating interactions with a dark sector and CP preserving interactions with the visible sector can act as a novel portal between dark matter and the Standard Model. In such theories, dark matter sources an axion field extending over the entire galaxy, the gradient of which can be searched for with precise spin precession experiments. A reinterpretation of existing co-magnetometer data already constrains theories that are consistent with astrophysical bounds, and near-future experiments will begin probing well-motivated models. The required interactions can arise from a confining hidden sector without necessitating fine-tuning of the axion's mass., Comment: 6 pages, 1 figure; comments welcome!

The Causaloid framework is an operational approach aimed to house both the radical aspects of General Relativity -- dynamic causal structure, and Quantum Theory -- indefiniteness, to provide a scaffolding that might be suitable for Quantum Gravity by providing a landscape of theories that allow for indefinite causal structure. One may consider it as a generalisation of generalised probability theories (or GPTs) where a priori regions are not assumed to have any given causal relationship, to incorporate the possibility of indefinite causal structure. Since its conception, there have been many advances in the field of indefinite causal structure mostly stemming from the work of Chiribella et al. on the quantum switch and supermaps and from Oreshkov et al. on causal inequalities and process matrices. These approaches have systems moving along wires and use Hilbert space structure. They violate the standard causality constraints of Quantum Theory and, in this sense, can be regarded as post-quantum. The Causaloid approach does not necessarily have systems moving along wires or Hilbert spaces. This is the first paper in a trilogy of papers aiming to close the gap between the Causaloid (that allows for GPTs) and post-quantum studies that employ Hilbert spaces. To do so in the present paper, we provide a diagrammatic language for the Causaloid framework along with new terminology for the three levels of physical compression (called Tomographic, Compositional, and Meta compression)., Comment: 29 pages, three figures, and many diagrammatic equations

Neutrinoless double beta decay is one of the most sensitive probes for new physics beyond the Standard Model of particle physics. One of the isotopes under investigation is $^{136}$Xe, which would double beta decay into $^{136}$Ba. Detecting the single $^{136}$Ba daughter provides a sort of ultimate tool in the discrimination against backgrounds. Previous work demonstrated the ability to perform single atom imaging of Ba atoms in a single-vacancy site of a solid xenon matrix. In this paper, the effort to identify signal from individual barium atoms is extended to Ba atoms in a hexa-vacancy site in the matrix and is achieved despite increased photobleaching in this site. Abrupt fluorescence turn-off of a single Ba atom is also observed. Significant recovery of fluorescence signal lost through photobleaching is demonstrated upon annealing of Ba deposits in the Xe ice. Following annealing, it is observed that Ba atoms in the hexa-vacancy site exhibit antibleaching while Ba atoms in the tetra-vacancy site exhibit bleaching. This may be evidence for a matrix site transfer upon laser excitation. Our findings offer a path of continued research toward tagging of Ba daughters in all significant sites in solid xenon., Comment: 9 pages, 8 figures

The rapid development of multimodal large language models (MLLMs), such as GPT-4V, has led to significant advancements. However, these models still face challenges in medical multimodal capabilities due to limitations in the quantity and quality of medical vision-text data, stemming from data privacy concerns and high annotation costs. While pioneering approaches utilize PubMed's large-scale, de-identified medical image-text pairs to address these limitations, they still fall short due to inherent data noise. To tackle this, we refined medical image-text pairs from PubMed and employed MLLMs (GPT-4V) in an 'unblinded' capacity to denoise and reformat the data, resulting in the creation of the PubMedVision dataset with 1.3 million medical VQA samples. Our validation demonstrates that: (1) PubMedVision can significantly enhance the medical multimodal capabilities of current MLLMs, showing significant improvement in benchmarks including the MMMU Health & Medicine track; (2) manual checks by medical experts and empirical results validate the superior data quality of our dataset compared to other data construction methods. Using PubMedVision, we train a 34B medical MLLM HuatuoGPT-Vision, which shows superior performance in medical multimodal scenarios among open-source MLLMs.

