Galaxy cluster gas temperatures ($T$) play a crucial role in many cosmological and astrophysical studies. However, it has been shown that $T$ measurements can vary between different X-ray telescopes. These $T$ biases can propagate to several cluster applications for which $T$ can be used. Thus, it is important to accurately cross-calibrate X-ray instruments to account for systematic biases. In this work, we present the cross-calibration between SRG/eROSITA and Chandra/ACIS, and between SRG/eROSITA and XMM-Newton/EPIC, using for the first time a large sample of galaxy cluster $T$. To do so, we use the first eROSITA All-Sky Survey data and a large X-ray flux-limited cluster catalog. We measure X-ray $T$ for 186 independent cluster regions with both SRG/eROSITA and Chandra/ACIS in a self-consistent way, for three energy bands; 0.7-7 keV (full), 0.5-4 keV (soft), and 1.5-7 keV (hard). We do the same with SRG/eROSITA and XMM-Newton/EPIC for 71 different cluster regions and all three bands. We find that SRG/eROSITA measures systematically lower $T$ than the other two instruments. For the full band, SRG/eROSITA returns 20$\%$ and 14$\%$ lower $T$ than Chandra/ACIS and XMM-Newton/EPIC respectively, when the two latter instruments measure $k_{\text{B}}T\approx 3$ keV each. The discrepancy increases to 38\% and 32\% when Chandra/ACIS and XMM-Newton/EPIC measure $k_{\text{B}}T\approx 10$ keV respectively. For low-$T$ galaxy groups, the discrepancy becomes milder. The soft band shows a marginally lower discrepancy than the full band. In the hard band, the cross-calibration of SRG/eROSITA and the other instruments show stronger differences. We could not identify any possible systematic biases that significantly alleviated the tension. Finally, we provide conversion factors between SRG/eROSITA, Chandra/ACIS, and XMM-Newton/EPIC $T$ which will be beneficial for future cluster studies.

As the closest galaxy cluster, the Virgo Cluster is an exemplary environment for the study of the large-scale filamentary structure and physical effects that are present in cluster outskirts but absent from the more easily studied inner regions. Here, we present an analysis of the SRG/eROSITA data from five all-sky surveys. eROSITA allows us to resolve the entire Virgo cluster and its outskirts on scales between 1 kpc and 3 Mpc, covering a total area on the sky of about 25$^\circ$ by 25$^\circ$. We utilized image manipulation techniques and surface brightness profiles to search for extended emission, surface brightness edges, and features in the outskirts. We employed a method of comparing mean and median profiles to measure gas clumping out to and beyond the virial radius. Surface brightness analysis of the cluster and individual sectors of the cluster reveal the full extent of previously identified cold fronts to the north and south. The emissivity bias due to gas clumping, which we quantify over three orders of magnitude in the radial range, is found to be mild, consistent with previous findings. We find uniform clumping measurements in all directions, with no enhancements along candidate filaments. We find an estimated virial gas mass of $M_{\mathrm{gas},r

Recent studies have claimed the detection of an active massive black hole (BH) in the low-metallicity blue compact dwarf galaxy SBS 0335-052 E based on near-infrared (NIR) time variability and broad H$\alpha$ wings. This interpretation remains questionable given the observed broad wings in forbidden [O III] emission. Based on spectroscopic properties derived from our KCWI/KCRM integral-field observation of super star clusters 1 and 2 (SSCs 1$\&$2), we propose instead that these BH signatures originate from a luminous blue variable (LBV) outburst in a binary system like $\eta$ Carinae. First, the [Fe II] emission-line ratio and detected O I 8446 pumped emission require high-density gas ($n_e \sim 10^6$ $\rm{cm^{-3}}$). This dense gas resides in the circumstellar medium (CSM) formed by pre-outburst stellar winds. Subsequent shock interaction between the LBV outburst ejecta and CSM efficiently produces warm dust and the corresponding NIR excess. Second, SSCs 1$\&$2 are nitrogen-enriched relative to other SSCs. This enrichment arises from ejections of CNO-cycled material by multiple LBV outbursts. Third, we detect asymmetric broad H$\alpha$ wings extending from $\sim -5\,000$ to $\sim 10\,000\ \rm{km\,s^{-1}}$. This asymmetry results from electron scattering in the expanding, optically thick CSM. The proposed CSM shock interaction naturally explains the luminosities of [Fe V] and ultra-luminous X-ray emission. Contrarily, [Fe II] and [Fe IV] emission originates primarily from gas photoionized by the cool primary LBV and hot secondary stars, respectively. Our results highlight how the shock interaction of massive stars with high-density CSM mimics active massive BH signatures in low-metallicity dwarf galaxies.

