Search

Large scale deep learning model, such as modern language models and diffusion architectures, have revolutionized applications ranging from natural language processing to computer vision. However, their deployment in distributed or decentralized environments raises significant privacy concerns, as sensitive data may be exposed during inference. Traditional techniques like secure multi-party computation, homomorphic encryption, and differential privacy offer partial remedies but often incur substantial computational overhead, latency penalties, or limited compatibility with non-linear network operations. In this work, we introduce Equivariant Encryption (EE), a novel paradigm designed to enable secure, "blind" inference on encrypted data with near zero performance overhead. Unlike fully homomorphic approaches that encrypt the entire computational graph, EE selectively obfuscates critical internal representations within neural network layers while preserving the exact functionality of both linear and a prescribed set of non-linear operations. This targeted encryption ensures that raw inputs, intermediate activations, and outputs remain confidential, even when processed on untrusted infrastructure. We detail the theoretical foundations of EE, compare its performance and integration complexity against conventional privacy preserving techniques, and demonstrate its applicability across a range of architectures, from convolutional networks to large language models. Furthermore, our work provides a comprehensive threat analysis, outlining potential attack vectors and baseline strategies, and benchmarks EE against standard inference pipelines in decentralized settings. The results confirm that EE maintains high fidelity and throughput, effectively bridging the gap between robust data confidentiality and the stringent efficiency requirements of modern, large scale model inference.

Active Galactic Nuclei (AGN) feedback is a key physical mechanism proposed to regulate star formation, primarily in massive galaxies. In particular, cosmic rays associated with AGN jets have the potential to efficiently suppress cooling flows and quench star formation. The locus of cosmic ray production and their coupling to gas play a crucial role in the overall self-regulation process. To investigate this in detail, we conduct high-resolution, non-cosmological MHD simulations of a massive $10^{14} {\rm M_\odot}$ halo using the FIRE-2 (Feedback In Realistic Environments) stellar feedback model. We explore a variety of AGN jet feedback scenarios with cosmic rays, examining different values for the cosmic ray energy fraction in jets, cosmic ray coupling sites (in the black hole vicinity versus at the large-scale jet-driven shock front), and jet precession parameters. Our findings indicate that when cosmic rays are injected near the black hole, they efficiently inhibit black hole accretion by suppressing the density before the jet propagates out to large radii. As a result, this leads to episodic black hole accretion, with the jet not having sufficient energy flux to reach large radii and impact cooling flows. Conversely, if the cosmic rays are injected at the jet-driven shock front, not only does the jet sustain a higher overall energy flux for an extended period, but it also disperses cosmic rays out to larger radii, more effectively suppressing the cooling flow. Furthermore, the period and angle of jet precession can influence the position of shock fronts. We identify an optimal range of jet precession periods ($\sim$ tens of Myr) that generates shocks at the inner circumgalactic medium, where cooling flows are most severe. We report that this specific configuration offers the most effective scenario for cosmic rays at the shock front to suppress the cooling flow and star formation., Comment: 17 pages, 12 figures

Reinforcement Learning from Human Feedback (RLHF) has achieved considerable success in aligning large language models (LLMs) by modeling human preferences with a learnable reward model and employing a reinforcement learning algorithm to maximize the reward model's scores. However, these reward models are susceptible to exploitation through various superficial confounding factors, with length bias emerging as a particularly significant concern. Moreover, while the pronounced impact of length bias on preference modeling suggests that LLMs possess an inherent sensitivity to length perception, our preliminary investigations reveal that fine-tuned LLMs consistently struggle to adhere to explicit length instructions. To address these two limitations, we propose a novel framework wherein the reward model explicitly differentiates between human semantic preferences and response length requirements. Specifically, we introduce a Response-conditioned Bradley-Terry (Rc-BT) model that enhances the reward model's capability in length bias mitigating and length instruction following, through training on our augmented dataset. Furthermore, we propose the Rc-DPO algorithm to leverage the Rc-BT model for direct policy optimization (DPO) of LLMs, simultaneously mitigating length bias and promoting adherence to length instructions. Extensive evaluations demonstrate that our approach substantially improves both preference modeling and length instruction compliance, with its effectiveness validated across various foundational models and preference datasets.

Existing foundation models, such as CLIP, aim to learn a unified embedding space for multimodal data, enabling a wide range of downstream web-based applications like search, recommendation, and content classification. However, these models often overlook the inherent graph structures in multimodal datasets, where entities and their relationships are crucial. Multimodal graphs (MMGs) represent such graphs where each node is associated with features from different modalities, while the edges capture the relationships between these entities. On the other hand, existing graph foundation models primarily focus on text-attributed graphs (TAGs) and are not designed to handle the complexities of MMGs. To address these limitations, we propose UniGraph2, a novel cross-domain graph foundation model that enables general representation learning on MMGs, providing a unified embedding space. UniGraph2 employs modality-specific encoders alongside a graph neural network (GNN) to learn a unified low-dimensional embedding space that captures both the multimodal information and the underlying graph structure. We propose a new cross-domain multi-graph pre-training algorithm at scale to ensure effective transfer learning across diverse graph domains and modalities. Additionally, we adopt a Mixture of Experts (MoE) component to align features from different domains and modalities, ensuring coherent and robust embeddings that unify the information across modalities. Extensive experiments on a variety of multimodal graph tasks demonstrate that UniGraph2 significantly outperforms state-of-the-art models in tasks such as representation learning, transfer learning, and multimodal generative tasks, offering a scalable and flexible solution for learning on MMGs., Comment: WWW 2025

In this paper, we introduce a linear stochastic volatility model driven by $\alpha$-stable processes, which admits a unique positive solution. To preserve positivity, we modify the classical forward Euler-Maruyama scheme and analyze its numerical properties. The scheme achieves a strong convergence order of $1/\alpha$. Numerical simulations are presented at the end to verify theoretical results.

Fractional vegetation coverage (FVC) and its spatio-temporal variations are critical indicators of regional ecological changes, which are of great significance to study the laws of surface variation and analyze regional ecosystem. Under the development of RS and GIS technology, this analysis employs Landsat satellite images in 1994, 2008, 2013 and 2016 to estimate FVC in Yalong River Basin based on the Dimidiate Pixel Model. With consideration of the vegetation coverage condition and land surface law in the study area, the research further analyzes the Spatio-temporal variations as well as the influencing factors of FVC in terms of topography and land use types respectively. The results show that since 1994, FVC in Yalong River Basin has experienced a downward trend yet displaying an uptick from 2013. Moreover, different land use types indicate the versatility of land covers in Yalong River Basin, with grassland and forest performing probably the most important factors that can induce changes to the stability of FVC in whole basin. Overall, the research reflects the impact of human activities on vegetation in Yalong River Basin, and provides available data and theoretical basis for ecological assessment, ecological restoration and environmental protection.