Time-reversal symmetry is a prevalent feature of microscopic physics, including operational quantum theory and classical general relativity. Previous works have studied indefinite causal structure using the language of operational quantum theory, however, these rely on time-asymmetric conditions to constrain both operations and the process matrix. Here, we use time-symmetric, operational probabilistic theory to develop a time-symmetric process matrix formalism for indefinite causal structure. This framework allows for more processes than previously considered and a larger set of causal inequalities. We demonstrate that this larger set of causal inequalities offers new opportunities for device-independent certification of causal non-separability by violating new inequalities. Additionally, we determined that the larger class of time-symmetric processes found here is equivalent to those with Indefinite Causal Order and Time Direction (ICOTD) considered by Chiribella and Liu, thereby providing a description of these processes in terms of process matrices., Comment: 38 pages

The increasing sophistication of cyber threats necessitates innovative approaches to cybersecurity. In this paper, we explore the potential of psychological profiling techniques, particularly focusing on the utilization of Large Language Models (LLMs) and psycholinguistic features. We investigate the intersection of psychology and cybersecurity, discussing how LLMs can be employed to analyze textual data for identifying psychological traits of threat actors. We explore the incorporation of psycholinguistic features, such as linguistic patterns and emotional cues, into cybersecurity frameworks. Our research underscores the importance of integrating psychological perspectives into cybersecurity practices to bolster defense mechanisms against evolving threats.

We describe a new mechanism, whose ingredients are realised in string compactifications, for the formation of cosmic (super)string networks. Oscillating string loops grow when their tension $\mu$ decreases with time. If $2H + \dot{\mu}/\mu < 0$, where $H$ is the Hubble parameter, loops grow faster than the scale factor and an initial population of isolated small loops (for example, produced by nucleation) can grow, percolate and form a network. This condition is satisfied for fundamental strings in the background of a kinating volume modulus rolling towards the asymptotic large volume region of moduli space. Such long kination epochs are motivated in string cosmology by both the electroweak hierarchy problem and the need to solve the overshoot problem. The tension of such a network today is set by the final vacuum; for phenomenologically appealing Large Volume Scenario (LVS) vacua, this would lead to a fundamental string network with $G \mu \sim 10^{-10}$., Comment: 6 pages

If millicharged particles (MCPs) exist they can be created in the atmosphere when high energy cosmic rays collide with nuclei and could subsequently be detected at neutrino experiments. We extend previous work, which considered MCPs from decays of light mesons and proton bremsstrahlung, by including production from $\Upsilon$ meson decays and the Drell-Yan process. MCPs with masses below a GeV primarily arise from proton bremsstrahlung, while heavier MCPs predominantly originate from heavy meson decays and Drell-Yan. We analyse the resulting single scatter and multiple scatter signals at SuperK and JUNO. Searches for low energy coincident signals at JUNO will be sensitive to MCPs with milli-charges up to an order of magnitude beyond current constraints for MCP masses between 2 GeV and 10 GeV., Comment: 13 pages, 8 figures

Electron-neutrino charged-current interactions with xenon nuclei were modeled in the nEXO neutrinoless double-beta decay detector (~5-tonne, 90% ${}^{136}$Xe, 10% ${}^{134}$Xe) to evaluate its sensitivity to supernova neutrinos. Predictions for event rates and detectable signatures were modeled using the MARLEY event generator. We find good agreement between MARLEY's predictions and existing theoretical calculations of the inclusive cross sections at supernova neutrino energies. The interactions modeled by MARLEY were simulated within the nEXO simulation framework and were run through an example reconstruction algorithm to determine the detector's efficiency for reconstructing these events. The simulated data, incorporating the detector response, were used to study the ability of nEXO to reconstruct the incident electron-neutrino spectrum and these results were extended to a larger xenon detector of the same isotope enrichment. We estimate that nEXO will be able to observe electron-neutrino interactions with xenon from supernovae as far as 5 to 8 kpc from earth, while the ability to reconstruct incident electron-neutrino spectrum parameters from observed interactions in nEXO is limited to closer supernovae., Comment: 17 pages, 16 figures

