arXiv subjects

Eran Ofek

Publications and source records attributed to Eran Ofek.

At least 19 recordsLinked to original sources

arXiv2026

Constraints on Late-Time Flaring from Luminous Fast Blue Optical Transients using the Transiting Exoplanet Survey Satellite and the Zwicky Transient Facility

The Luminous Fast Blue Optical Transient (LFBOT) AT2022tsd exhibited minutes-timescale optical flares in the tens of days following the initial transient event, likely due to a central engine -- either an accreting black hole or a magnetar. In this paper, we use data from the Transiting Exoplanet Survey Satellite (TESS) and the Zwicky Transient Facility (ZTF) to constrain the occurrence of similar flares in the 12 (of 14) known LFBOTs that had observational coverage with TESS from tens of days to thousands of days after the transient's initial emission. We find seven flare-like signals at the locations of four unique LFBOTs; all seven can likely be attributed to a solar system object (SSO) moving through the TESS aperture. Assuming all seven flares arise from SSOs, for the LFBOT AT2024qfm we rule out flaring with a similar timescale (40--65 d) and luminosity ($\nu L_\nu\sim10^{43}$ erg s$^{-1}$) as in AT2022tsd, while for AT2022tsd itself we rule out flares between 380--430 d after the initial transient that were as luminous as the earlier flares. This observation suggests that the engine power in AT2022tsd declined or shut off on a timescale of hundreds of days. We also find that there is no late-time activity detectable in TESS thousands of days after the prototype LFBOT, AT2018cow. We discuss our constraints on the duty cycle of such flaring and then present estimates for the number of minutes-duration flares detectable with ongoing and upcoming high-cadence ($\ll1$ d) wide-field surveys.

astro-ph.HE↗

arXiv2026

Thinking to Recall: How Reasoning Unlocks Parametric Knowledge in LLMs

While reasoning in LLMs plays a natural role in math, code generation, and multi-hop factual questions, its effect on simple, single-hop factual questions remains unclear. Such questions do not require step-by-step logical decomposition, making the utility of reasoning highly counterintuitive. Nevertheless, we find that enabling reasoning substantially expands the capability boundary of the model's parametric knowledge recall, unlocking correct answers that are otherwise effectively unreachable. Why does reasoning aid parametric knowledge recall when there are no complex reasoning steps to be done? To answer this, we design a series of hypothesis-driven controlled experiments, and identify two key driving mechanisms: (1) a computational buffer effect, where the model uses the generated reasoning tokens to perform latent computation independent of their semantic content; and (2) factual priming, where generating topically related facts acts as a semantic bridge that facilitates correct answer retrieval. Importantly, this latter generative self-retrieval mechanism carries inherent risks: we demonstrate that hallucinating intermediate facts during reasoning increases the likelihood of hallucinations in the final answer. Finally, we show that our insights can be harnessed to directly improve model accuracy by prioritizing reasoning trajectories that contain hallucination-free factual statements.

cs.CL↗

arXiv2026

Empty Shelves or Lost Keys? Recall Is the Bottleneck for Parametric Factuality

Standard factuality evaluations of LLMs treat all errors alike, obscuring whether failures arise from missing knowledge (empty shelves) or from limited access to encoded facts (lost keys). We propose a behavioral framework that profiles factual knowledge at the level of facts rather than questions, characterizing each fact by whether it is encoded, and then by how accessible it is: cannot be recalled, can be directly recalled, or can only be recalled with inference-time computation (thinking). To support such profiling, we introduce WikiProfile, a new benchmark constructed via an automated pipeline with a prompted LLM grounded in web search. Across 4 million responses from 13 LLMs, we find that encoding is nearly saturated in frontier models on our benchmark, with GPT-5 and Gemini-3 encoding 95--98% of facts. However, recall remains a major bottleneck: many errors previously attributed to missing knowledge instead stem from failures to access it. These failures are systematic and disproportionately affect long-tail facts and reverse questions. Finally, we show that thinking improves recall and can recover a substantial fraction of failures, indicating that future gains may rely less on scaling and more on methods that improve how models utilize what they already encode.

cs.CL↗

arXiv2025

The FACTS Leaderboard: A Comprehensive Benchmark for Large Language Model Factuality

