Search

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.

Despite having different astronomical characteristics, the studies of mira variables and ultra-cool dwarfs frequently show similar red colors, which could cause leading to photometric misclassification. This study uses photometric data from the WISE, 2MASS, and Pan-STARRS surveys to construct color-based selection criteria for red dwarfs, brown dwarfs, and Mira variables. On analyzing the color indices, we developed empirical rules that separate these objects with an overall classification accuracy of approximately 91%-92%. While the differentiation between red dwarfs and both Mira variables and brown dwarfs is effective, challenges remain in distinguishing Mira variables from brown dwarfs due to overlapping color indices. The robustness of our classification technique was validated by a bootstrap analysis, highlighting the significance of color indices in large photometric surveys for stellar classification., Comment: Accepted to Peremennye Zvezdy (Variable Stars), comments are welcomed

Sixty years after the discovery of brown dwarfs, the search for these objects continues, particularly in the vicinity of the Sun. Objects near the Sun are characterized by large proper motions, making them seen as fast-moving objects. While the Gaia DR3 catalogue is a comprehensive source of proper motions, it lacks the depth needed for discovering fainter objects. Modern multi-epoch surveys, with their greater depth, offer a new opportunity for systematic search for ultra-cool dwarfs. The study aims to systematically search for high proper motion objects using the newly released catalogue of epochal WISE data in order to identify new brown dwarf candidates in the solar neighborhood, estimate their spectral types, distances and spatial velocities. We used recently released unTimely catalogue of epochal detections in unWISE coadds to search for objects with high proper motions using simple motion detection algorithm. This method was used to identify objects with proper motions exceeding approximately 0.6 arcseconds per year. The identified objects were then cross-referenced with data from other large-scale sky surveys to further analyze their characteristics. The search yielded 3245 moving objects with significant proper motions, 32 of which had not been previously published. Among these, at least 15 were identified as reliable new brown dwarf candidates, with estimated distances closer than 50 parsecs and spectral types later than T0., Comment: Table 1 is available online at https://zenodo.org/records/13362690. Submitted to A&A

We establish a version of the Landen's transformation for Weierstrass functions and invariants that is applicable to general lattices in complex plane. Using it we present an effective method for computing Weierstrass functions, their periods, and elliptic integral in Weierstrass form given Weierstrass invariants $g_2$ and $g_3$ of an elliptic curve. Similarly to the classical Landen's method our algorithm has quadratic rate of convergence.

The observed energy spectra of accelerated particles at interplanetary shocks often do not match the diffusive shock acceleration (DSA) theory predictions. In some cases, the particle flux forms a plateau over a wide range of energies, extending upstream of the shock for up to seven flux's e-folds before submerging into the background spectrum. Remarkably, at and behind the shock that we have studied in detail, the flux falls off in energy as $\epsilon^{-1}$, consistent with the DSA prediction for a strong shock. The upstream plateau suggests a different particle transport mechanism than those traditionally employed in DSA models. We show that a standard (linear) DSA solution based on a widely accepted diffusive particle transport with an underlying resonant wave-particle interaction is inconsistent with the plateau in the particle flux. To resolve this contradiction, we modify the DSA theory in two ways. First, we include a dependence of the particle diffusivity $\kappa$ on the particle flux $F$ (nonlinear particle transport). Second, we invoke short-scale magnetic perturbations that are self-consistently generated by, but not resonant with, accelerated particles. They lead to the particle diffusivity increasing with the particle energy as $\propto\epsilon^{3/2}$ that simultaneously decreases with the particle flux as $1/F$. The combination of these two trends results in the flat spectrum upstream., Comment: 13 pages 3 figures 1 table