We show that if dark matter consists of QCD axions in the post-inflationary scenario more than ten percent of it efficiently collapses into Bose stars at matter-radiation equality. Such a result is mostly independent of the present uncertainties on the axion mass. This large population of solitons, with asteroid masses and Earth-Moon distance sizes, might plausibly survive until today, with potentially interesting implications for phenomenology and experimental searches., Comment: 23 pages + appendices

Mental illness remains one of the most critical public health issues. Despite its importance, many mental health professionals highlight a disconnect between their training and actual real-world patient practice. To help bridge this gap, we propose PATIENT-{\Psi}, a novel patient simulation framework for cognitive behavior therapy (CBT) training. To build PATIENT-{\Psi}, we construct diverse patient cognitive models based on CBT principles and use large language models (LLMs) programmed with these cognitive models to act as a simulated therapy patient. We propose an interactive training scheme, PATIENT-{\Psi}-TRAINER, for mental health trainees to practice a key skill in CBT -- formulating the cognitive model of the patient -- through role-playing a therapy session with PATIENT-{\Psi}. To evaluate PATIENT-{\Psi}, we conducted a comprehensive user study of 13 mental health trainees and 20 experts. The results demonstrate that practice using PATIENT-{\Psi}-TRAINER enhances the perceived skill acquisition and confidence of the trainees beyond existing forms of training such as textbooks, videos, and role-play with non-patients. Based on the experts' perceptions, PATIENT-{\Psi} is perceived to be closer to real patient interactions than GPT-4, and PATIENT-{\Psi}-TRAINER holds strong promise to improve trainee competencies. Our code and data are released at \url{https://github.com/ruiyiw/patient-psi}., Comment: EMNLP 2024 Main, 9 pages, 5 figures

This work examines the reproducibility and benchmarking of state-of-the-art real-time object detection models. As object detection models are often used in real-world contexts, such as robotics, where inference time is paramount, simply measuring models' accuracy is not enough to compare them. We thus compare a large variety of object detection models' accuracy and inference speed on multiple graphics cards. In addition to this large benchmarking attempt, we also reproduce the following models from scratch using PyTorch on the MS COCO 2017 dataset: DETR, RTMDet, ViTDet and YOLOv7. More importantly, we propose a unified training and evaluation pipeline, based on MMDetection's features, to better compare models. Our implementation of DETR and ViTDet could not achieve accuracy or speed performances comparable to what is declared in the original papers. On the other hand, reproduced RTMDet and YOLOv7 could match such performances. Studied papers are also found to be generally lacking for reproducibility purposes. As for MMDetection pretrained models, speed performances are severely reduced with limited computing resources (larger, more accurate models even more so). Moreover, results exhibit a strong trade-off between accuracy and speed, prevailed by anchor-free models - notably RTMDet or YOLOx models. The code used is this paper and all the experiments is available in the repository at https://github.com/Don767/segdet_mlcr2024., Comment: Authors are presented in alphabetical order, each having equal contribution to the work. Copyright may be transferred without notice, after which this version may no longer be accessible

Extreme temperature outages can lead to not just economic losses but also various non-energy impacts (NEI) due to significant degradation of indoor operating conditions caused by service disruptions. However, existing resilience assessment approaches lack specificity for extreme temperature conditions. They often overlook temperature-related mortality and neglect the customer characteristics and grid response in the calculation, despite the significant influence of these factors on NEI-related economic losses. This paper aims to address these gaps by introducing a comprehensive framework to estimate the impact of resilience enhancement not only on the direct economic losses incurred by customers but also on potential NEI, including mortality and the value of statistical life during extreme temperature-related outages. The proposed resilience valuation integrates customer characteristics and grid response variables based on a scalable grid simulation environment. This study adopts a holistic approach to quantify customer-oriented economic impacts, utilizing probabilistic loss scenarios that incorporate health-related factors and damage/loss models as a function of exposure for valuation. The proposed methodology is demonstrated through comparative resilient outage planning, using grid response models emulating a Texas weather zone during the 2021 winter storm Uri. The case study results show that enhanced outage planning with hardened infrastructure can improve the system resilience and thereby reduce the relative risk of mortality by 16% and save the total costs related to non-energy impacts by 74%. These findings underscore the efficacy of the framework by assessing the financial implications of each case, providing valuable insights for decision-makers and stakeholders involved in extreme-weather related resilience planning for risk management and mitigation strategies.