The rise of single-cell sequencing technologies has revolutionized the exploration of drug resistance, revealing the crucial role of cellular heterogeneity in advancing precision medicine. By building computational models from existing single-cell drug response data, we can rapidly annotate cellular responses to drugs in subsequent trials. To this end, we developed scGSDR, a model that integrates two computational pipelines grounded in the knowledge of cellular states and gene signaling pathways, both essential for understanding biological gene semantics. scGSDR enhances predictive performance by incorporating gene semantics and employs an interpretability module to identify key pathways contributing to drug resistance phenotypes. Our extensive validation, which included 16 experiments covering 11 drugs, demonstrates scGSDR's superior predictive accuracy, when trained with either bulk-seq or scRNA-seq data, achieving high AUROC, AUPR, and F1 Scores. The model's application has extended from single-drug predictions to scenarios involving drug combinations. Leveraging pathways of known drug target genes, we found that scGSDR's cell-pathway attention scores are biologically interpretable, which helped us identify other potential drug-related genes. Literature review of top-ranking genes in our predictions such as BCL2, CCND1, the AKT family, and PIK3CA for PLX4720; and ICAM1, VCAM1, NFKB1, NFKBIA, and RAC1 for Paclitaxel confirmed their relevance. In conclusion, scGSDR, by incorporating gene semantics, enhances predictive modeling of cellular responses to diverse drugs, proving invaluable for scenarios involving both single drug and combination therapies and effectively identifying key resistance-related pathways, thus advancing precision medicine and targeted therapy development.

Generative models serve as powerful tools for modeling the real world, with mainstream diffusion models, particularly those based on the latent diffusion model paradigm, achieving remarkable progress across various tasks, such as image and video synthesis. Latent diffusion models are typically trained using Variational Autoencoders (VAEs), interacting with VAE latents rather than the real samples. While this generative paradigm speeds up training and inference, the quality of the generated outputs is limited by the latents' quality. Traditional VAE latents are often seen as spatial compression in pixel space and lack explicit semantic representations, which are essential for modeling the real world. In this paper, we introduce ReaLS (Representation-Aligned Latent Space), which integrates semantic priors to improve generation performance. Extensive experiments show that fundamental DiT and SiT trained on ReaLS can achieve a 15% improvement in FID metric. Furthermore, the enhanced semantic latent space enables more perceptual downstream tasks, such as segmentation and depth estimation.

We propose a theoretical framework demonstrating that popular Large Language Model (LLM) alignment methods, including Reinforcement Learning from Human Feedback (RLHF) and alternatives, fundamentally function as divergence estimators between aligned (preferred or safe) and unaligned (less-preferred or harmful) distributions. This explains the separation phenomenon between safe and harmful prompts in the model hidden representation after alignment. Inspired by the theoretical results, we identify that some alignment methods are better than others in terms of separation and, introduce a new method, KLDO, and further demonstrate the implication of our theories. We advocate for compliance-refusal datasets over preference datasets to enhance safety alignment, supported by both theoretical reasoning and empirical evidence. Additionally, to quantify safety separation, we leverage a distance metric in the representation space and statistically validate its efficacy as a statistical significant indicator of LLM resilience against jailbreak attacks.

Cutting thin-walled deformable structures is common in daily life, but poses significant challenges for simulation due to the introduced spatial discontinuities. Traditional methods rely on mesh-based domain representations, which require frequent remeshing and refinement to accurately capture evolving discontinuities. These challenges are further compounded in reduced-space simulations, where the basis functions are inherently geometry- and mesh-dependent, making it difficult or even impossible for the basis to represent the diverse family of discontinuities introduced by cuts. Recent advances in representing basis functions with neural fields offer a promising alternative, leveraging their discretization-agnostic nature to represent deformations across varying geometries. However, the inherent continuity of neural fields is an obstruction to generalization, particularly if discontinuities are encoded in neural network weights. We present Wind Lifter, a novel neural representation designed to accurately model complex cuts in thin-walled deformable structures. Our approach constructs neural fields that reproduce discontinuities precisely at specified locations, without baking in the position of the cut line. Crucially, our approach does not embed the discontinuity in the neural network's weights, opening avenues to generalization of cut placement. Our method achieves real-time simulation speeds and supports dynamic updates to cut line geometry during the simulation. Moreover, the explicit representation of discontinuities makes our neural field intuitive to control and edit, offering a significant advantage over traditional neural fields, where discontinuities are embedded within the network's weights, and enabling new applications that rely on general cut placement.

Predicting web service traffic has significant social value, as it can be applied to various practical scenarios, including but not limited to dynamic resource scaling, load balancing, system anomaly detection, service-level agreement compliance, and fraud detection. Web service traffic is characterized by frequent and drastic fluctuations over time and are influenced by heterogeneous web user behaviors, making accurate prediction a challenging task. Previous research has extensively explored statistical approaches, and neural networks to mine features from preceding service traffic time series for prediction. However, these methods have largely overlooked the causal relationships between services. Drawing inspiration from causality in ecological systems, we empirically recognize the causal relationships between web services. To leverage these relationships for improved web service traffic prediction, we propose an effective neural network module, CCMPlus, designed to extract causal relationship features across services. This module can be seamlessly integrated with existing time series models to consistently enhance the performance of web service traffic predictions. We theoretically justify that the causal correlation matrix generated by the CCMPlus module captures causal relationships among services. Empirical results on real-world datasets from Microsoft Azure, Alibaba Group, and Ant Group confirm that our method surpasses state-of-the-art approaches in Mean Squared Error (MSE) and Mean Absolute Error (MAE) for predicting service traffic time series. These findings highlight the efficacy of leveraging causal relationships for improved predictions., Comment: time series, web service, web traffic, causality

The intercalated two-dimensional (2D) transition metal dichalcogenides (TMDCs) have attracted much attention for their designable structure and novel properties. Among this family, host materials with low symmetry such as 1T' phase TMDCs are particularly interesting because of their potentials in inducing unconventional phenomena. However, such systems typically have low quality and poor stability, hindering further study in the structure-property relationship and applications. In this work, we intercalated Cu into 1T' MoS2 with high crystallinity and high thermal stability up to ~300 oC. We identified the distribution and arrangement of Cu intercalators for the first time, and the results show that Cu occupy partial of the tetrahedral interstices aligned with Mo sites. The obtained Cu-1T' MoS2 exhibits an insulating hopping transport behavior with a large temperature coefficient of resistance reaching -4 ~ -2 % K-1. This work broadens the artificial intercalated structure library and promotes structure design and property modulation of layered materials., Comment: 20 pages, 4 figures. Accepted