We introduce The FACTS Leaderboard, an online leaderboard suite and associated set of benchmarks that comprehensively evaluates the ability of language models to generate factually accurate text across diverse scenarios. The suite provides a holistic measure of factuality by aggregating the performance of models on four distinct sub-leaderboards: (1) FACTS Multimodal, which measures the factuality of responses to image-based questions; (2) FACTS Parametric, which assesses models' world knowledge by answering closed-book factoid questions from internal parameters; (3) FACTS Search, which evaluates factuality in information-seeking scenarios, where the model must use a search API; and (4) FACTS Grounding (v2), which evaluates whether long-form responses are grounded in provided documents, featuring significantly improved judge models. Each sub-leaderboard employs automated judge models to score model responses, and the final suite score is an average of the four components, designed to provide a robust and balanced assessment of a model's overall factuality. The FACTS Leaderboard Suite will be actively maintained, containing both public and private splits to allow for external participation while guarding its integrity. It can be found at https://www.kaggle.com/benchmarks/google/facts .

cs.CL↗

arXiv2025

DRAGged into Conflicts: Detecting and Addressing Conflicting Sources in Search-Augmented LLMs

Retrieval Augmented Generation (RAG) is a commonly used approach for enhancing large language models (LLMs) with relevant and up-to-date information. However, the retrieved sources can often contain conflicting information and it remains unclear how models should address such discrepancies. In this work, we first propose a novel taxonomy of knowledge conflict types in RAG, along with the desired model behavior for each type. We then introduce CONFLICTS, a high-quality benchmark with expert annotations of conflict types in a realistic RAG setting. CONFLICTS is the first benchmark that enables tracking progress on how models address a wide range of knowledge conflicts. We conduct extensive experiments on this benchmark, showing that LLMs often struggle to appropriately resolve conflicts between sources. While prompting LLMs to explicitly reason about the potential conflict in the retrieved documents significantly improves the quality and appropriateness of their responses, substantial room for improvement in future research remains.

cs.CL↗

arXiv2025

Inside-Out: Hidden Factual Knowledge in LLMs

This work presents a framework for assessing whether large language models (LLMs) encode more factual knowledge in their parameters than what they express in their outputs. While a few studies hint at this possibility, none has clearly defined or demonstrated this phenomenon. We first propose a formal definition of knowledge, quantifying it for a given question as the fraction of correct-incorrect answer pairs where the correct one is ranked higher. This gives rise to external and internal knowledge, depending on the information used to score individual answer candidates: either the model's observable token-level probabilities or its intermediate computations. Hidden knowledge arises when internal knowledge exceeds external knowledge. We then present a case study, applying this framework to three popular open-weights LLMs in a closed-book QA setup. Our results indicate that: (1) LLMs consistently encode more factual knowledge internally than what they express externally, with an average relative gap of 40%. (2) Surprisingly, some knowledge is so deeply hidden that a model can internally know an answer perfectly, yet fail to generate it even once, despite large-scale repeated sampling of 1,000 answers. This reveals fundamental limitations in the generation capabilities of LLMs, which (3) put a practical constraint on scaling test-time compute via repeated answer sampling in closed-book QA: significant performance improvements remain inaccessible because some answers are practically never sampled, yet if they were, we would be guaranteed to rank them first.

cs.CL↗

arXiv2025

Confidence Improves Self-Consistency in LLMs

Self-consistency decoding enhances LLMs' performance on reasoning tasks by sampling diverse reasoning paths and selecting the most frequent answer. However, it is computationally expensive, as sampling many of these (lengthy) paths is required to increase the chances that the correct answer emerges as the most frequent one. To address this, we introduce Confidence-Informed Self-Consistency (CISC). CISC performs a weighted majority vote based on confidence scores obtained directly from the model. By prioritizing high-confidence paths, it can identify the correct answer with a significantly smaller sample size. When tested on nine models and four datasets, CISC outperforms self-consistency in nearly all configurations, reducing the required number of reasoning paths by over 40% on average. In addition, we introduce the notion of within-question confidence evaluation, after showing that standard evaluation methods are poor predictors of success in distinguishing correct and incorrect answers to the same question. In fact, the most calibrated confidence method proved to be the least effective for CISC. Lastly, beyond these practical implications, our results and analyses show that LLMs can effectively judge the correctness of their own outputs, contributing to the ongoing debate on this topic.

