Search arXiv⌕ Search

arXiv subjects

Bo Li

Publications and source records attributed to Bo Li.

At least 109 records · Page 6Linked to original sources
arXiv2026

VIPER: Visual In-Context Physics Reasoning for Physically Plausible Video Generation

Modern video generation models can synthesize visually compelling and temporally coherent clips, yet controlling their physical behavior remains difficult with standard text and image conditions. The core challenge is a conditioning bottleneck: material response, contact interaction, deformation, and motion trajectory are continuous and relational physical cues that are hard to specify exhaustively in language but can be demonstrated naturally by video. We propose VIPER, a Visual In-Context Physics Reasoning framework for reference-guided image-to-video generation. Given a target image, a brief target prompt, and a reference video, VIPER treats the reference as a dense visual demonstration of the desired physical process rather than an appearance template. It uses a Multimodal Large Language Model (MLLM) to extract reference-derived physical cues and guide a pretrained image-to-video generator through a hierarchical training strategy, enabling physical behavior transfer while preserving the visual prior of the base generator. To support this setting, we construct VIPER-19K, a curated dataset with material, trajectory, and physical-impact annotations, together with filtered reference-target pairs. Experiments on an unseen validation set show that VIPER achieves stronger reference-video physical similarity and higher human preference than representative video generation and video-as-prompt baselines, while maintaining competitive general video quality. Qualitative results further demonstrate that VIPER can transfer reference-derived physical behavior to new target scenes without requiring carefully engineered prompts.

cs.CV↗
arXiv2026

MineValiCoder: Reliable Code Generation with Test Case Quality Mining and Bipartite Graph-Based Mutual Validation

Large Language Model (LLM)-based Test-Driven Development (TDD) has advanced automated code generation. However, existing approaches depend heavily on human-crafted test cases and cannot operate effectively when only natural-language requirements are available. Although recent work enables automatic test generation, it often overlooks the inherent stochasticity of LLMs, leading to two key defects: faulty tests generate misleading feedback that distorts code optimization, while mixed-quality test cases produce conflicting evaluation signals that hinder reliable code selection. To address these challenges, we propose MineValiCoder, a collaborative closed-loop TDD framework based on the mutual reinforcement of test-case quality and code quality. MineValiCoder comprises three modules. The Test Case Quality Mining (TCQM) module filters faulty test cases through self-validation, providing reliable optimization supervision. The Parallel TDD Refinement module iteratively optimizes code and generates diverse high-quality code candidates using validated test-case feedback. The Bipartite Graph-Based Code-Test Mutual Validation (BiCoTeV) module dynamically models code-test interactions and performs mutual validation scoring for stable and reliable optimal-code selection. Extensive evaluations across four LLMs and mainstream benchmarks show that MineValiCoder significantly outperforms state-of-the-art methods. Specifically, it achieves Pass@1 scores of 96.34% on HumanEval, 87.40% on MBPP, 64.00% on APPS, and 51.33% on LiveCodeBench. These results demonstrate the effectiveness of MineValiCoder in mitigating LLM stochasticity and improving the reliability of automated code generation.

cs.SE↗
arXiv2026

MagicMakeup: A Region-Controllable Diffusion Transformer for High-Fidelity Makeup-Transfer

Makeup-transfer applies the reference makeup to the source face while preserving the source identity. Despite advances in full-face editing by diffusion-based methods, strong regional controllability, makeup fidelity, and identity preservation remain challenging. The reasons are (i) pixel-to-attention misalignment that causes spillover into non-target areas and weakens regional control; (ii) unclear transfer/preservation concept separation under two-image conditioning, leading to coupling between makeup attributes and identity; and (iii) the lack of a high-resolution dataset that is identity-consistent and region-labeled for fine-grained supervision. In this paper, we propose MagicMakeup, a diffusion transformer-based framework for region-controllable and high-fidelity makeup transfer, built on spatial constraints and concept disentanglement. To enable precise region-specific editing while preserving identity, we propose Token-Aligned Region Gating, which aligns pixel masks with attention and applies region-specific logit gating. To clarify the concepts of transfer and preservation, we further introduce Cross-Modal Perception Guidance, which aligns text and image features to enhance cross-modal concept perception. We also design a pipeline for the generation of 1024 x 1024 data pairs through region-specific makeup removal and establish a unified benchmark in synthetic and real settings. Extensive quantitative and qualitative experiments show that MagicMakeup improves regional controllability, makeup fidelity, and identity preservation, with strong robustness across styles, races, and poses.

cs.CV↗
arXiv2026

Do Pathology Vision-Language Models Truly See Pathology?

