Sheng Li

Yun Luo, Futing Wang, Qianjia Cheng et al.

The transition from symbolic manipulation to science-grade reasoning represents a pivotal frontier for Large Language Models (LLMs), with physics serving as the critical test anchor for binding abstract logic to physical reality. Physics demands that a model maintain physical consistency with the laws governing the universe, a task that fundamentally requires multimodal perception to ground abstract logic in reality. At the Olympiad level, diagrams are often constitutive rather than illustrative, containing essential constraints, such as boundary conditions and spatial symmetries, that are absent from the text. To bridge this visual-logical gap, we introduce P1-VL, a family of open-source vision-language models engineered for advanced scientific reasoning. Our method harmonizes Curriculum Reinforcement Learning, which employs progressive difficulty expansion to stabilize post-training, with Agentic Augmentation, enabling iterative self-verification at inference. Evaluated on HiPhO, a rigorous benchmark of 13 exams from 2024-2025, our flagship P1-VL-235B-A22B becomes the first open-source Vision-Language Model (VLM) to secure 12 gold medals and achieves the state-of-the-art performance in the open-source models. Our agent-augmented system achieves the No.2 overall rank globally, trailing only Gemini-3-Pro. Beyond physics, P1-VL demonstrates remarkable scientific reasoning capacity and generalizability, establishing significant leads over base models in STEM benchmarks. By open-sourcing P1-VL, we provide a foundational step toward general-purpose physical intelligence to better align visual perceptions with abstract physical laws for machine scientific discovery.

Yuxin Zuo, Kaiyan Zhang, Li Sheng et al. · pku, tsinghua

This paper investigates Reinforcement Learning (RL) on data without explicit labels for reasoning tasks in Large Language Models (LLMs). The core challenge of the problem is reward estimation during inference while not having access to ground-truth information. While this setting appears elusive, we find that common practices in Test-Time Scaling (TTS), such as majority voting, yield surprisingly effective rewards suitable for driving RL training. In this work, we introduce Test-Time Reinforcement Learning (TTRL), a novel method for training LLMs using RL on unlabeled data. TTRL enables self-evolution of LLMs by utilizing the priors in the pre-trained models. Our experiments demonstrate that TTRL consistently improves performance across a variety of tasks and models. Notably, TTRL boosts the pass@1 performance of Qwen-2.5-Math-7B by approximately 211% on the AIME 2024 with only unlabeled test data. Furthermore, although TTRL is only supervised by the maj@n metric, TTRL has demonstrated performance to consistently surpass the upper limit of the initial model maj@n, and approach the performance of models trained directly on test data with ground-truth labels. Our experimental findings validate the general effectiveness of TTRL across various tasks and highlight TTRL's potential for broader tasks and domains. GitHub: https://github.com/PRIME-RL/TTRL

Miaomiao Wang, Guang Hua, Sheng Li et al.

Virtual faces are crucial content in the metaverse. Recently, attempts have been made to generate virtual faces for privacy protection. Nevertheless, these virtual faces either permanently remove the identifiable information or map the original identity into a virtual one, which loses the original identity forever. In this study, we first attempt to address the conflict between privacy and identifiability in virtual faces, where a key-driven face anonymization and authentication recognition (KFAAR) framework is proposed. Concretely, the KFAAR framework consists of a head posture-preserving virtual face generation (HPVFG) module and a key-controllable virtual face authentication (KVFA) module. The HPVFG module uses a user key to project the latent vector of the original face into a virtual one. Then it maps the virtual vectors to obtain an extended encoding, based on which the virtual face is generated. By simultaneously adding a head posture and facial expression correction module, the virtual face has the same head posture and facial expression as the original face. During the authentication, we propose a KVFA module to directly recognize the virtual faces using the correct user key, which can obtain the original identity without exposing the original face image. We also propose a multi-task learning objective to train HPVFG and KVFA. Extensive experiments demonstrate the advantages of the proposed HPVFG and KVFA modules, which effectively achieve both facial anonymity and identifiability.