cs.CL↗

arXiv2024

A cosmic formation site of silicon and sulphur revealed by a new type of supernova explosion

The cores of stars are the cosmic furnaces where light elements are fused into heavier nuclei. The fusion of hydrogen to helium initially powers all stars. The ashes of the fusion reactions are then predicted to serve as fuel in a series of stages, eventually transforming massive stars into a structure of concentric shells. These are composed of natal hydrogen on the outside, and consecutively heavier compositions inside, predicted to be dominated by helium, carbon/oxygen, oxygen/neon/magnesium, and oxygen/silicon/sulphur. Silicon and sulphur are fused into inert iron, leading to the collapse of the core and either a supernova explosion or the direct formation of a black hole. Stripped stars, where the outer hydrogen layer has been removed and the internal He-rich layer (in Wolf-Rayet WN stars) or even the C/O layer below it (in Wolf-Rayet WC/WO stars) are exposed, provide evidence for this shell structure, and the cosmic element production mechanism it reflects. The types of supernova explosions that arise from stripped stars embedded in shells of circumstellar material (most notably Type Ibn supernovae from stars with outer He layers, and Type Icn supernovae from stars with outer C/O layers) confirm this scenario. However, direct evidence for the most interior shells, which are responsible for the production of elements heavier than oxygen, is lacking. Here, we report the discovery of the first-of-its-kind supernova arising from a star peculiarly stripped all the way to the silicon and sulphur-rich internal layer. Whereas the concentric shell structure of massive stars is not under debate, it is the first time that such a thick, massive silicon and sulphur-rich shell, expelled by the progenitor shortly before the SN explosion, has been directly revealed.

astro-ph.HE↗

arXiv2024

HighSpec: A High-Resolution Spectrograph for the MAST Telescope Array

We present the updated design of HighSpec, a high-resolution $\mathcal{R} \sim 20,000$ spectrograph designed for the Multi Aperture Spectroscopic Telescope (MAST). HighSpec offers three observing modes centered at the Ca II H&K, Mg b triplet, and H$\alpha$ lines. Each mode is supported by a highly optimized ion-etched grating, contributing to an exceptional instrument peak efficiency of $\gtrsim85\%$ for the two latter bands ($\gtrsim55\%$ for the Ca II H&K band). Optimizing throughput over wavelength coverage ($\Delta \lambda=10-17$ nm), HighSpec enables the precise measurement of spectral lines from faint targets. This approach is especially relevant for stellar object studies, specifically of WDs, which are intrinsically faint and have few spectroscopic lines. Each observing mode was tailored to target spectral features essential for WD research. Its integration with MAST, an array of 20 custom-designed telescopes that can function as a single large telescope (equivalent to a $2.7$ m telescope in collecting area) or multiplexing over the entire sky, provides unique adaptability for extensive and effective spectroscopic campaigns. Currently in its final assembly and testing stages, HighSpec's on-sky commissioning is scheduled for 2025.

astro-ph.IM↗

arXiv2023

What You See is What You Read? Improving Text-Image Alignment Evaluation

Automatically determining whether a text and a corresponding image are semantically aligned is a significant challenge for vision-language models, with applications in generative text-to-image and image-to-text tasks. In this work, we study methods for automatic text-image alignment evaluation. We first introduce SeeTRUE: a comprehensive evaluation set, spanning multiple datasets from both text-to-image and image-to-text generation tasks, with human judgements for whether a given text-image pair is semantically aligned. We then describe two automatic methods to determine alignment: the first involving a pipeline based on question generation and visual question answering models, and the second employing an end-to-end classification approach by finetuning multimodal pretrained models. Both methods surpass prior approaches in various text-image alignment tasks, with significant improvements in challenging cases that involve complex composition or unnatural images. Finally, we demonstrate how our approaches can localize specific misalignments between an image and a given text, and how they can be used to automatically re-rank candidates in text-to-image generation.

cs.CL↗

arXiv2022

The prevalence and influence of circumstellar material around hydrogen-rich supernova progenitors