Pathology vision-language models (VLMs) have recently progressed rapidly and are commonly evaluated by answer accuracy on pathology VQA benchmarks. However, we dig into current evaluations and identify three overlooked issues: 1) Visual evidence is not always necessary. For instance, Gemini-3-Pro achieves 53.5% average accuracy across 5 VQA benchmarks without any visual input. 2) Domain training can improve accuracy without proportional gains in visual binding. Compared with Qwen2.5-VL-7B, Patho-R1-7B exhibits a 5.8-point lower multimodal gain and a 3.7-point lower attention IoU. 3) Entity-level attention is diffuse and weakly query-specific. On PathVG, attention maps remain highly correlated across different entity queries. These issues can lead to substantial misjudgments of pathology VLMs' actual multimodal capabilities. To this end, we present PathBind, a benchmark comprising 2,600 samples: PathBind-VQA with 1,500 questions across six dimensions, PathBind-PTA with 600 questions from a private pathology teaching atlas, and PathBind-Grounding with 500 expert-curated region-level samples. Each component undergoes task-specific automated filtering and expert review to reduce textual shortcuts and improve entity-region correspondence. We evaluate 18 representative VLMs on VQA samples of PathBind and five existing pathology VQA benchmarks, and further evaluate 10 VLMs on PathBind-Grounding and PathVG. Results show that current pathology VLMs still exhibit a substantial gap between answer-side performance and visual-semantic binding.

cs.CV↗
arXiv2026

Beyond Episodic Evaluation: Memory Architectural Bottlenecks in Sequential Embodied Question Answering

Embodied question answering (EQA) is traditionally evaluated under an episodic formulation, where agents solve each task independently and reset internal state between episodes. However, real-world robots operate continuously and must accumulate, retain, and selectively reuse information acquired from prior interactions. Despite this practical requirement, the architectural mechanisms needed to support sequential memory in EQA remain underexplored. In this work, we investigate how different memory architectures behave when EQA agents are evaluated sequentially, with multiple questions answered in the same scene while memory is carried forward across queries. We find that simply preserving existing memory is often insufficient. Agents that retain only traversability information, such as 2D occupancy maps, remember where the robot has explored but not the visual-semantic evidence needed for later questions. Agents trained on short-horizon episodic data face a different challenge: when exposed to continuous, multi-query histories, their inherited context suffers from severe temporal mismatch, rather than forming a reusable scene representation. To overcome this architectural bottleneck, we highlight the necessity of structured, spatially grounded memory: architectures that map persistent visual observations onto metric 3D geometry preserve visual-semantic evidence in a coherent scene representation. Extensive experiments in simulated environments reveal that this form of memory breaks the accuracy-efficiency tradeoff in sequential settings, simultaneously achieving higher answer accuracy and lower navigation costs. We further validate these findings on a real-world mobile robot, demonstrating that spatially grounded visual memory is critical for enabling continuous, intelligent operation in physical environments.

cs.RO↗
arXiv2026

Twin Agent: Context Residual Compression for Privilege Separated Agents

Large language model (LLM) agents are vulnerable to security risks, such as prompt injection attacks from untrusted context that manipulate downstream reasoning and tool use. Existing secure-by-design approaches mitigate this risk by separating untrusted observations from privileged execution and careful control of information flow, but often degrade utility and require extensive task-specific engineering. We thus propose Twin Agent, a general privilege separation design pattern inspired by residual coding in the agent context. Twin Agent consists of two nearly symmetric agents: an Explore Agent that inspects untrusted information and a Safe Agent that executes privileged actions. The Explore Agent is conditioned on the Safe Agent's current context and communicates only compact hints to the Safe Agent about the next action to take. This design reduces the information needed to preserve task utility and thus achieves a better security--utility tradeoff, which we empirically verify by measuring how utility and attack success change as the length of hints varies. We evaluate Twin Agent on long-horizon software engineering tasks with SWE-bench Lite and on heterogeneous multi-tool interaction tasks with AgentDojo and DecodingTrust-Agent. Across both benchmarks, Twin Agent preserves high task utility while preventing prompt injection attacks, outperforming both undefended agents and privilege separation baselines.

