The misuse of large language models (LLMs) requires precise detection of synthetic text.Existing works mainly follow binary or ternary classification settings, which can only distinguish pure human/LLM text or collaborative text at best.This remains insufficient for the nuanced regulation, as the LLM-polished human text and humanized LLM text often trigger different policy consequences.In this paper, we explore fine-grained LLM-generated text detection under a rigorous four-class setting.To handle such complexities, we propose RACE (Rhetorical Analysis for Creator-Editor Modeling), a fine-grained detection method that characterizes the distinct signatures of creator and editor.Specifically, RACE utilizes Rhetorical Structure Theory (RST) to construct a logic graph for the creator's foundation while extracting Elementary Discourse Unit (EDU)-level features for the editor's style.Experiments show that RACE outperforms 12 baselines in identifying fine-grained types with low false alarms, offering a policy-aligned solution for LLM regulation.

Generative artificial intelligence and large model agents are revolutionizing the higher education, influencing everything from talent development frameworks and teaching methodologies to knowledge acquisition processes and research paradigms. Meanwhile, in the context of digital transformation and technological innovation, data as a production factor and emerging productive forces are reshaping the requirements and demands for cultivating high-level interdisciplinary engineering talents. This paper draws on the joint educational achievements between Beijing University of Posts and Telecommunications and Queen Mary University of London, focusing on the coconstruction and sharing of experimental resources, as well as innovation-driven entrepreneurship education. It introduces a digital-intelligence-powered educational platform aimed at fostering internationally-minded, innovative, and outstanding talents. The platform's core philosophy, development strategy, functional modules, and technical framework are detailed. The wide recognition and interest among stakeholders further validate its potential to support the development of a cross-disciplinary, cross-professional, and cross-national ecosystem for innovation and entrepreneurship education. As a key outcome of the 20th Anniversary Development Conference of Joint Education of our two universities, I-LEAD is a platform designed to cultivate students' comprehensive innovative capabilities. Leveraging large language models (LLMs) and multi-agent technology, we have developed BUPT iMentor, an intelligent agent for innovation and entrepreneurship guidance; and BUPT EnPower, a cultivation assistant for personalized longlife learning. By LLMs with a robust knowledge based augmented generation and fine tuning, this tool effectively addresses common student challenges during innovation & entrepreneurship projects, such as idea generation and validation, access to relevant learning resources. I-LEAD provides comprehensive support, including customized course creation, problem-solving guidance, real-time interactive Q&A, and learning progress monitoring. This empowers students to independently plan their learning journeys and holistically enhance their academic and innovative skills. The effectiveness and practicality of the platform have been validated through a questionnaire survey. Therefore, we are extensively gathering feedback and continuously optimizing the platform's services. The teacher-student collaborative learning represents the future of higher education. This student-centered education system provides a platform for that.

Though safety alignment has been applied to most large language models (LLMs), LLM service providers generally deploy a subsequent moderation as the external safety guardrail in real-world products. Existing moderators mainly practice a conventional full detection, which determines the harmfulness based on the complete LLM output, causing high service latency. Recent works pay more attention to partial detection where moderators oversee the generation midway and early stop the output if harmfulness is detected, but they directly apply moderators trained with the full detection paradigm to incomplete outputs, introducing a training-inference gap that lowers the performance. In this paper, we explore how to form a data-and-model solution that natively supports partial detection. For the data, we construct **FineHarm**, a dataset consisting of 29K prompt-response pairs with fine-grained token-level annotations to provide reasonable supervision for token-level training. Then, we propose the **Streaming Content Monitor (SCM)**, which is trained with dual supervision of response- and token-level labels and can follow the output stream of LLM to make a timely judgment of harmfulness. Experiments show that SCM gains 0.95+ in macro F1 score that is comparable to full-detection, by only seeing the first 18% of tokens in responses on average. Moreover, the SCM can serve as a pseudo-harmfulness annotator for improving safety alignment and lead to a higher harmlessness score than DPO.