Despite the advancements and impressive performance of Multimodal Large Language Models (MLLMs) on benchmarks, their effectiveness in real-world, long-context, and multi-image tasks is unclear due to the benchmarks' limited scope. Existing benchmarks often focus on single-image and short-text samples, and when assessing multi-image tasks, they either limit the image count or focus on specific task (e.g time-series captioning), potentially obscuring the performance challenges of MLLMs. To address these limitations, we introduce MileBench, a pioneering benchmark designed to test the MultImodal Long-contExt capabilities of MLLMs. This benchmark comprises not only multimodal long contexts, but also multiple tasks requiring both comprehension and generation. We establish two distinct evaluation sets, diagnostic and realistic, to systematically assess MLLMs' long-context adaptation capacity and their ability to complete tasks in long-context scenarios. Our experimental results, obtained from testing 22 models, revealed that while the closed-source GPT-4o outperforms others, most open-source MLLMs struggle in long-context situations. Interestingly, the performance gap tends to widen with an increase in the number of images. We strongly encourage an intensification of research efforts towards enhancing MLLMs' long-context capabilities, especially in scenarios involving multiple images., Comment: 31 pages, 13 figures, 14 tables; We add results of GPT-4o in this version

Whilst holding great promise for low noise, ease of operation and networking, useful photonic quantum computing has been precluded by the need for beyond-state-of-the-art components, manufactured by the millions. Here we introduce a manufacturable platform for quantum computing with photons. We benchmark a set of monolithically-integrated silicon photonics-based modules to generate, manipulate, network, and detect photonic qubits, demonstrating dual-rail photonic qubits with $99.98\% \pm 0.01\%$ state preparation and measurement fidelity, Hong-Ou-Mandel quantum interference between independent photon sources with $99.50\%\pm0.25\%$ visibility, two-qubit fusion with $99.22\%\pm0.12\%$ fidelity, and a chip-to-chip qubit interconnect with $99.72\%\pm0.04\%$ fidelity, not accounting for loss. In addition, we preview a selection of next generation technologies, demonstrating low-loss silicon nitride waveguides and components, fabrication-tolerant photon sources, high-efficiency photon-number-resolving detectors, low-loss chip-to-fiber coupling, and barium titanate electro-optic phase shifters., Comment: 8 pages, 5 figures

Current human pose estimation systems focus on retrieving an accurate 3D global estimate of a single person. Therefore, this paper presents one of the first 3D multi-person human pose estimation systems that is able to work in real-time and is also able to handle basic forms of occlusion. First, we adjust an off-the-shelf 2D detector and an unsupervised 2D-3D lifting model for use with a 360$^\circ$ panoramic camera and mmWave radar sensors. We then introduce several contributions, including camera and radar calibrations, and the improved matching of people within the image and radar space. The system addresses both the depth and scale ambiguity problems by employing a lightweight 2D-3D pose lifting algorithm that is able to work in real-time while exhibiting accurate performance in both indoor and outdoor environments which offers both an affordable and scalable solution. Notably, our system's time complexity remains nearly constant irrespective of the number of detected individuals, achieving a frame rate of approximately 7-8 fps on a laptop with a commercial-grade GPU.