Recent measurements of primary and secondary CR spectra, their arrival directions, and our improved knowledge of the magnetic field geometry around the heliosphere allow us to set a bound on the distance beyond which a puzzling 10-TeV ``bump'' and certain related spectral features \emph{cannot} originate. The sharpness of the spectral breaks associated with the bump, the abrupt change of the CR intensity across the local magnetic equator ($90^{\circ}$ pitch angle), and the similarity between the primary and secondary CR spectral patterns point to a local reacceleration of the bump particles out of the background CRs. We argue that, owing to a steep preexisting CR spectrum, a nearby shock may generate such a bump by boosting particle rigidity by a mere factor of $\sim$1.5 in the range below 50 TV. Reaccelerated particles below $\sim$0.5 TV are convected with the interstellar medium flow and do not reach the Sun. The particles above this rigidity then form the bump. This single universal process is responsible for the observed spectral features of all CR nuclei, primary and secondary, in the rigidity range below 100 TV. We propose that one viable candidate is the system of shocks associated with $\epsilon$ Eridani star at 3.2 pc of the Sun, which is well aligned with the direction of the local magnetic field. Other shocks, such as old supernova shells, may produce a similar effect. We provide a simple formula that reproduces the spectra of all CR species with only three parameters uniquely derived from the CR proton data. We show how our formalism predicts helium, boron, carbon, oxygen, and iron spectra. Our model thus unifies all the CR spectral features observed below 50 TV., Comment: 12 pages Cospar 2022 conf. To appear in Adv. Space Res

Gaia Data Release 3 (DR3) provides extensive information on the astrophysical properties of stars, such as effective temperature, surface gravity, metallicity, and luminosity, for over 470 million objects. However, as Gaia's stellar parameters in GSP-Phot module are derived through model-dependent methods and indirect measurements, it can lead to additional systematic errors in the derived parameters. In this study, we compare GSP-Phot effective temperature estimates with two high-resolution and high signal-to-noise spectroscopic catalogues: APOGEE DR17 and GALAH DR3, aiming to assess the reliability of Gaia's temperatures. We introduce an approach to distinguish good-quality Gaia DR3 effective temperatures using machine-learning methods such as XGBoost, CatBoost and LightGBM. The models create quality flags, which can help one to distinguish good-quality GSP-Phot effective temperatures. We test our models on three independent datasets, including PASTEL, a compilation of spectroscopically derived stellar parameters from different high-resolution studies. The results of the test suggest that with these models it is possible to filter effective temperatures as accurate as 250 K with ~ 90 per cent precision even in complex regions, such as the Galactic plane. Consequently, the models developed herein offer a valuable quality assessment tool for GSP-Phot effective temperatures in Gaia DR3. Consequently, the developed models offer a valuable quality assessment tool for GSP-Phot effective temperatures in Gaia DR3. The dataset with flags for all GSP-Phot effective temperature estimates, is publicly available, as are the models themselves., Comment: 13 pages, 10 figures

Understanding the ability of particles to maneuver through disordered environments is a central problem in innumerable settings, from active matter and biology to electronics. Macroscopic particles ultimately exhibit diffusive motion when their energy exceeds the characteristic potential barrier of the random landscape. In stark contrast, wave-particle duality causes electrons in sufficiently disordered media to come to rest even when the potential is weak -- a remarkable phenomenon known as Anderson localization. Here, we present a hydrodynamic active system with wave-particle features, a millimetric droplet self-guided by its own wave field over a submerged random topography, whose dynamics exhibits localized statistics analogous to those of electronic systems. Consideration of an ensemble of particle trajectories reveals a suppression of diffusion when the guiding wave field extends over the disordered topography. We rationalize mechanistically the emergent statistics by virtue of the wave-mediated resonant coupling between the droplet and topography, which produces an attractive wave potential about the localization region. This hydrodynamic analog, which demonstrates how a classical particle may localize like a wave, suggests new directions for future research in various areas, including active matter, wave localization, many-body localization, and topological matter.

