'I have to do it': Why one of the world's most brilliant AI scientists left the US for China In 2020, after spending half his life in the US, Song-Chun Zhu took a one-way ticket to China. By the time Song-Chun Zhu was six years old, he had encountered death more times than he could count. This was the early 1970s, the waning years of the Cultural Revolution, and his father ran a village supply store in rural China . There was little to do beyond till the fields and study Mao Zedong at home, and so the shop became a refuge where people could rest, recharge and share tales. Zhu grew up in that shop, absorbing a lifetime's worth of tragedies: a family friend lost in a car crash, a relative from an untreated illness, stories of suicide or starvation. "That was really tough," Zhu recalled recently. The young Zhu became obsessed with what people left behind after they died. One day, he came across a book that contained his family genealogy. When he asked the bookkeeper why it included his ancestors' dates of birth and death but nothing about their lives, the man told him matter of factly that they were peasants, so there was nothing worth recording. He resolved that his fate would be different. Today, at 56, Zhu is one of the world's leading authorities in artificial intelligence. In 1992, he left China for the US to pursue a PhD in computer science at Harvard. Later, at University of California, Los Angeles (UCLA), he led one of the most prolific AI research centres in the world, won numerous major awards, and attracted prestigious research grants from the Pentagon and the National Science Foundation. He was celebrated for his pioneering research into how machines can spot patterns in data, which helped lay the groundwork for modern AI systems such as ChatGPT and DeepSeek. He and his wife, and their two US-born daughters, lived in a hilltop home on Los Angeles's Mulholland Drive. He thought he would never leave. But in August 2020, after 28 years in the US, Zhu astonished his colleagues and friends by suddenly moving back to China, where he took up professorships at two top Beijing universities and a directorship in a state-sponsored AI institute.

We propose FACTORS, a framework that combines design of experiments with Shapley decomposition to address performance and stability issues that are sensitive to combinations of training factors. Our approach consistently estimates main effects and two-factor interactions, then integrates them into a risk-adjusted objective function that jointly accounts for uncertainty and cost, enabling reliable selection of configurations under a fixed budget. Effect estimation is implemented through two complementary paths: a plug-in path based on conditional means, and a least-squares path that reconstructs Shapley contributions from samples. These paths are designed to work complementarily even when design density and bias levels differ. By incorporating standardization of estimates, bias correction, and uncertainty quantification, our procedure ensures comparability across heterogeneous factor spaces and designs, while a lightweight search routine yields configurations within practical time even for large factor spaces. On the theoretical side, we provide error decompositions, sample complexity analysis, and upper bounds on optimality gaps. On the interpretive side, we summarize main effects and interactions in map form, highlighting adjustment priorities and safe improvement pathways. Across diverse datasets and design conditions, our approach improves rank preservation and optimal configuration identification, reduces decision-making risks, and offers a tuning foundation that delivers interpretable justification alongside stable performance gains even under budget constraints.

The integration of Large Language Models (LLMs) into legal practice raises pressing concerns about judicial fairness, particularly due to the nature of their "black-box" processes. This study introduces JustEva, a comprehensive, open-source evaluation toolkit designed to measure LLM fairness in legal tasks. JustEva features several advantages: (1) a structured label system covering 65 extra-legal factors; (2) three core fairness metrics - inconsistency, bias, and imbalanced inaccuracy; (3) robust statistical inference methods; and (4) informative visualizations. The toolkit supports two types of experiments, enabling a complete evaluation workflow: (1) generating structured outputs from LLMs using a provided dataset, and (2) conducting statistical analysis and inference on LLMs' outputs through regression and other statistical methods. Empirical application of JustEva reveals significant fairness deficiencies in current LLMs, highlighting the lack of fair and trustworthy LLM legal tools. JustEva offers a convenient tool and methodological foundation for evaluating and improving algorithmic fairness in the legal domain.