Large language models are rapidly transforming social science research by enabling the automation of labor-intensive tasks like data annotation and text analysis. However, LLM outputs vary significantly depending on the implementation choices made by researchers (e.g., model selection or prompting strategy). Such variation can introduce systematic biases and random errors, which propagate to downstream analyses and cause Type I (false positive), Type II (false negative), Type S (wrong sign), or Type M (exaggerated effect) errors. We call this phenomenon where configuration choices lead to incorrect conclusions LLM hacking. We find that intentional LLM hacking is strikingly simple. By replicating 37 data annotation tasks from 21 published social science studies, we show that, with just a handful of prompt paraphrases, virtually anything can be presented as statistically significant. Beyond intentional manipulation, our analysis of 13 million labels from 18 different LLMs across 2361 realistic hypotheses shows that there is also a high risk of accidental LLM hacking, even when following standard research practices. We find incorrect conclusions in approximately 31% of hypotheses for state-of-the-art LLMs, and in half the hypotheses for smaller language models. While higher task performance and stronger general model capabilities reduce LLM hacking risk, even highly accurate models remain susceptible. The risk of LLM hacking decreases as effect sizes increase, indicating the need for more rigorous verification of LLM-based findings near significance thresholds. We analyze 21 mitigation techniques and find that human annotations provide crucial protection against false positives. Common regression estimator correction techniques can restore valid inference but trade off Type I vs. Type II errors. We publish a list of practical recommendations to prevent LLM hacking.

Training high-performing Small Language Models (SLMs) remains costly, even with knowledge distillation and pruning from larger teacher models. Existing work often faces three key challenges: (1) information loss from hard pruning, (2) inefficient alignment of representations, and (3) underutilization of informative activations, particularly from Feed-Forward Networks (FFNs). To address these challenges, we introduce Low-Rank Clone (LRC), an efficient pre-training method that constructs SLMs aspiring to behavioral equivalence with strong teacher models. LRC trains a set of low-rank projection matrices that jointly enable soft pruning by compressing teacher weights, and activation clone by aligning student activations, including FFN signals, with those of the teacher. This unified design maximizes knowledge transfer while removing the need for explicit alignment modules. Extensive experiments with open-source teachers (e.g., Llama-3.2-3B-Instruct, Qwen2.5-3B/7B-Instruct) show that LRC matches or surpasses state-of-the-art models trained on trillions of tokens--while using only 20B tokens, achieving over 1,000x training efficiency. Our codes and model checkpoints are available at this https URL and this https URL.

Generative retrieval (GR) reformulates information retrieval (IR) by framing it as the generation of document identifiers (docids), thereby enabling an end-to-end optimization and seamless integration with generative language models (LMs). Despite notable progress under supervised training, GR still struggles to generalize to zero-shot IR scenarios, which are prevalent in real-world applications. To tackle this challenge, we propose \textsc{ZeroGR}, a zero-shot generative retrieval framework that leverages natural language instructions to extend GR across a wide range of IR tasks. Specifically, \textsc{ZeroGR} is composed of three key components: (i) an LM-based docid generator that unifies heterogeneous documents (e.g., text, tables, code) into semantically meaningful docids; (ii) an instruction-tuned query generator that generates diverse types of queries from natural language task descriptions to enhance corpus indexing; and (iii) a reverse annealing decoding strategy to balance precision and recall during docid generation. We investigate the impact of instruction fine-tuning scale and find that performance consistently improves as the number of IR tasks encountered during training increases. Empirical results on the BEIR and MAIR benchmarks demonstrate that \textsc{ZeroGR} outperforms strong dense retrieval and generative baselines in zero-shot settings, establishing a new state-of-the-art for instruction-driven GR.