To gain panoramic awareness of spectrum coverage in complex wireless environments, data-driven learning approaches have recently been introduced for radio map estimation (RME). While existing deep learning based methods conduct RME given spectrum measurements gathered from dispersed sensors in the region of interest, they rely on centralized data at a fusion center, which however raises critical concerns on data privacy leakages and high communication overloads. Federated learning (FL) enhance data security and communication efficiency in RME by allowing multiple clients to collaborate in model training without directly sharing local data. However, the performance of the FL-based RME can be hindered by the problem of task heterogeneity across clients due to their unavailable or inaccurate landscaping information. To fill this gap, in this paper, we propose a physics-inspired distributed RME solution in the absence of landscaping information. The main idea is to develop a novel distributed RME framework empowered by leveraging the domain knowledge of radio propagation models, and by designing a new distributed learning approach that splits the entire RME model into two modules. A global autoencoder module is shared among clients to capture the common pathloss influence on radio propagation pattern, while a client-specific autoencoder module focuses on learning the individual features produced by local shadowing effects from the unique building distributions in local environment. Simulation results show that our proposed method outperforms the benchmarks in achieving higher performance.

Brain-inspired computing architectures attempt to emulate the computations performed in the neurons and the synapses in human brain. Memristors with continuously tunable resistances are ideal building blocks for artificial synapses. Through investigating the memristor behaviors in a La0.7Sr0.3MnO3/BaTiO3/La0.7Sr0.3MnO3 multiferroic tunnel junction, it was found that the ferroelectric domain dynamics characteristics are influenced by the relative magnetization alignment of the electrodes, and the interfacial spin polarization is manipulated continuously by ferroelectric domain reversal, enriching our understanding of the magnetoelectric coupling fundamentally. This creates a functionality that not only the resistance of the memristor but also the synaptic plasticity form can be further manipulated, as demonstrated by the spike-timing-dependent plasticity investigations. Density functional theory calculations are carried out to describe the obtained magnetoelectric coupling, which is probably related to the Mn-Ti intermixing at the interfaces. The multiple and controllable plasticity characteristic in a single artificial synapse, to resemble the synaptic morphological alteration property in a biological synapse, will be conducive to the development of artificial intelligence., Comment: 5 figures, 20 pages

Functionalities in crystalline materials are determined by 3-dimensional collective interactions of atoms. The confinement of dimensionality in condensed matter provides an exotic research direction to understand the interaction of atoms, thus can be used to tailor or create new functionalities in material systems. In this study, a 2-dimensional transition metal oxide monolayer is constructed inside complex oxide heterostructures based on the theoretical predictions. The electrostatic boundary conditions of oxide monolayer in the heterostructure is carefully designed to tune the chemical, electronic, and magnetic states of oxide monolayer. The challenge of characterizing such an oxide monolayer is overcome by a combination of transmission electron microscopy, x-ray absorption spectroscopy, cross-sectional scanning tunneling microscopy, and electrical transport measurements. An intriguing metal-insulator transition associated with a magnetic transition is discovered in the MnO2 monolayer. This study paves a new route to understand the confinement of dimensionality and explore new intriguing phenomena in condensed matters., Comment: 23 pages, 6 figures

With the rapid development of wireless communication technology, the efficient utilization of spectrum resources, optimization of communication quality, and intelligent communication have become critical. Radio map reconstruction is essential for enabling advanced applications, yet challenges such as complex signal propagation and sparse data hinder accurate reconstruction. To address these issues, we propose the **Radio Map Diffusion Model (RMDM)**, a physics-informed framework that integrates **Physics-Informed Neural Networks (PINNs)** to incorporate constraints like the **Helmholtz equation**. RMDM employs a dual U-Net architecture: the first ensures physical consistency by minimizing PDE residuals, boundary conditions, and source constraints, while the second refines predictions via diffusion-based denoising. By leveraging physical laws, RMDM significantly enhances accuracy, robustness, and generalization. Experiments demonstrate that RMDM outperforms state-of-the-art methods, achieving **NMSE of 0.0031** and **RMSE of 0.0125** under the Static RM (SRM) setting, and **NMSE of 0.0047** and **RMSE of 0.0146** under the Dynamic RM (DRM) setting. These results establish a novel paradigm for integrating physics-informed and data-driven approaches in radio map reconstruction, particularly under sparse data conditions.

Despite the great performance of deep learning models in many areas, they still make mistakes and underperform on certain subsets of data, i.e. error slices. Given a trained model, it is important to identify its semantically coherent error slices that are easy to interpret, which is referred to as the error slice discovery problem. However, there is no proper metric of slice coherence without relying on extra information like predefined slice labels. Current evaluation of slice coherence requires access to predefined slices formulated by metadata like attributes or subclasses. Its validity heavily relies on the quality and abundance of metadata, where some possible patterns could be ignored. Besides, current algorithms cannot directly incorporate the constraint of coherence into their optimization objective due to the absence of an explicit coherence metric, which could potentially hinder their effectiveness. In this paper, we propose manifold compactness, a coherence metric without reliance on extra information by incorporating the data geometry property into its design, and experiments on typical datasets empirically validate the rationality of the metric. Then we develop Manifold Compactness based error Slice Discovery (MCSD), a novel algorithm that directly treats risk and coherence as the optimization objective, and is flexible to be applied to models of various tasks. Extensive experiments on the benchmark and case studies on other typical datasets demonstrate the superiority of MCSD.

Large language models (LLMs) such as OpenAI's o1 have demonstrated remarkable abilities in complex reasoning tasks by scaling test-time compute and exhibiting human-like deep thinking. However, we identify a phenomenon we term underthinking, where o1-like LLMs frequently switch between different reasoning thoughts without sufficiently exploring promising paths to reach a correct solution. This behavior leads to inadequate depth of reasoning and decreased performance, particularly on challenging mathematical problems. To systematically analyze this issue, we conduct experiments on three challenging test sets and two representative open-source o1-like models, revealing that frequent thought switching correlates with incorrect responses. We introduce a novel metric to quantify underthinking by measuring token efficiency in incorrect answers. To address underthinking, we propose a decoding strategy with thought switching penalty TIP that discourages premature transitions between thoughts, encouraging deeper exploration of each reasoning path. Experimental results demonstrate that our approach improves accuracy across challenging datasets without requiring model fine-tuning. Our findings contribute to understanding reasoning inefficiencies in o1-like LLMs and offer a practical solution to enhance their problem-solving capabilities.