cs.CR↗
arXiv2026

Mediator: Memory-efficient LLM Merging with Less Parameter Conflicts and Uncertainty Based Routing

Model merging aggregates Large Language Models (LLMs) finetuned on different tasks into a stronger one. However, parameter conflicts between models leads to performance degradation in averaging. While model routing addresses this issue by selecting individual models during inference, it imposes excessive storage and compute costs, and fails to leverage the common knowledge from different models. In this work, we observe that different layers exhibit varying levels of parameter conflicts. Building on this insight, we average layers with minimal parameter conflicts and use a novel task-level expert routing for layers with significant conflicts. To further reduce storage costs, inspired by task arithmetic sparsity, we decouple multiple fine-tuned experts into a dense expert and several sparse experts. Considering the out-of-distribution samples, we select and merge appropriate experts based on the task uncertainty of the input data. We conduct extensive experiments on both LLaMA and Qwen with varying parameter scales, and evaluate on real-world reasoning tasks. Results demonstrate that our method consistently achieves significant performance improvements while requiring less system cost.

cs.LG↗
arXiv2026

Imaginarity as a Resource within Quantum Coherence: Geometric Decomposition and Operational Conversion

We establish a rigorous framework that identifies imaginarity as a fundamental resource inherent in quantum coherence. By means of a geometric decomposition, we partition coherence into distinct imaginarity and residual components, thereby revealing a universal hierarchical relationship among these resources. For bipartite systems, this decomposition provides explicit bounds on the extent to which nonlocal correlations and imaginarity limit local coherence generation. Furthermore, we devise an explicit operational protocol that converts imaginarity into usable coherence, demonstrating the direct interconvertibility of these resources under physically admissible operations. The dynamical evolution under diagonal Hamiltonians is fully characterized, showing that while total coherence is conserved, imaginarity and residual coherence exhibit complementary oscillations. Our results provide a rigorous geometric and operational characterization of imaginarity as a fundamental constituent of quantum coherence, offering concrete insights for resource management in distributed quantum technologies. The geometric framework and theoretical bounds established herein are fully general, while the explicit conversion protocol and dynamical analysis serve as a compelling proof-of-principle demonstration in qubit systems.

quant-ph↗
arXiv2026

A Unified Credit Expansion Theory on Housing Cycle: Causal Evidence for Within- and Cross-Metro Patterns in the Prior, Boom, Bust, and Recovery Periods

During the 1999-2019 U.S. housing cycle, three empirical facts present a puzzle: in the boom period, the correlation between income growth and mortgage growth is (1) negative across ZIP codes within a metropolitan area, but (2) positive across metropolitan areas, and (3) the metropolitan areas that experience the worst bust also show the strongest recovery. I develop a unified credit expansion theory that explains both within- and cross-metro patterns in the prior, boom, bust, and recovery periods (including the three facts above) and generates new testable implications of ``double differences" (cross ZIP codes and cross metros) for the four periods. Following the idea of ``Economic Base Theory", I construct local economic exposure to net export growth as the driving force of local economy and credit expansion. For the identification strategy, I use a new instrumental variable approach from the International trade literature for the following empirical results. First, I show that high-net-export-growth metros experience a stronger boom-bust-recovery housing cycle due to credit expansion in private-label mortgages (PLMs), rather than in government-sponsored enterprise mortgages (GSEMs), because only the former can legally respond to local economic conditions. Second, for the ``double differences", I define a low-minus-high (LMH) factor as the private-label mortgage (and house price) growth in low-income ZIP codes minus that in high-income ZIP codes within the same metropolitan area. I show that this low-minus-high factor (as a measure of credit expansion) in the high-net-export-growth metros is more positive during the boom period, more negative during the bust period, and slightly more positive in the recovery period than in the low-net-export-growth metros. Lastly, I employ five tests to demonstrate that ``speculation" is unlikely to play a dominant role in this housing cycle.

q-fin.GN↗
arXiv2026

Beyond Looking Up, Try Looking Around: Harmonizing Global Structure and Local Consistency in Optimal Transport for Short Text Clustering