We present JWST observations of the environments surrounding two high-redshift quasars -- J0252$-$0503 at $z = 7.0$ and J1007$+$2115 at $z = 7.5$ -- which enable the first constraints on quasar-galaxy clustering at $z \sim 7.3$. Galaxies in the vicinity of the quasars are selected through ground-based and JWST/NIRCam imaging and then spectroscopically confirmed with JWST/NIRSpec using the multi-shutter assembly (MSA). Over both fields, we identify 51 $z>5$ galaxies, of which eight are found within a $\Delta v_{\textrm{LOS}}=\pm1500 \rm{km} \rm{s}^{-1}$ line-of-sight velocity window from the quasars and another eight in the background. The galaxy J0252\_8713, located just $7\,\rm{pkpc}$ and $\Delta v_{\textrm{LOS}} \approx 360\,\rm{km}\,\rm{s}^{-1}$ from quasar J0252$-$0503, emerges as a compelling candidate for one of the most distant quasar-galaxy mergers. Combining the galaxy discoveries over the two fields, we measure the quasar-galaxy cross-correlation and obtain a correlation length of $r_0^{\rm{QG}}\approx7.6_{-1.6}^{+1.7}\,h^{-1}\,\rm{cMpc}$, based on a power-law model with a fixed slope of $\gamma_{\rm{QG}} = 2.0$. Under the assumption that quasars and galaxies trace the same underlying dark matter density fluctuations, we infer a minimum dark matter halo mass for $z\simeq7.3$ quasars of $\log_{10}(M_{\textrm{halo, min}}/\textrm{M}_{\odot})= 11.6\pm0.6$ in a halo model framework. Compared to measurements from EIGER at $\langle z \rangle = 6.25$ and ASPIRE at $\langle z \rangle = 6.7$ (where $\log_{10}(M_{\textrm{halo, min}}/\textrm{M}_{\odot}) \gtrsim 12.3$), our clustering results provide tentative evidence for a non-monotonic redshift evolution of quasar clustering properties. We further estimate a quasar duty cycle of $f_{\rm{duty}}\approx0.1\%$, consistent with constraints from quasar proximity zones and IGM damping wings. (abridged)

This article presents experiments performed using a computational laboratory environment for language acquisition experiments. It implements a multi-agent system consisting of two agents: an adult language model and a daughter language model that aims to learn the mother language. Crucially, the daughter agent does not have access to the internal knowledge of the mother language model but only to the language exemplars the mother agent generates. These experiments illustrate how this system can be used to acquire abstract grammatical knowledge. We demonstrate how statistical analyses of patterns in the input data corresponding to grammatical categories yield discrete grammatical rules. These rules are subsequently added to the grammatical knowledge of the daughter language model. To this end, hierarchical agglomerative cluster analysis was applied to the utterances consecutively generated by the mother language model. It is argued that this procedure can be used to acquire structures resembling grammatical categories proposed by linguists for natural languages. Thus, it is established that non-trivial grammatical knowledge has been acquired. Moreover, the parameter configuration of this computational laboratory environment determined using training data generated by the mother language model is validated in a second experiment with a test set similarly resulting in the acquisition of non-trivial categories.

Large language models (LLMs) are increasingly entrusted with high-stakes decisions that affect human welfare. However, the principles and values that guide these models when distributing scarce societal resources remain largely unexamined. To address this, we introduce the Social Welfare Function (SWF) Benchmark, a dynamic simulation environment where an LLM acts as a sovereign allocator, distributing tasks to a heterogeneous community of recipients. The benchmark is designed to create a persistent trade-off between maximizing collective efficiency (measured by Return on Investment) and ensuring distributive fairness (measured by the Gini coefficient). We evaluate 20 state-of-the-art LLMs and present the first leaderboard for social welfare allocation. Our findings reveal three key insights: (i) A model's general conversational ability, as measured by popular leaderboards, is a poor predictor of its allocation skill. (ii) Most LLMs exhibit a strong default utilitarian orientation, prioritizing group productivity at the expense of severe inequality. (iii) Allocation strategies are highly vulnerable, easily perturbed by output-length constraints and social-influence framing. These results highlight the risks of deploying current LLMs as societal decision-makers and underscore the need for specialized benchmarks and targeted alignment for AI governance.