Narrow transient emission lines (flash-ionization features) in early supernova (SN) spectra trace the presence of circumstellar material (CSM) around the massive progenitor stars of core-collapse SNe. The lines disappear within days after the SN explosion, suggesting that this material is spatially confined, and originates from enhanced mass loss shortly (months to a few years) prior to explosion. We performed a systematic survey of H-rich (Type II) SNe discovered within less than two days from explosion during the first phase of the Zwicky Transient Facility (ZTF) survey (2018-2020), finding thirty events for which a first spectrum was obtained within $< 2$ days from explosion. The measured fraction of events showing flash ionisation features ($>36\%$ at $95\%$ confidence level) confirms that elevated mass loss in massive stars prior to SN explosion is common. We find that SNe II showing flash ionisation features are not significantly brighter, nor bluer, nor more slowly rising than those without. This implies that CSM interaction does not contribute significantly to their early continuum emission, and that the CSM is likely optically thin. We measured the persistence duration of flash ionisation emission and find that most SNe show flash features for $\approx 5 $ days. Rarer events, with persistence timescales $>10$ days, are brighter and rise longer, suggesting these may be intermediate between regular SNe II and strongly-interacting SNe IIn.

astro-ph.HE↗

arXiv2022

The scientific payload of the Ultraviolet Transient Astronomy Satellite (ULTRASAT)

The Ultraviolet Transient Astronomy Satellite (ULTRASAT) is a space-borne near UV telescope with an unprecedented large field of view (200 sq. deg.). The mission, led by the Weizmann Institute of Science and the Israel Space Agency in collaboration with DESY (Helmholtz association, Germany) and NASA (USA), is fully funded and expected to be launched to a geostationary transfer orbit in Q2/3 of 2025. With a grasp 300 times larger than GALEX, the most sensitive UV satellite to date, ULTRASAT will revolutionize our understanding of the hot transient universe, as well as of flaring galactic sources. We describe the mission payload, the optical design and the choice of materials allowing us to achieve a point spread function of ~10arcsec across the FoV, and the detector assembly. We detail the mitigation techniques implemented to suppress out-of-band flux and reduce stray light, detector properties including measured quantum efficiency of scout (prototype) detectors, and expected performance (limiting magnitude) for various objects.

astro-ph.IM↗

arXiv2021

A Search for Extragalactic Fast Blue Optical Transients in ZTF and the Rate of AT2018cow-like Transients

We present a search for extragalactic fast blue optical transients (FBOTs) during Phase I of the Zwicky Transient Facility (ZTF). We identify 38 candidates with durations above half-maximum light 1 d < t1/2 < 12 d, of which 28 have blue (g-r<-0.2 mag) colors at peak light. Of the 38 transients (28 FBOTs), 19 (13) can be spectroscopically classified as core-collapse supernovae (SNe): 11 (8) H- or He-rich (Type II/IIb/Ib) SNe, 6 (4) interacting (Type IIn/Ibn) SNe, and 2 (1) H&He-poor (Type Ic/Ic-BL) SNe. Two FBOTs (published previously) had high-S/N predominantly featureless spectra and luminous radio emission: AT2018lug and AT2020xnd. Seven (five) did not have a definitive classification: AT 2020bdh showed tentative broad H$\alpha$ in emission, and AT 2020bot showed unidentified broad features and was 10 kpc offset from the center of an early-type galaxy. Ten (six) have no spectroscopic observations or redshift measurements. We present multiwavelength (radio, millimeter, and/or X-ray) observations for five FBOTs (three Type Ibn, one Type IIn/Ibn, one Type IIb). Additionally, we search radio-survey (VLA and ASKAP) data to set limits on the presence of radio emission for 22 of the transients. All X-ray and radio observations resulted in non-detections; we rule out AT2018cow-like X-ray and radio behavior for five FBOTs and more luminous emission (such as that seen in the Camel) for four additional FBOTs. We conclude that exotic transients similar to AT2018cow, the Koala, and the Camel represent a rare subset of FBOTs, and use ZTF's SN classification experiments to measure the rate to be at most 0.1% of the local core-collapse SN rate.

astro-ph.HE↗

arXiv2020

A large fraction of hydrogen-rich supernova progenitors experience elevated mass loss shortly prior to explosion