Motivated by systems where a high temperature non-Fermi liquid gives way to low temperature $\mathbb{Z}_3$ Potts nematic order, we studied a three-orbital Sachdev-Ye-Kitaev (SYK) model in the large-$N$ limit. In the single-site limit, this model exhibits a spontaneous orbital-selective transition which preserves average particle-hole symmetry, with two orbitals becoming insulators while the third orbital remains a non-Fermi liquid down to zero temperature. We extend this study to lattice models of three-orbital SYK dots, exploring uniform symmetry broken states on the triangular and cubic lattices. At high temperature, these lattice models exhibit an isotropic non-Fermi liquid metal phase. On the three-dimensional (3D) cubic lattice, the low temperature uniform $\mathbb{Z}_3$ nematic state corresponds to an orbital selective layered state which preserves particle-hole symmetry at small hopping and spontaneously breaks the particle-hole symmetry at large hopping. Over a wide range of temperature, the transport in this layered state shows metallic in-plane resistivity but insulating out-of-plane resistivity. On the 2D triangular lattice, the low temperature state with uniform orbital order is also a correlated $\mathbb{Z}_3$ nematic with orbital-selective transport but it remains metallic in both principal directions. We discuss a Landau theory with $\mathbb{Z}_3$ clock terms which captures salient features of the phase diagram and nematic order in all these models. We also present results on the approximate wavevector dependent orbital susceptibility of the isotropic non-Fermi liquid states., Comment: 23 pages, 23 figures

Large vision-language models (LVLMs) have shown premise in a broad range of vision-language tasks with their strong reasoning and generalization capabilities. However, they require considerable computational resources for training and deployment. This study aims to bridge the performance gap between traditional-scale LVLMs and resource-friendly lite versions by adopting high-quality training data. To this end, we propose a comprehensive pipeline for generating a synthetic dataset. The key idea is to leverage strong proprietary models to generate (i) fine-grained image annotations for vision-language alignment and (ii) complex reasoning visual question-answering pairs for visual instruction fine-tuning, yielding 1.3M samples in total. We train a series of lite VLMs on the synthetic dataset and experimental results demonstrate the effectiveness of the proposed scheme, where they achieve competitive performance on 17 benchmarks among 4B LVLMs, and even perform on par with 7B/13B-scale models on various benchmarks. This work highlights the feasibility of adopting high-quality data in crafting more efficient LVLMs. We name our dataset \textit{ALLaVA}, and open-source it to research community for developing better resource-efficient LVLMs for wider usage., Comment: 22 pages

Adopting human and large language models (LLM) as judges (a.k.a human- and LLM-as-a-judge) for evaluating the performance of LLMs has recently gained attention. Nonetheless, this approach concurrently introduces potential biases from human and LLMs, questioning the reliability of the evaluation results. In this paper, we propose a novel framework that is free from referencing groundtruth annotations for investigating Misinformation Oversight Bias, Gender Bias, Authority Bias and Beauty Bias on LLM and human judges. We curate a dataset referring to the revised Bloom's Taxonomy and conduct thousands of evaluations. Results show that human and LLM judges are vulnerable to perturbations to various degrees, and that even the cutting-edge judges possess considerable biases. We further exploit these biases to conduct attacks on LLM judges. We hope that our work can notify the community of the bias and vulnerability of human- and LLM-as-a-judge, as well as the urgency of developing robust evaluation systems., Comment: EMNLP2024

A major obstacle to understanding high-Tc cuprates is that superconductivity precludes observing normal-state properties at low temperatures. One prime example is the normal-state spin susceptibility: although its decrease upon cooling far above Tc typifies pseudogap behavior, its behavior at low temperatures is generally unknown. Here, our measurements in high magnetic fields expose the spin susceptibility of YBa2Cu3Oy down to low temperatures. Even though superconductivity is suppressed by the field, we uncover two thermally-activated contributions alongside a residual susceptibility at T=0 due to gapless excitations. We relate these two distinct gaps to short-range charge-density waves and to the formation of spin singlets similar to those found in certain quantum spin systems. These phenomena thus collectively contribute to the pseudogap in the spin susceptibility at low temperature, supplementing short-lived antiferromagnetism known to initiate pseudogap behavior at high temperatures. We therefore propose that the pseudogap should be regarded as a composite property.