Financial disclosures such as 10-K filings present challenging retrieval problems due to their length, regulatory section hierarchy, and domain-specific language, which standard retrieval-augmented generation (RAG) models underuse. We introduce FinGEAR (Financial Mapping-Guided Enhanced Answer Retrieval), a retrieval framework tailored to financial documents. FinGEAR combines a finance lexicon for Item-level guidance (FLAM), dual hierarchical indices for within-Item search (Summary Tree and Question Tree), and a two-stage cross-encoder reranker. This design aligns retrieval with disclosure structure and terminology, enabling fine-grained, query-aware context selection. Evaluated on full 10-Ks with queries aligned to the FinQA dataset, FinGEAR delivers consistent gains in precision, recall, F1, and relevancy, improving F1 by up to 56.7% over flat RAG, 12.5% over graph-based RAGs, and 217.6% over prior tree-based systems, while also increasing downstream answer accuracy with a fixed reader. By jointly modeling section hierarchy and domain lexicon signals, FinGEAR improves retrieval fidelity and provides a practical foundation for high-stakes financial analysis.

New technologies in generative AI can enable deeper analysis into our nation's supply chains but truly informative insights require the continual updating and aggregation of massive data in a timely manner. Large Language Models (LLMs) offer unprecedented analytical opportunities however, their knowledge base is constrained to the models' last training date, rendering these capabilities unusable for organizations whose mission impacts rely on emerging and timely information. This research proposes an innovative approach to supply chain analysis by integrating emerging Retrieval-Augmented Generation (RAG) preprocessing and retrieval techniques with advanced web-scraping technologies. Our method aims to reduce latency in incorporating new information into an augmented-LLM, enabling timely analysis of supply chain disruptors. Through experimentation, this study evaluates the combinatorial effects of these techniques towards timeliness and quality trade-offs. Our results suggest that in applying RAG systems to supply chain analysis, fine-tuning the embedding retrieval model consistently provides the most significant performance gains, underscoring the critical importance of retrieval quality. Adaptive iterative retrieval, which dynamically adjusts retrieval depth based on context, further enhances performance, especially on complex supply chain queries. Conversely, fine-tuning the LLM yields limited improvements and higher resource costs, while techniques such as downward query abstraction significantly outperforms upward abstraction in practice.

An intrinsic aspect of every conversation is the way talk-time is shared between multiple speakers. Conversations can be balanced, with each speaker claiming a similar amount of talk-time, or imbalanced when one talks disproportionately. Such overall distributions are the consequence of continuous negotiations between the speakers throughout the conversation: who should be talking at every point in time, and for how long? In this work we introduce a computational framework for quantifying both the conversation-level distribution of talk-time between speakers, as well as the lower-level dynamics that lead to it. We derive a typology of talk-time sharing dynamics structured by several intuitive axes of variation. By applying this framework to a large dataset of video-chats between strangers, we confirm that, perhaps unsurprisingly, different conversation-level distributions of talk-time are perceived differently by speakers, with balanced conversations being preferred over imbalanced ones, especially by those who end up talking less. Then we reveal that -- even when they lead to the same level of overall balance -- different types of talk-time sharing dynamics are perceived differently by the participants, highlighting the relevance of our newly introduced typology. Finally, we discuss how our framework offers new tools to designers of computer-mediated communication platforms, for both human-human and human-AI communication.

As large language models (LLMs) are increasingly deployed in high-stakes settings, the risk of generating harmful or toxic content remains a central challenge. Post-hoc alignment methods are brittle: once unsafe patterns are learned during pretraining, they are hard to remove. In this work, we present a data-centric pretraining framework that builds safety into the model from the start. Our framework consists of four key steps: (i) Safety Filtering: building a safety classifier to classify webdata into safe and unsafe categories; (ii) Safety Rephrasing: we recontextualize unsafe webdata into safer narratives; (iii) Native Refusal: we develop RefuseWeb and Moral Education pretraining datasets that actively teach model to refuse on unsafe content and the moral reasoning behind it, and (iv) Harmfulness-Tag annotated pretraining: we flag unsafe content during pretraining using a special token, and use it to steer model away from unsafe generations at inference. Our safety-pretrained models reduce attack success rates from 38.8\% to 8.4\% on standard LLM safety benchmarks with no performance degradation on general tasks.