Spectroscopic detection of narrow emission lines traces the presence of circumstellar mass distributions around massive stars exploding as core-collapse supernovae. Transient emission lines disappearing shortly after the supernova explosion suggest that the spatial extent of such material is compact, and hence imply an increased mass loss shortly prior to explosion. Here, we present a systematic survey for such transient emission lines (Flash Spectroscopy) among Type II supernovae detected in the first year of the Zwicky Transient Facility (ZTF) survey. We find that at least six out of ten events for which a spectrum was obtained within two days of estimated explosion time show evidence for such transient flash lines. Our measured flash event fraction ($>30\%$ at $95\%$ confidence level) indicates that elevated mass loss is a common process occurring in massive stars that are about to explode as supernovae.

astro-ph.HE↗

arXiv2020

PTF11rka: an interacting supernova at the crossroads of stripped-envelope and H-poor super-luminous stellar core collapses

The hydrogen-poor supernova PTF11rka (z = 0.0744), reported by the Palomar Transient Factory, was observed with various telescopes starting a few days after the estimated explosion time of 2011 Dec. 5 UT and up to 432 rest-frame days thereafter. The rising part of the light curve was monitored only in the R_PTF filter band, and maximum in this band was reached ~30 rest-frame days after the estimated explosion time. The light curve and spectra of PTF11rka are consistent with the core-collapse explosion of a ~10 Msun carbon-oxygen core evolved from a progenitor of main-sequence mass 25--40 Msun, that liberated a kinetic energy (KE) ~ 4 x 10^{51} erg, expelled ~8 Msun of ejecta (Mej), and synthesised ~0.5 Msun of 56Nichel. The photospheric spectra of PTF11rka are characterised by narrow absorption lines that point to suppression of the highest ejecta velocities ~>15,000 km/s. This would be expected if the ejecta impacted a dense, clumpy circumstellar medium. This in turn caused them to lose a fraction of their energy (~5 x 10^50 erg), less than 2% of which was converted into radiation that sustained the light curve before maximum brightness. This is reminiscent of the superluminous SN 2007bi, the light-curve shape and spectra of which are very similar to those of PTF11rka, although the latter is a factor of 10 less luminous and evolves faster in time. PTF11rka is in fact more similar to gamma-ray burst supernovae (GRB-SNe) in luminosity, although it has a lower energy and a lower KE/Mej ratio.

astro-ph.HE↗

arXiv2019

The Zwicky Transient Facility: Surveys and Scheduler

We present a novel algorithm for scheduling the observations of time-domain imaging surveys. Our Integer Linear Programming approach optimizes an observing plan for an entire night by assigning targets to temporal blocks, enabling strict control of the number of exposures obtained per field and minimizing filter changes. A subsequent optimization step minimizes slew times between each observation. Our optimization metric self-consistently weights contributions from time-varying airmass, seeing, and sky brightness to maximize the transient discovery rate. We describe the implementation of this algorithm on the surveys of the Zwicky Transient Facility and present its on-sky performance.

astro-ph.IM↗

arXiv2019

The Dynamic Infrared Sky

Opening up the dynamic infrared sky for systematic time-domain exploration would yield many scientific advances. Multi-messenger pursuits such as localizing gravitational waves from neutron star mergers and quantifying the nucleosynthetic yields require the infrared. Another multi-messenger endeavor that needs infrared surveyors is the study of the much-awaited supernova in our own Milky Way. Understanding shocks in novae, true rates of supernovae and stellar mergers are some other examples of stellar evolution and high energy physics wherein the answers are buried in the infrared. We discuss some of the challenges in the infrared and pathfinders to overcome them. We conclude with recommendations on both infrared discovery engines and infrared follow-up machines that would enable this field to flourish in the next decade.

astro-ph.HE↗

arXiv2019

The Zwicky Transient Facility: System Overview, Performance, and First Results

The Zwicky Transient Facility (ZTF) is a new optical time-domain survey that uses the Palomar 48-inch Schmidt telescope. A custom-built wide-field camera provides a 47 deg$^2$ field of view and 8 second readout time, yielding more than an order of magnitude improvement in survey speed relative to its predecessor survey, the Palomar Transient Factory (PTF). We describe the design and implementation of the camera and observing system. The ZTF data system at the Infrared Processing and Analysis Center provides near-real-time reduction to identify moving and varying objects. We outline the analysis pipelines, data products, and associated archive. Finally, we present on-sky performance analysis and first scientific results from commissioning and the early survey. ZTF's public alert stream will serve as a useful precursor for that of the Large Synoptic Survey Telescope.

astro-ph.IM↗