Newton-Okounkov bodies serve as a bridge between algebraic geometry and convex geometry, enabling the application of combinatorial and geometric methods to the study of linear systems on algebraic varieties. This paper contributes to understanding the algebro-geometric information encoded in the collection of all Newton-Okounkov bodies on a given surface, focusing on Newton-Okounkov polygons with few vertices and on elliptic K3 surfaces. Let S be an algebraic surface and mv(S) be the maximum number of vertices of the Newton-Okounkov bodies of S. We prove that mv(S) = 4 if and only if its Picard number \rho(S) is at least 2 and S contains no negative irreducible curve. Additionally, if S contains a negative curve, then \rho(S) = 2 if and only if mv(S) = 5. Furthermore, we provide an example involving two elliptic K3 surfaces to demonstrate that when \rho(S) \geq 3, mv(S) no longer determines the Picard number \rho(S).

As LLMs increasingly impact safety-critical applications, ensuring their safety using guardrails remains a key challenge. This paper proposes GuardReasoner, a new safeguard for LLMs, by guiding the guard model to learn to reason. Concretely, we first create the GuardReasonerTrain dataset, which consists of 127K samples with 460K detailed reasoning steps. Then, we introduce reasoning SFT to unlock the reasoning capability of guard models. In addition, we present hard sample DPO to further strengthen their reasoning ability. In this manner, GuardReasoner achieves better performance, explainability, and generalizability. Extensive experiments and analyses on 13 benchmarks of 3 guardrail tasks demonstrate its superiority. Remarkably, GuardReasoner 8B surpasses GPT-4o+CoT by 5.74% and LLaMA Guard 3 8B by 20.84% F1 score on average. We release the training data, code, and models with different scales (1B, 3B, 8B) of GuardReasoner : https://github.com/yueliu1999/GuardReasoner/., Comment: 22 pages, 18 figures

Rapid progress in text-to-motion generation has been largely driven by diffusion models. However, existing methods focus solely on temporal modeling, thereby overlooking frequency-domain analysis. We identify two key phases in motion denoising: the **semantic planning stage** and the **fine-grained improving stage**. To address these phases effectively, we propose **Fre**quency **e**nhanced **t**ext-**to**-**m**otion diffusion model (**Free-T2M**), incorporating stage-specific consistency losses that enhance the robustness of static features and improve fine-grained accuracy. Extensive experiments demonstrate the effectiveness of our method. Specifically, on StableMoFusion, our method reduces the FID from **0.189** to **0.051**, establishing a new SOTA performance within the diffusion architecture. These findings highlight the importance of incorporating frequency-domain insights into text-to-motion generation for more precise and robust results.

Pancreatic NEuroendocrine Tumors (pNETs) are very rare endocrine neoplasms that account for less than 5% of all pancreatic malignancies, with an incidence of only 1-1.5 cases per 100,000. Early detection of pNETs is critical for improving patient survival, but the rarity of pNETs makes segmenting them from CT a very challenging problem. So far, there has not been a dataset specifically for pNETs available to researchers. To address this issue, we propose a pNETs dataset, a well-annotated Contrast-Enhanced Computed Tomography (CECT) dataset focused exclusively on Pancreatic Neuroendocrine Tumors, containing data from 469 patients. This is the first dataset solely dedicated to pNETs, distinguishing it from previous collections. Additionally, we provide the baseline detection networks with a new slice-wise weight loss function designed for the UNet-based model, improving the overall pNET segmentation performance. We hope that our dataset can enhance the understanding and diagnosis of pNET Tumors within the medical community, facilitate the development of more accurate diagnostic tools, and ultimately improve patient outcomes and advance the field of oncology.

Quantum networks promise an unprecedented leap in semi-device-independent communication and security by capitalizing on quantum steering. However, current methods for assessing quantum network steering are constrained to specific cases. In this work, we introduce the network-Clauser-Horn-Shimony-Holt-like inequality for investigating network steering independent of entanglement source characteristics. We employ this inequality to detect full network steering in both single-node and multinode repeater networks and assess the tolerance of various noise models. Under a specific noise model, our method is used to compute the bound of semi-device-independent communication distance. Through case studies, we also demonstrate that our method, as a semi-device-independent entanglement witness, is more suitable for settings requiring Bell measurements compared to network Bell inequality. These findings open new avenues for broader ways to detect quantum steering independent of entanglement sources., Comment: 14 pages, 7 figures

Modern recommendation systems can be broadly divided into two key stages: the ranking stage, where the system predicts various user engagements (e.g., click-through rate, like rate, follow rate, watch time), and the value model stage, which aggregates these predictive scores through a function (e.g., a linear combination defined by a weight vector) to measure the value of each content by a single numerical score. Both stages play roughly equally important roles in real industrial systems; however, how to optimize the model weights for the second stage still lacks systematic study. This paper focuses on optimizing the second stage through auto-tuning technology. Although general auto-tuning systems and solutions - both from established production practices and open-source solutions - can address this problem, they typically require weeks or even months to identify a feasible solution. Such prolonged tuning processes are unacceptable in production environments for recommendation systems, as suboptimal value models can severely degrade user experience. An effective auto-tuning solution is required to identify a viable model within 2-3 days, rather than the extended timelines typically associated with existing approaches. In this paper, we introduce a practical auto-tuning system named HyperZero that addresses these time constraints while effectively solving the unique challenges inherent in modern recommendation systems. Moreover, this framework has the potential to be expanded to broader tuning tasks within recommendation systems.