A red team simulates adversary attacks to help defenders find effective strategies to defend their systems in a real-world operational setting. As more enterprise systems adopt AI, red-teaming will need to evolve to address the unique vulnerabilities and risks posed by AI systems. We take the position that AI systems can be more effectively red-teamed if AI red-teaming is recognized as a domain-specific evolution of cyber red-teaming. Specifically, we argue that existing Cyber Red Teams who adopt this framing will be able to better evaluate systems with AI components by recognizing that AI poses new risks, has new failure modes to exploit, and often contains unpatchable bugs that re-prioritize disclosure and mitigation strategies. Similarly, adopting a cybersecurity framing will allow existing AI Red Teams to leverage a well-tested structure to emulate realistic adversaries, promote mutual accountability with formal rules of engagement, and provide a pattern to mature the tooling necessary for repeatable, scalable engagements. In these ways, the merging of AI and Cyber Red Teams will create a robust security ecosystem and best position the community to adapt to the rapidly changing threat landscape.

Purpose: The governance of artificial iintelligence (AI) systems requires a structured approach that connects high-level regulatory principles with practical implementation. Existing frameworks lack clarity on how regulations translate into conformity mechanisms, leading to gaps in compliance and enforcement. This paper addresses this critical gap in AI governance. Methodology/Approach: A five-layer AI governance framework is proposed, spanning from broad regulatory mandates to specific standards, assessment methodologies, and certification processes. By narrowing its scope through progressively focused layers, the framework provides a structured pathway to meet technical, regulatory, and ethical requirements. Its applicability is validated through two case studies on AI fairness and AI incident reporting. Findings: The case studies demonstrate the framework's ability to identify gaps in legal mandates, standardization, and implementation. It adapts to both global and region-specific AI governance needs, mapping regulatory mandates with practical applications to improve compliance and risk management. Practical Implications - By offering a clear and actionable roadmap, this work contributes to global AI governance by equipping policymakers, regulators, and industry stakeholders with a model to enhance compliance and risk management. Social Implications: The framework supports the development of policies that build public trust and promote the ethical use of AI for the benefit of society. Originality/Value: This study proposes a five-layer AI governance framework that bridges high-level regulatory mandates and implementation guidelines. Validated through case studies on AI fairness and incident reporting, it identifies gaps such as missing standardized assessment procedures and reporting mechanisms, providing a structured foundation for targeted governance measures.

Artificial Intelligence Generated Content (AIGC) has rapidly emerged with the capability to generate different forms of content, including text, images, videos, and other modalities, which can achieve a quality similar to content created by humans. As a result, AIGC is now widely applied across various domains such as digital marketing, education, and public health, and has shown promising results by enhancing content creation efficiency and improving information delivery. However, there are few studies that explore the latest progress and emerging challenges of AIGC across different domains. To bridge this gap, this paper brings together 16 scholars from multiple disciplines to provide a cross-domain perspective on the trends and challenges of AIGC. Specifically, the contributions of this paper are threefold: (1) It first provides a broader overview of AIGC, spanning the training techniques of Generative AI, detection methods, and both the spread and use of AI-generated content across digital platforms. (2) It then introduces the societal impacts of AIGC across diverse domains, along with a review of existing methods employed in these contexts. (3) Finally, it discusses the key technical challenges and presents research propositions to guide future work. Through these contributions, this vision paper seeks to offer readers a cross-domain perspective on AIGC, providing insights into its current research trends, ongoing challenges, and future directions.