Wenjie Wang

Wenjie Wang, Yue Huang, Zhengqing Yuan et al.

As large language models (LLMs) are increasingly deployed in real-world applications, alignment is no longer governed by a single universal notion of safety or helpfulness, but instead by provider- or application-specific model specifications. These specifications are typically long, structured, and frequently updated, yet existing alignment pipelines lack a systematic mechanism to operationalize them as training signals. In this paper, we propose specification-grounded alignment, a new alignment paradigm that treats provider-authored model specifications as the primary alignment target rather than abstract principles or static benchmarks. To instantiate this paradigm, we introduce SpecAlign, a framework that synthesizes alignment data directly from specification documents. SpecAlign combines structured rule annotation, controllable specification instantiation, and multi-agent adversarial data synthesis to generate fine-grained, boundary-aware preference pairs that capture both compliant behaviors and meaningful specification violations. Experiments across multiple model specifications and backbone models demonstrate that training with SpecAlign consistently improves rule compliance while preserving general capabilities and avoiding over-conservative behavior. These results suggest that grounding alignment in explicit model specifications enables rapid, precise, and scalable adaptation of LLM behavior to evolving policy requirements.

Wenjie Wang, Yue Huang, Zipeng Ling et al.

Large language models (LLMs) are increasingly deployed as UX judges that inspect interfaces, diagnose usability problems, and propose repairs. Yet no controlled benchmark measures whether the resulting critiques are reliable and actionable across heterogeneous product surfaces. We introduce UXBench, a benchmark for evaluating LLMs as interaction-grounded UX judges. UXBench comprises local-first runnable web fixtures spanning ten product-surface families, paired with coverage-gated browser exploration that forces models to collect interaction evidence before reporting. Each judge model produces a structured UX report over seven rubric dimensions; report quality is measured by whether a fixed downstream repair agent can improve the interface based on the critique. We evaluate eight frontier models under both an automated repair-lift protocol and a blind human validation study. Results show that UX judging is neither saturated nor one dimensional: models differ meaningfully in report actionability, exhibit distinct rubric-level repair signatures, vary in fixture-level reliability, and trade leadership across surface categories

Wenjie Wang, Hao Chen, Ran Shu et al.

Accurate and adaptive modeling of vehicle dynamics is paramount for the safety of autonomous driving systems, particularly under extreme maneuvers and time-varying parameters. While Deep Koopman operator theory offers a promising global linearization framework, its online application faces a theoretical bottleneck: the high-dimensional lifted state space inherently induces a rank-deficient problem, rendering traditional recursive least squares based updates numerically unstable. To address this, we propose a novel tire-force-driven modeling framework with guaranteed online stability. First, an offline Deep Koopman model is constructed by embedding 7DOF dynamic equilibrium constraints into the learning objective, ensuring the structural fidelity and physical interpretability of the lifted manifold. Second, we theoretically reformulate the operator update in the rank-deficient space as a minimum-norm solution problem. A Physics-Informed Variable Step-Size Normalized Least Mean Squares (PI-VSS-NLMS) algorithm is proposed, which leverages the projection property of NLMS to act as a stable pseudo-inverse solver while incorporating an anchoring mechanism to suppress parameter drift. Extensive simulations on CarSim and Hardware-in-the-Loop validation on dSPACE MicroAutobox III confirm the superiority of the proposed algorithm. It achieves robust prediction accuracy under unseen excitations while guaranteeing real-time feasibility with an average execution time of 0.421 ms, thus bridging the gap between theoretical models and practical deployment.