Topological Floquet energy pumps -- which use periodic driving to create a topologically protected quantized energy current -- have been proposed and studied theoretically, but have never been observed directly. Previous work proposed that such a pump could be realized with a strongly-driven superconducting qubit coupled to a cavity. Here, we experimentally demonstrate that the proposed hierarchy of energy scales and drive frequencies can be realized using a transmon qubit. We develop an experimental toolkit to realize the adiabatic driving field required for energy pumping using coordinated frequency modulation of the transmon and amplitude modulation of an applied resonant microwave drive. With this toolkit, we measure adiabatic evolution of the qubit under the applied field for times comparable to $T_1$, which far exceed the bare qubit dephasing time. This result paves the way for direct experimental observation of topological energy pumping., Comment: 19 pages, 14 figures

Supervised Fine-Tuning (SFT) is commonly used to train language models to imitate annotated responses for given instructions. In this paper, we challenge this paradigm and propose Critique Fine-Tuning (CFT), a strategy where models learn to critique noisy responses rather than simply imitate correct ones. Inspired by human learning processes that emphasize critical thinking, CFT encourages deeper analysis and nuanced understanding-traits often overlooked by standard SFT. To validate the effectiveness of CFT, we construct a 50K-sample dataset from WebInstruct, using GPT-4o as the teacher to generate critiques in the form of ([query; noisy response], critique). CFT on this dataset yields a consistent 4-10% improvement over SFT on six math benchmarks with different base models like Qwen2.5, Qwen2.5-Math and DeepSeek-Math. We further expand to MetaMath and NuminaMath datasets and observe similar gains over SFT. Notably, our model Qwen2.5-Math-CFT only requires 1 hour training on 8xH100 over the 50K examples. It can match or outperform strong competitors like Qwen2.5-Math-Instruct on most benchmarks, which use over 2M samples. Moreover, it can match the performance of SimpleRL, which is a deepseek-r1 replication trained with 140x more compute. Ablation studies show that CFT is robust to the source of noisy response and teacher critique model. Through these findings, we argue that CFT offers a more effective alternative to advance the reasoning of language models.

We present MultiChallenge, a pioneering benchmark evaluating large language models (LLMs) on conducting multi-turn conversations with human users, a crucial yet underexamined capability for their applications. MultiChallenge identifies four categories of challenges in multi-turn conversations that are not only common and realistic among current human-LLM interactions, but are also challenging to all current frontier LLMs. All 4 challenges require accurate instruction-following, context allocation, and in-context reasoning at the same time. We also develop LLM as judge with instance-level rubrics to facilitate an automatic evaluation method with fair agreement with experienced human raters. Despite achieving near-perfect scores on existing multi-turn evaluation benchmarks, all frontier models have less than 50% accuracy on MultiChallenge, with the top-performing Claude 3.5 Sonnet (June 2024) achieving just a 41.4% average accuracy.

Simulation plays a crucial role in assessing autonomous driving systems, where the generation of realistic multi-agent behaviors is a key aspect. In multi-agent simulation, the primary challenges include behavioral multimodality and closed-loop distributional shifts. In this study, we revisit mixture models for generating multimodal agent behaviors, which can cover the mainstream methods including continuous mixture models and GPT-like discrete models. Furthermore, we introduce a closed-loop sample generation approach tailored for mixture models to mitigate distributional shifts. Within the unified mixture model~(UniMM) framework, we recognize critical configurations from both model and data perspectives. We conduct a systematic examination of various model configurations, including positive component matching, continuous regression, prediction horizon, and the number of components. Moreover, our investigation into the data configuration highlights the pivotal role of closed-loop samples in achieving realistic simulations. To extend the benefits of closed-loop samples across a broader range of mixture models, we further address the shortcut learning and off-policy learning issues. Leveraging insights from our exploration, the distinct variants proposed within the UniMM framework, including discrete, anchor-free, and anchor-based models, all achieve state-of-the-art performance on the WOSAC benchmark.

The increasing scarcity of spectrum resources and the rapid growth of wireless device have made efficient management of radio networks a critical challenge. Cognitive Radio Technology (CRT), when integrated with deep learning (DL), offers promising solutions for tasks such as radio signal classification (RSC), signal denoising, and spectrum allocation. However, existing DL-based CRT frameworks are often task-specific and lack scalability to diverse real-world scenarios. Meanwhile, Large Language Models (LLMs) have demonstrated exceptional generalization capabilities across multiple domains, making them a potential candidate for advancing CRT technologies. In this paper, we introduce RadioLLM, a novel framework that incorporates Hybrid Prompt and Token Reprogramming (HPTR) and a Frequency Attuned Fusion (FAF) module to enhance LLMs for CRT tasks. HPTR enables the integration of radio signal features with expert knowledge, while FAF improves the modeling of high-frequency features critical for precise signal processing. These innovations allow RadioLLM to handle diverse CRT tasks, bridging the gap between LLMs and traditional signal processing methods. Extensive empirical studies on multiple benchmark datasets demonstrate that the proposed RadioLLM achieves superior performance over current baselines.

Here we present results from our spectroscopic follow-up of SDSS J2320+0024, a candidate binary supermassive black hole (SMBH) with a suspected sub-pc separation, identified by a 278-day periodicity observed in its multi-band optical light curves. We investigate the dramatic variability of the complex Mg II emission line profile aiming to test the alignments of the observed photometric light curves and the spectroscopic signatures in the context of the binary SMBH system. We extract the pure broad Mg II line from the newly obtained Gemini and Magellan spectra and measure the emission line parameters to reveal fundamental dynamical parameters of the SMBHs binary system. We adopt the PoSKI sub-pc binary SMBH model, which includes broad-line region (BLR) around less massive component and a circumbinary BLR, to interpret the observed variability in the spectral profile. We find that the Mg II line profile has a distinctive complex shape with the asymmetry and two peaks present which is varying across recent and archival observations. The temporal variability of the Mg II line profile may be associated with the emission from the binary SMBH system consisting of components with masses $M_1 = 2 \times 10^7 \, M_{\odot}$ and $M_2 = 2 \times 10^8 \, M_{\odot}$, and eccentricity e = 0.1. With an total estimated mass of $\sim 10^9 M_{\odot}$ and a subannual orbital period, this system may be a rare example of high-mass compact candidate of SMBH binary, thus important for further investigations of the evolution of the binary system. This study is a prototype of synergies of spectroscopic follow-up and future massive time-domain photometric surveys like Vera C. Rubin Observatory Legacy Survey of Space and Time., Comment: Accepted for publication in A&A

Utilizing the phenomenological theory based on crystal symmetry operation, we have established the complete angular dependencies of the galvanomagnetic effects, encompassing both anisotropic magnetoresistance (AMR) and the planar Hall effect (PHE), for the ferromagnetic films with C4v symmetry. These dependencies were experimentally confirmed via comprehensive angular-mapping of AMR and PHE in single-crystal Fe(001) films at room temperature. We demonstrated that the intrinsic magnetization-induced effects are independent of the field strength by carefully separating the field-induced and magnetization-induced galvanomagnetic effects. Our theoretical and experimental findings highlight the absence of in-plane four-fold angular dependence in PHE, a feature prohibited by the Onsager relation in systems with C4 symmetry. This study affirms that the universal angular dependencies of AMR and PHE in single crystals can be accurately predicted by the conventional phenomenological theory., Comment: 32 pages, 6 figures, with supplemental material