Pseudo-labeling based on Optimal Transport (OT) has become an effective mechanism for enhancing short text clustering. Existing OT methods are short in modeling semantic consistencies between samples, which may assign different pseudo-labels to semantically similar samples. These erroneous pseudo-labels can cause the model to produce inferior clusters. This paper proposes a novel short text clustering framework, which remedies the neglect of semantic consistency in existing OT methods, generating reliable pseudo-labels to facilitate clustering. Specifically, the proposed approach first designs an instance-level attention mechanism to capture semantic relationships between samples, which are then integrated into the OT formulation to endow the transport process with neighborhood semantic awareness. By solving the proposed OT formulation, reliable pseudo-labels are obtained that simultaneously account for sample-to-sample semantic consistency and sample-to-cluster global structure information. These pseudo-labels are then used as supervisory signals to guide the model to achieve accurate clustering. Extensive experiments demonstrate that the proposed approach outperforms state-of-the-art methods. The code is available at: \href{https://github.com/YZH0905/CAOT-STC}{https://github.com/YZH0905/CAOT-STC}.

stat.ML↗
arXiv2026

H3D: Benchmarking Unsupervised Text Hashing for Fine-Grained Document Deduplication

Document hashing provides compact representations for efficient similarity search and document deduplication, but existing studies rarely compare hashing pipelines under a unified protocol for fine-grained scientific documents. H3D is an unsupervised text hashing benchmark for fine-grained document deduplication. It evaluates representative unsupervised non-learning hashing approaches (MinHash, SimHash, Winnowing, FuzzyHash, FlyHash) together with semantic-sensitive methods built from frozen BGE embeddings and two quantization strategies (BGE-BIHash and BGE-LSHash). The non-learning methods generate hash fingerprints through manually designed mathematical rules without training or labeled similarity pairs, which distinguishes them from neural semantic hashing models. We benchmark all methods on CSFCube and RELISH, two datasets that provide complementary evaluation settings: facet-level analysis for scientific-document similarity and larger-scale split-level evaluation for biomedical similarity search. H3D jointly reports ranking quality (MAP, NDCG@20), efficiency, and robustness under controlled text compression. The results show a consistent trade-off: lexical and structural fingerprints are competitive for near-duplicate matching, while semantic-sensitive representations better preserve similarity under content rewriting, at higher computational cost. We further analyze when different similarity measures become rank-equivalent for specific hash representations, improving the interpretability and reproducibility of method comparisons.

cs.IR↗
arXiv2026

Charge radii of calcium isotopes within relativistic configuration-interaction density functional theory

The charge radii of calcium isotopes are investigated within the framework of relativistic configuration-interaction density functional (ReCD) theory. The ReCD theory microscopically incorporates beyond-mean-field correlations through rotational symmetry restoration and configuration mixing among quasiparticle excited states, and treats even-even and odd-$A$ isotopes on the same footing. It is found that beyond-mean-field correlations significantly soften the potential energy surfaces of calcium isotopes and shift the energy minima from nearly spherical mean-field solutions to deformed shapes. The quadrupole deformation parameters predicted by the ReCD theory show much better agreement with the available experimental data than the mean-field results, supporting the reliability of the calculated potential energy surfaces and highlighting the important role of beyond-mean-field correlations. Owing to the sensitive dependence of charge radii on nuclear deformation, the charge radii obtained within the ReCD framework are generally larger than the mean-field predictions. The nearly identical charge radii of $^{40}\mathrm{Ca}$ and $^{48}\mathrm{Ca}$, as well as the unexpectedly large charge radius of $^{52}\mathrm{Ca}$, are well reproduced. Compared with the mean-field calculations, the description of the odd-even staggering is improved, especially for the enhanced charge radii of $^{42}\mathrm{Ca}$ and $^{44}\mathrm{Ca}$. It is also worth noting that secondary local minima appear in the ReCD-based potential energy surfaces of the odd-$A$ calcium isotopes $^{41,43,47}\mathrm{Ca}$. The present results suggest that shape mixing between different local minima, which is not fully included in the present calculation, may further improve the description of the pronounced odd-even staggering observed in calcium isotopes.