We have completed our observational program to search for wide binary systems with non-coeval components in the southern sky and report our results here. The final set of four systems was spectroscopically investigated in this paper. No binary systems with components of different ages were found among them. Taking into account our previous studies, we estimate the fraction of such binaries (i.e., binaries formed, presumably, by capture) not higher than 0.06~\%. The study will be continued on the northern sky., Comment: 9 pages, 3 figures, 2 tables. Published in Research in Astronomy and Astrophysics

A shock wave propagating perpendicularly to an ambient magnetic field accelerates particles considerably faster than in the parallel propagation regime. However, the perpendicular acceleration stops after the shock overruns a circular particle orbit. At the same time, it may continue in flows resulting from supersonically colliding plasmas bound by a pair of perpendicular shocks. Although the double-shock acceleration mechanism, which we consider in detail, is not advantageous for thermal particles, pre-energized particles may avoid the premature end of acceleration. We argue that if their gyroradius exceeds the dominant turbulence scale between the shocks, these particles might traverse the intershock space repeatedly before being carried away by the shocked plasma. Moreover, entering the space between the shocks of similar velocities $u_{1}\approx u_{2}\approx c$, such particles start bouncing between the shocks at a fixed angle $\approx 35.3^{\circ}$ to the shock surface. Their drift along the shock fronts is slow, $V_{d}\sim\left|u_{2}-u_{1}\right|\ll c$, so that it will take $N\sim Lc/\left|u_{2}-u_{1}\right|d\gg1$ bounces before they escape the accelerator (here, $L$ is the size of the shocks and $d$ is the gap between them). Since these particles more than ten-fold their energy per cycle (two consecutive bounces), we invoke other possible losses that can limit the acceleration. They include drifts due to rippled shocks, the nonparallel mutual orientation of the upstream magnetic fields, and radiative losses., Comment: 21 pages, 13 figure, improved in V.2

State-of-the-art face recognition systems require vast amounts of labeled training data. Given the priority of privacy in face recognition applications, the data is limited to celebrity web crawls, which have issues such as limited numbers of identities. On the other hand, self-supervised revolution in the industry motivates research on the adaptation of related techniques to facial recognition. One of the most popular practical tricks is to augment the dataset by the samples drawn from generative models while preserving the identity. We show that a simple approach based on fine-tuning pSp encoder for StyleGAN allows us to improve upon the state-of-the-art facial recognition and performs better compared to training on synthetic face identities. We also collect large-scale unlabeled datasets with controllable ethnic constitution -- AfricanFaceSet-5M (5 million images of different people) and AsianFaceSet-3M (3 million images of different people) -- and we show that pretraining on each of them improves recognition of the respective ethnicities (as well as others), while combining all unlabeled datasets results in the biggest performance increase. Our self-supervised strategy is the most useful with limited amounts of labeled training data, which can be beneficial for more tailored face recognition tasks and when facing privacy concerns. Evaluation is based on a standard RFW dataset and a new large-scale RB-WebFace benchmark. The code and data are made publicly available at https://github.com/seva100/stylegan-for-facerec., Comment: 16 pages, 9 figures, 11 tables; accepted to ICCV 2023

Pre-trained language models (PLMs) are fundamental for natural language processing applications. Most existing PLMs are not tailored to the noisy user-generated text on social media, and the pre-training does not factor in the valuable social engagement logs available in a social network. We present TwHIN-BERT, a multilingual language model productionized at Twitter, trained on in-domain data from the popular social network. TwHIN-BERT differs from prior pre-trained language models as it is trained with not only text-based self-supervision, but also with a social objective based on the rich social engagements within a Twitter heterogeneous information network (TwHIN). Our model is trained on 7 billion tweets covering over 100 distinct languages, providing a valuable representation to model short, noisy, user-generated text. We evaluate our model on various multilingual social recommendation and semantic understanding tasks and demonstrate significant metric improvement over established pre-trained language models. We open-source TwHIN-BERT and our curated hashtag prediction and social engagement benchmark datasets to the research community.

In this paper, we investigate the orbital and stellar parameters of low- and intermediate-mass close binary systems. We use models, presented in the catalogue of (Han et al. 2000) and calculate parameters of accretors. We also construct distributions of systems along luminosity, semi-major axis and angular momentum, and make some conclusions on their evolution with time. We made a comparison of the results with observational data and it shows a good agreement. The set of theoretical models published in (Han et al. 2000) quite adequately describes the observational data and, consequently, can be used to determine the evolutionary path of specific close binary systems, their initial parameters values and final stages., Comment: 13 pages, 7 figures