The rapidly expanding Internet of Things (IoT) landscape is shifting toward cloudless architectures, removing reliance on centralized cloud services but exposing devices directly to the internet and increasing their vulnerability to cyberattacks. Our research revealed an unexpected pattern of substantial Tor network traffic targeting cloudless IoT devices. suggesting that attackers are using Tor to anonymously exploit undisclosed vulnerabilities (possibly obtained from underground markets). To delve deeper into this phenomenon, we developed TORCHLIGHT, a tool designed to detect both known and unknown threats targeting cloudless IoT devices by analyzing Tor traffic. TORCHLIGHT filters traffic via specific IP patterns, strategically deploys virtual private server (VPS) nodes for cost-effective detection, and uses a chain-of-thought (CoT) process with large language models (LLMs) for accurate threat identification. Our results are significant: for the first time, we have demonstrated that attackers are indeed using Tor to conceal their identities while targeting cloudless IoT devices. Over a period of 12 months, TORCHLIGHT analyzed 26 TB of traffic, revealing 45 vulnerabilities, including 29 zero-day exploits with 25 CVE-IDs assigned (5 CRITICAL, 3 HIGH, 16 MEDIUM, and 1 LOW) and an estimated value of approximately $312,000. These vulnerabilities affect around 12.71 million devices across 148 countries, exposing them to severe risks such as information disclosure, authentication bypass, and arbitrary command execution. The findings have attracted significant attention, sparking widespread discussion in cybersecurity circles, reaching the top 25 on Hacker News, and generating over 190,000 views., Comment: 27 pages, 14 figure, 9 tables

3D Gaussian Splatting (3DGS) has advanced radiance field reconstruction by enabling real-time rendering. However, its reliance on Gaussian kernels for geometry and low-order Spherical Harmonics (SH) for color encoding limits its ability to capture complex geometries and diverse colors. We introduce Deformable Beta Splatting (DBS), a deformable and compact approach that enhances both geometry and color representation. DBS replaces Gaussian kernels with deformable Beta Kernels, which offer bounded support and adaptive frequency control to capture fine geometric details with higher fidelity while achieving better memory efficiency. In addition, we extended the Beta Kernel to color encoding, which facilitates improved representation of diffuse and specular components, yielding superior results compared to SH-based methods. Furthermore, Unlike prior densification techniques that depend on Gaussian properties, we mathematically prove that adjusting regularized opacity alone ensures distribution-preserved Markov chain Monte Carlo (MCMC), independent of the splatting kernel type. Experimental results demonstrate that DBS achieves state-of-the-art visual quality while utilizing only 45% of the parameters and rendering 1.5x faster than 3DGS-based methods. Notably, for the first time, splatting-based methods outperform state-of-the-art Neural Radiance Fields, highlighting the superior performance and efficiency of DBS for real-time radiance field rendering.

Assisted Reality (aR) is a subfield of Augmented Reality (AR) that overlays information onto a user's immediate view via see-through head-mounted displays (OST-HMDs). This technology has proven to be effective and energy-efficient to support the user and information interaction for everyday wearable intelligent systems. The aR viewing experience, however, is affected by varying real-world backgrounds, lighting, and user movements, which makes designing for aR challenging. Designers have to test their designs in-situ across multiple real-world settings, which can be time-consuming and labor-intensive. We propose SimulataR, a cost-effective desktop-based approach for rapid aR prototyping using first-person-view context videos blended with design prototypes to simulate an aR experience. A field study involving 12 AR users comparing SimulataR to real OST-HMDs found that SimulataR can approximate the aR experience, particularly for indoors and in low-to-moderate lit outdoor environments. Case studies with two designers who used SimulataR in their design process demonstrates the potential of design-blended videos for rapid aR prototyping.

Understanding the physical world is a fundamental challenge in embodied AI, critical for enabling agents to perform complex tasks and operate safely in real-world environments. While Vision-Language Models (VLMs) have shown great promise in reasoning and task planning for embodied agents, their ability to comprehend physical phenomena remains extremely limited. To close this gap, we introduce PhysBench, a comprehensive benchmark designed to evaluate VLMs' physical world understanding capability across a diverse set of tasks. PhysBench contains 10,002 entries of interleaved video-image-text data, categorized into four major domains: physical object properties, physical object relationships, physical scene understanding, and physics-based dynamics, further divided into 19 subclasses and 8 distinct capability dimensions. Our extensive experiments, conducted on 75 representative VLMs, reveal that while these models excel in common-sense reasoning, they struggle with understanding the physical world -- likely due to the absence of physical knowledge in their training data and the lack of embedded physical priors. To tackle the shortfall, we introduce PhysAgent, a novel framework that combines the generalization strengths of VLMs with the specialized expertise of vision models, significantly enhancing VLMs' physical understanding across a variety of tasks, including an 18.4\% improvement on GPT-4o. Furthermore, our results demonstrate that enhancing VLMs' physical world understanding capabilities can help embodied agents such as MOKA. We believe that PhysBench and PhysAgent offer valuable insights and contribute to bridging the gap between VLMs and physical world understanding., Comment: ICLR 2025. Project page: https://physbench.github.io/ Dataset: https://huggingface.co/datasets/USC-GVL/PhysBench

We study the Out-of-Distribution (OoD) generalization ability of three SotA trajectory prediction models with comparable In-Distribution (ID) performance but different model designs. We investigate the influence of inductive bias, size of training data and data augmentation strategy by training the models on Argoverse 2 (A2) and testing on Waymo Open Motion (WO) and vice versa. We find that the smallest model with highest inductive bias exhibits the best OoD generalization across different augmentation strategies when trained on the smaller A2 dataset and tested on the large WO dataset. In the converse setting, training all models on the larger WO dataset and testing on the smaller A2 dataset, we find that all models generalize poorly, even though the model with the highest inductive bias still exhibits the best generalization ability. We discuss possible reasons for this surprising finding and draw conclusions about the design and test of trajectory prediction models and benchmarks., Comment: arXiv admin note: text overlap with arXiv:2407.13431