nucl-th↗
arXiv2026

Involution game with migration and spatial heterogeneity of social resources

Involution -- a phenomenon of excessive competition with diminishing returns -- has become a pressing socio-economic concern in contemporary China, prompting both academic inquiry and policy interventions. This paper proposes an evolutionary game model of involution that incorporates agent migration and spatial heterogeneity in resource distribution. The model captures realistic features such as effort-based resource allocation, local interactions on a lattice, and mobility driven by payoff comparisons. We explore how varying conditions of migration and resource allocation influence the dynamics of involution. The key findings from our simulations are as follows: when total resources are held constant, similar resource levels across different regions tend to suppress involution, whereas a large disparity between regions promotes it. Furthermore, increasing the total amount of resources exacerbates involution. In addition, the probability of migration does not significantly affect the final evolutionary outcome. We further identify threshold effects in the effort ratio and utility multiplier, revealing critical conditions under which involution emerges or subsides. To further elucidate these simulation results, we conduct a theoretical analysis using mean-field theory, which provides analytical expressions for the equilibria and stability conditions. The theoretical predictions are in excellent qualitative agreement with simulation outcomes. Finally, we discuss real-world counterparts of the model, including competition among food delivery riders and between stores offering similar services.

physics.soc-ph↗
arXiv2026

Constant Weighted Maximin Share Approximations for Chores

We study the fair allocation of indivisible chores among agents with asymmetric weights. Among the various fairness notions, weighted maximin share (WMMS) stands out as particularly compelling. Despite its appeal, the existence of a constant-factor approximation for WMMS has remained an important open problem in weighted fair division [Aziz et al., 2022, Suksompong, 2025]. Prior to our work, the best known approximation ratio was $O(\log n)$, where $n$ is the number of agents. In this paper, we make significant progress by presenting the first constant-factor approximation algorithm for WMMS. Our main contributions are as follows: [1] We design the first algorithm that guarantees a 12-approximate WMMS allocation, substantially improving upon the previous $O(\log n)$ upper bound. Our approach introduces a novel analytical framework based on canonical instance reductions, agent delegation, and proxy cost functions to effectively bound agents' costs. Additionally, we provide a polynomial-time implementation for any approximate WMMS algorithm, incurring a factor of 2 loss in the approximation ratio. [2] We present an improved worst-case lower bound, showing that no algorithm can achieve better than 2-approximate WMMS, thereby strengthening the previous best lower bound of 1.366. We further construct a general hard instance, which provides lower bounds for an arbitrary number of agents. [3] Beyond worst-case bounds, we precisely characterize the optimal approximation ratio curve for every possible weight distribution in the two-agent case. Notably, our results imply that a WMMS allocation may not exist for any two agents with different weights, in sharp contrast to the symmetric case where an MMS allocation always exists.

cs.GT↗
arXiv2026

PolicyShiftGuard: Benchmarking and Improving Policy-Adaptive Image Guardrails

Image guardrails are typically trained and evaluated under a fixed safety policy, implicitly treating safety as an intrinsic property of an image. Real deployments are different: the same image may be allowed in one product, restricted in another, and newly disallowed when a policy boundary changes. We study policy-adaptive image guardrailing, where a model must decide whether an image violates the currently supplied policy and generalize to held-out policy definitions. We introduce PolicyShiftBench, a comprehensive benchmark with 2,000 policy-discriminative instances over 265 images, where each image is paired with 7.55 policy-conditioned prompts on average to test whether models adapt to the active policy rather than relying on image-level safety priors. We then propose PolicyShiftGuard, a compact policy-conditioned guardrail trained with a two-stage training recipe that combines Randomized Policy SFT (RP-SFT) with Boundary-Pair Policy Adaptation (BP-Adapt). BP-Adapt trains matched prompts for the same image and risk category using standard label supervision and a pairwise comparison loss that separates blocking policies from passing policies. Experiments show that existing VLMs and specialized guardrails remain brittle under policy shifts, while PolicyShiftGuard substantially improves policy-sensitive performance. The 7B model achieves SOTA performance of 76.9 Avg. F1 and 72.1 Avg. PSS on PolicyShiftBench, transfers well to UnSafeBench and SafeEditBench, and improves the latency-performance trade-off with a concise output format. Ablations confirm that matched pass/block boundary pairs are essential for stable policy adaptation.