We have estimated masses of components of visual binaries from their spectral classification. We have selected pairs, where the less massive component looks more evolved. Spectral observations of some of such pairs were made, and at least one pair, HD~156331, was confirmed to have components of different age. Since mass exchange is excluded in wide binaries, it means that HD~156331 can be formed by the capture., Comment: 6 pages, 2 figures. accepted to Open Astronomy

The production of the artificial 51Cr neutrino source with activity > 3 MCi for the experiment BEST is presented. This procedure consisted of making a 50Cr target and irradiating it with thermal neutrons in a nuclear reactor SM-3. The production of a target in the form of disks with a thickness of 4 mm and a diameter of 84 and 88 mm included enrichment (to 96.5% in 50Cr) of natural chromium in the form of oxyfluoride by gas centrifugation, electrolytic reduction and refining of metallic chromium, as well as the formation of chromium disks by spark plasma sintering. Simulations of various source geometries, neutron flux and nuclear transmutation were carried out to validate the design of the source, the irradiation device and the transport container, the required chemical purity of the target and the irradiation schedule in the reactor. The calculated activity of the source after 75 effective days of irradiation was 3.55 MCi. The activity of the source was measured by the calorimetric method and amounted to 3.41 MCi at the time of its delivery to the Baksan Neutrino Observatory. This is the most intense chemically pure neutrino source ever produced., Comment: 18 pages, 8 figures

3350 objects from the Sixth catalog of orbits of visual binary stars (ORB6) are investigated to validate Gaia EDR3 parallaxes and provide mass estimates for the systems. We show that 2/3 of binaries with 0.2 - 0.5 arcsec separation are left without a parallax solution in EDR3. A special attention is paid to 521 pairs with parallax known separately for both components. We find 16 entries that are deemed to be chance alignments of unrelated stars. At once we show examples of high-confidence binary systems with significant differences in the reported parallaxes of their components. Next we conclude that the reported Gaia EDR3 parallax errors are underestimated, at least by a factor of 3 for sources with large RUWE. Parallaxes are needed to estimate stellar masses. Since nearly 30\% of ORB6 entries lack 5 or 6-parameter solution in EDR3, we attempt to enrich the astrometric data. Distant companions of ORB6 entries are revealed in EDR3 by analysis of stellar proper motions and Hipparcos parallaxes. Notably, in 28 cases intrinsic EDR3 parallaxes of the binary components appear to be less reliable than the parallax of the outer companions. Gaia DR2, TGAS and Hipparcos parallaxes are used when EDR3 data is unavailable. Synthetic mass-luminosity relation in the G band for main sequence stars is obtained to provide mass estimates along with dynamical masses calculated via Kepler's Third Law., Comment: Accepted for publication in MNRAS 28.09.2022, 15 pages, 11 figures

We have analyzed the retrieval of the relation between mass and absolute Gaia G magnitude as obtained from theoretical models by \cite{2012MNRAS.427..127B} for the stars of different luminosity classes. For subgiant and main-sequence stars, we provide approximate analytical direct and inverse relations based on the most probable value of G magnitude in the course of evolution within the given stage. A comparison with (albeit few) observational data confirms that our results can be used for the estimation of the stellar mass from Gaia photometry in the range of 1 to 10 solar mass. We argue that similar relations for other luminosity classes are not informative., Comment: 5 pages, 3 figures

Social networks, such as Twitter, form a heterogeneous information network (HIN) where nodes represent domain entities (e.g., user, content, advertiser, etc.) and edges represent one of many entity interactions (e.g, a user re-sharing content or "following" another). Interactions from multiple relation types can encode valuable information about social network entities not fully captured by a single relation; for instance, a user's preference for accounts to follow may depend on both user-content engagement interactions and the other users they follow. In this work, we investigate knowledge-graph embeddings for entities in the Twitter HIN (TwHIN); we show that these pretrained representations yield significant offline and online improvement for a diverse range of downstream recommendation and classification tasks: personalized ads rankings, account follow-recommendation, offensive content detection, and search ranking. We discuss design choices and practical challenges of deploying industry-scale HIN embeddings, including compressing them to reduce end-to-end model latency and handling parameter drift across versions.