Motion deblurring addresses the challenge of image blur caused by camera or scene movement. Event cameras provide motion information that is encoded in the asynchronous event streams. To efficiently leverage the temporal information of event streams, we employ Spiking Neural Networks (SNNs) for motion feature extraction and Artificial Neural Networks (ANNs) for color information processing. Due to the non-uniform distribution and inherent redundancy of event data, existing cross-modal feature fusion methods exhibit certain limitations. Inspired by the visual attention mechanism in the human visual system, this study introduces a bioinspired dual-drive hybrid network (BDHNet). Specifically, the Neuron Configurator Module (NCM) is designed to dynamically adjusts neuron configurations based on cross-modal features, thereby focusing the spikes in blurry regions and adapting to varying blurry scenarios dynamically. Additionally, the Region of Blurry Attention Module (RBAM) is introduced to generate a blurry mask in an unsupervised manner, effectively extracting motion clues from the event features and guiding more accurate cross-modal feature fusion. Extensive subjective and objective evaluations demonstrate that our method outperforms current state-of-the-art methods on both synthetic and real-world datasets., Comment: 11 pages, 8 figures

The existence of a quantum critical point (QCP) and fluctuations around it are believed to be important for understanding the phase diagram in unconventional superconductors such as cuprates, iron pnictides, and heavy fermion superconductors. However, the QCP is usually buried deep within the superconducting dome and is difficult to investigate. The connection between quantum critical fluctuations and superconductivity remains an outstanding problem in condensed matter. Here combining both electrical transport and Nernst experiments, we explicitly demonstrate the onset of superconductivity at an unconventional QCP in gate-tuned monolayer tungsten ditelluride (WTe2), with features incompatible with the conventional Bardeen-Cooper-Schrieffer (BCS) scenario. The results lead to a novel superconducting phase diagram that is distinguished from other known superconductors. Two distinct gate-tuned quantum phase transitions are observed at the ends of the superconducting dome. We find that quantum fluctuations around the QCP of the underdoped regime are essential for understanding how the monolayer superconductivity is established. The unconventional phase diagram we report here illustrates a previously unknown relation between superconductivity and QCP.

As is known, hybrid quadratic and subquadratic attention models in multi-head architectures have surpassed both Transformer and Linear RNN models , with these works primarily focusing on reducing KV complexity and improving efficiency. For further research on expressiveness, we introduce our series of models distilled from Qwen 2.5, based on pure native RWKV-7 attention, which aims to make RNN more expressive and demonstrates state tracking ability beyond transformers. We work with QRWK 32B based on RWKV-6 architecture, another approach that reduces the entire knowledge processing time to just 8 hours using 16 AMD MI300X GPUs while maintaining Qwen 2.5's performance. In fact, the distillation process can utilize any LLM, not just Qwen, and enables knowledge transfer from larger LLMs to smaller ones with more fewer tokens. We will explain the detailed process and share our insights on building more powerful foundation models. Please note that this is an ongoing work that will be updated continuously. The model checkpoints and source code are available at \href{https://github.com/yynil/RWKVInside}{https://github.com/yynil/RWKVInside}, \href{https://huggingface.co/RWKV-Red-Team/ARWKV-7B-Preview-0.1}{https://huggingface.co/RWKV-Red-Team/ARWKV-7B-Preview-0.1}.

In this paper, we examine the performance of randomised Euler-Maruyama (EM) method for additive time-inhomogeneous SDEs with an irregular drift. In particular, the drift is assumed to be $\alpha$-H\"older continuous in time and bounded $\beta$-H\"older continuous in space with $\alpha,\beta\in (0,1]$. The strong order of convergence of the randomised EM in $L^p$-norm is shown to be $1/2+(\alpha \wedge (\beta/2))-\epsilon$ for an arbitrary $\epsilon\in (0,1/2)$, higher than the one of standard EM, which is $\alpha \wedge (1/2+\beta/2-\epsilon)$. The proofs highly rely on the stochastic sewing lemma, where we also provide an alternative proof when handling time irregularity for a comparison.

We present the TinyLLaVA-Video, a video understanding model with parameters not exceeding 4B that processes video sequences in a simple manner, without the need for complex architectures, supporting both fps sampling and uniform frame sampling. Our model is characterized by modularity and scalability, allowing training and inference with limited computational resources and enabling users to replace components based on their needs. We validate the effectiveness of this framework through experiments, the best model achieving performance comparable to certain existing 7B models on multiple video understanding benchmarks. The code and training recipes are fully open source, with all components and training data publicly available. We hope this work can serve as a baseline for practitioners exploring small-scale multimodal models for video understanding. It is available at \url{https://github.com/ZhangXJ199/TinyLLaVA-Video}., Comment: code and training recipes are available at https://github.com/ZhangXJ199/TinyLLaVA-Video

Configuration in software systems helps to ensure efficient operation and meet diverse user needs. Yet, some, if not all, configuration options have profound implications for the system's performance. Configuration performance analysis, wherein the key is to understand (or infer) the configuration options' relations and their impacts on performance, is crucial. Two major modalities exist that serve as the source information in the analysis: either the manual or source code. However, it remains unclear what roles they play in configuration performance analysis. Much work that relies on manuals claims their benefits of information richness and naturalness; while work that trusts the source code more prefers the structural information provided therein and criticizes the timeliness of manuals. To fill such a gap, in this paper, we conduct an extensive empirical study over 10 systems, covering 1,694 options, 106,798 words in the manual, and 22,859,552 lines-of-code for investigating the usefulness of manual and code in two important tasks of configuration performance analysis, namely performance-sensitive options identification and the associated dependencies extraction. We reveal several new findings and insights, such as it is beneficial to fuse the manual and code modalities for both tasks; the current automated tools that rely on a single modality are far from being practically useful and generally remain incomparable to human analysis. All those pave the way for further advancing configuration performance analysis., Comment: accepted by ICSE 2025

Cooperative-integrated sensing and communication (C-ISAC) networks have emerged as promising solutions for communication and target sensing. However, imperfect channel state information (CSI) estimation and time synchronization (TS) errors degrade performance, affecting communication and sensing accuracy. This paper addresses these challenges {by employing} {movable antennas} (MAs) to enhance C-ISAC robustness. We analyze the impact of CSI errors on achievable rates and introduce a hybrid Cramer-Rao lower bound (HCRLB) to evaluate the effect of TS errors on target localization accuracy. Based on these models, we derive the worst-case achievable rate and sensing precision under such errors. We optimize cooperative beamforming, {base station (BS)} selection factor and MA position to minimize power consumption while ensuring accuracy. {We then propose a} constrained deep reinforcement learning (C-DRL) approach to solve this non-convex optimization problem, using a modified deep deterministic policy gradient (DDPG) algorithm with a Wolpertinger architecture for efficient training under complex constraints. {Simulation results show that the proposed method significantly improves system robustness against CSI and TS errors, where robustness mean reliable data transmission under poor channel conditions.} These findings demonstrate the potential of MA technology to reduce power consumption in imperfect CSI and TS environments.