cs.CV↗
arXiv2026

Interfacial Noncollinear Filtering of Spin Hall Currents

Spin Hall currents generated in nonmagnetic materials are conventionally regarded as bulk responses whose polarization is fixed by crystal symmetry. This view has motivated the search for intrinsically low-symmetry spin sources when unconventional spin polarizations are required. Here we point out that, in realistic heterostructures, the device-relevant quantity is not the fully symmetry-averaged bulk spin Hall current, but the emitted spin current transmitted across the interface. We therefore establish emitted spin currents as bulk-interface hybrid responses and propose interfacial noncollinear filtering as a mechanism to bypass the bulk-symmetry constraint. A low-symmetry interfacial spin-orbit field, generally noncollinear with the momentum-resolved spin polarization of the incident spin Hall current, imposes spin-dependent transmission and converts hidden momentum-resolved spin-polarization components into an observable unconventional emitted spin current. Using both a rotationally symmetric minimal model and a realistic high-symmetry Dirac-semimetal model, we show that conventional spin Hall sources can emit sizable out-of-plane spin currents when their hidden bulk spin Hall textures are selectively transmitted by the interfacial spin-orbit field. Our results reveal that spin-current polarization emerges from the cooperative action of bulk and interfacial responses, providing a strategy for reprogramming spin-current polarization in high-efficiency, CMOS-compatible spin Hall materials without relying on intrinsically low-symmetry bulk crystals or external symmetry-breaking schemes.

cond-mat.mes-hall↗
arXiv2026

Physics-Informed Feature Engineering 1D-CNN for Multilayer Cloud Detection from Geostationary Satellites

Multilayer cloud detection from active--passive observation is vital for numerical weather prediction. In this study, channel selections derived from threshold-based algorithms are embedded as feature-engineering priors into a 1D-CNN, and machine learning (ML) is used to learn latent physical relationships to simplify physical retrievals for operational deployment. The results show that the 1D-CNN achieves a multilayer-cloud probability of detection ($\mathrm{POD}{\mathrm{mul}}$) of 0.620 and a false alarm rate ($\mathrm{FAR}{\mathrm{mul}}$) of 0.240, outperforming the conventional threshold algorithm ($\mathrm{POD}{\mathrm{mul}} = 0.558$, $\mathrm{FAR}{\mathrm{mul}} = 0.369$). These results demonstrate that prior physical knowledge derived from radiative transfer theory can serve as an effective feature-engineering prior. Further experiments show that ML-revealed physical mechanisms can also enhance traditional algorithms. Replacing AGRI channel 12 (C12, centered at $10.8~μ\mathrm{m}$) with channel 13 (C13, centered at $12.0~μ\mathrm{m}$) increased $\mathrm{POD}{\mathrm{mul}}$ from 0.558 to 0.609 without materially affecting $\mathrm{FAR}{\mathrm{mul}}$. However, for AHI, substituting the $11.2~μ\mathrm{m}$ channel with the $12.3~μ\mathrm{m}$ channel yielded negligible improvement. In addition to spectral response function (SRF) mismatches, a primary contributing factor is the channels' on-orbit radiometric stability. Hence, physics-informed machine-learning methods appear promising for advancing remote-sensing AI, while sensor-specific characteristics must be considered during operational transfer.

eess.SP↗
arXiv2026

AIR-BENCH Live: An Evolving Safety Benchmark for Foundation Models

Foundation-model safety benchmarks capture the AI risks of their time of publication: as models improve and governments pass new AI-safety legislation, their risk taxonomies become incomprehensive and their attack prompts become ineffective. We present AIR-BENCH Live, a self-evolving successor to AIR-BENCH 2024. An automated update pipeline monitors government regulation and classifies new policies against the current four-tier risk taxonomy, either matching them to existing categories or proposing new granular categories. Then, a multi-agent, persona-driven prompt generation algorithm generates realistic, multilingual prompts with minimal human review, leaving room for improvement with modern jail breaking techniques. This algorithm is used to overhaul legacy prompts and generate prompts for new categories. In our current version, the pipeline has expanded the benchmark from 314 to 335 granular risks, with the 21 new categories drawing from 31 truly novel policy clauses across seven jurisdictions. Evaluating 14 recent models, we find a wide safety spread (from 0.17 to 0.89 among the models judged on their own behavior), that the modernized prompts are on average 0.06 points harder than the 2024 set, with the largest drops concentrated among the most compliant models, and that most models are modestly less safe on non-English prompts. By continuously absorbing new regulation and regenerating prompts, AIR-BENCH Live is designed to evolve alongside a fast-moving field.

cs.AI↗