As the size of transistors shrinks and power density increases, thermal simulation has become an indispensable part of the device design procedure. However, existing works for advanced technology transistors use simplified empirical models to calculate effective thermal conductivity in the simulations. In this work, we present a dataset of size-dependent effective thermal conductivity with electron and phonon properties extracted from ab initio computations. Absolute in-plane and cross-plane thermal conductivity data of eight semiconducting materials (Si, Ge, GaN, AlN, 4H-SiC, GaAs, InAs, BAs) and four metallic materials (Al, W, TiN, Ti) with the characteristic length ranging from 5 to 50 nanometers have been provided. Besides the absolute value, normalized effective thermal conductivity is also given, in case it needs to be used with updated bulk thermal conductivity in the future. The dataset presented in this paper are openly available at https://doi.org/10.57760/sciencedb.j00113.00154.

Artificial intelligence is reshaping creative domains, yet its co-creative processes, especially in group settings with novice users, remain under explored. To bridge this gap, we conducted a case study in a college-level course where nine undergraduate students were tasked with creating three original music tracks using AI tools over 10 weeks. The study spanned the entire creative journey from ideation to releasing these songs on Spotify. Participants leveraged AI for music and lyric production, cover art, and distribution. Our findings highlight how AI transforms creative workflows: accelerating ideation but compressing the traditional preparation stage, and requiring novices to navigate a challenging idea selection and validation phase. We also identified a new "collaging and refinement" stage, where participants creatively combined diverse AI-generated outputs into cohesive works. Furthermore, AI influenced group social dynamics and role division among human creators. Based on these insights, we propose the Human-AI Co-Creation Stage Model and the Human-AI Agency Model, offering new perspectives on collaborative co-creation with AI.

Multi-modal recommender systems (MMRS) have gained significant attention due to their ability to leverage information from various modalities to enhance recommendation quality. However, existing negative sampling techniques often struggle to effectively utilize the multi-modal data, leading to suboptimal performance. In this paper, we identify two key challenges in negative sampling for MMRS: (1) producing cohesive negative samples contrasting with positive samples and (2) maintaining a balanced influence across different modalities. To address these challenges, we propose NegGen, a novel framework that utilizes multi-modal large language models (MLLMs) to generate balanced and contrastive negative samples. We design three different prompt templates to enable NegGen to analyze and manipulate item attributes across multiple modalities, and then generate negative samples that introduce better supervision signals and ensure modality balance. Furthermore, NegGen employs a causal learning module to disentangle the effect of intervened key features and irrelevant item attributes, enabling fine-grained learning of user preferences. Extensive experiments on real-world datasets demonstrate the superior performance of NegGen compared to state-of-the-art methods in both negative sampling and multi-modal recommendation.

We study the atomic collapse phenomenon in $K$ and $K'$ valley of gapped graphene. Bound states induced by Coulomb impurity in the gap turn into atomic collapse resonances as the charge increases beyond the supercritical charge $Z_c$. $Z_c$ increases sublinear with the band gap $\Delta$. The atomic collapse resonances result in peaks in the LDOS at the same energies in $K$ and $K'$ valley, but the strong (weak) LDOS peaks in $K$ valley are weak (strong) LDOS peaks in $K'$ valley reminiscent of pseudospin polarization phenomenon. From a spatial LDOS analysis of the atomic collapse resonance states, we assign specific atomic orbitals to the atomic collapse resonances. Remarkably, the two $p$ atomic orbital atomic collapse states are no longer degenerate and splits into two having lobes in different directions in the graphene plane.

This paper introduces DeepFlow, a scalable and serverless AI platform designed to efficiently serve large language models (LLMs) at scale in cloud environments. DeepFlow addresses key challenges such as resource allocation, serving efficiency, and cold start latencies through four main design components. First, it uses a simple serverless abstraction called the request-job-task model, which helps manage AI workloads across post-training and model serving tasks. Second, it builds an in-house serving engine FlowServe using a microkernel-inspired design, NPU-centric execution, and SPMD-based parallelism to optimize LLM serving. The system also includes novel scheduling policies tailored for both PD-disaggregated and PD-colocated configurations. With optimizations like pre-warmed pods, DRAM pre-loading, and NPU-fork, DeepFlow can scale up to 64 instances in seconds. DeepFlow has been in production for over a year, operating on a large Ascend NPU cluster and providing industrystandard APIs for fine-tuning, agent serving, and model serving to our customers.

The Potts model describes interacting spins with $Q$ different components, which is a direct generalization of the Ising model ($Q=2$). Compared to the existing exact solutions in 2D, the phase transitions and critical phenomena in the 3D Potts model have been less explored. Here, we systematically investigate a quantum $(2+1)$-D Potts model with $Q=3$ using a fuzzy sphere regularization scheme. We first construct a microscopic model capable of achieving a magnetic phase transition that separates a spin $S_3$ permutationally symmetric paramagnet and a spontaneous symmetry-breaking ferromagnet. Importantly, the energy spectrum at the phase transition point exhibits an approximately conformal symmetry, implying that an underlying conformal field theory may govern this transition. Moreover, when tuning along the phase transition line in the mapped phase diagram, we find that the dimension of the subleading $S_3$ singlet operator flows and drifts around the critical value $\sim 3$, which is believed to be crucial for understanding this phase transition, although determining its precise value remains challenging due to the limitations of our finite-size calculations. These findings suggest a discontinuous transition in the 3D 3-state Potts model, characterized by pseudo-critical behavior, which we argue results from a nearby multicritical or complex fixed point.

Benchmarks are important tools for tracking the rapid advancements in large language model (LLM) capabilities. However, benchmarks are not keeping pace in difficulty: LLMs now achieve over 90\% accuracy on popular benchmarks like MMLU, limiting informed measurement of state-of-the-art LLM capabilities. In response, we introduce Humanity's Last Exam (HLE), a multi-modal benchmark at the frontier of human knowledge, designed to be the final closed-ended academic benchmark of its kind with broad subject coverage. HLE consists of 3,000 questions across dozens of subjects, including mathematics, humanities, and the natural sciences. HLE is developed globally by subject-matter experts and consists of multiple-choice and short-answer questions suitable for automated grading. Each question has a known solution that is unambiguous and easily verifiable, but cannot be quickly answered via internet retrieval. State-of-the-art LLMs demonstrate low accuracy and calibration on HLE, highlighting a significant gap between current LLM capabilities and the expert human frontier on closed-ended academic questions. To inform research and policymaking upon a clear understanding of model capabilities, we publicly release HLE at https://lastexam.ai., Comment: 26 pages, 6 figures