William Yeoh

Knowledge distillation is a widely adopted technique for transferring capabilities from LLMs to smaller, more efficient student models. However, unauthorized use of knowledge distillation takes unfair advantage of the considerable effort and cost put into developing frontier models. We investigate methods for modifying teacher-generated reasoning traces to achieve two objectives that deter unauthorized distillation: (1) anti-distillation, or degrading the training usefulness of query responses, and (2) API watermarking, which embeds verifiable signatures in student models. We introduce several approaches for dynamically rewriting a teacher’s reasoning outputs while preserving answer correctness and semantic coherence. Two of these leverage the rewriting capabilities of LLMs, while others use gradient-based techniques. Our experiments show that a simple instruction-based rewriting approach achieves a strong anti-distillation effect while maintaining or even improving teacher performance. Furthermore, we show that our rewriting approach also enables embedding watermarks that can be reliably detected with essentially no false alarms. Our code is available at https://github.com/xhOwenMa/trace-rewriting.

A wide variety of resource allocation problems operate under resource constraints that are managed by a central arbitrator, with agents who evaluate and communicate preferences over these resources. We formulate this broad class of problems as Distributed Evaluation, Centralized Allocation (DECA) problems and propose methods to learn fair and efficient policies in centralized resource allocation. Our methods are applied to learning long-term fairness in a novel and general framework for fairness in multi-agent systems. We show three different methods based on Double Deep Q-Learning: (1) A joint weighted optimization of fairness and utility, (2) a split optimization, learning two separate Q-estimators for utility and fairness, and (3) an online policy perturbation to guide existing black-box utility functions toward fair solutions. Our methods outperform existing fair MARL approaches on multiple resource allocation domains, even when evaluated using diverse fairness functions, and allow for flexible online trade-offs between utility and fairness.

Computational argumentation offers formal frameworks for transparent, verifiable reasoning but has traditionally been limited by its reliance on domain-specific information and extensive feature engineering. In contrast, LLMs excel at processing unstructured text, yet their opaque nature makes their reasoning difficult to evaluate and trust. We argue that the convergence of these fields will lay the foundation for a new paradigm: Argumentative Human-AI Decision-Making. We analyze how the synergy of argumentation framework mining, argumentation framework synthesis, and argumentative reasoning enables agents that do not just justify decisions, but engage in dialectical processes where decisions are contestable and revisable – reasoning with humans rather than for them. This convergence of computational argumentation and LLMs is essential for human-aware, trustworthy AI in high-stakes domains.

Distributed Constraint Reasoning (DCR) has long provided a principled framework for modeling and solving multi-agent coordination and optimization problems. However, its practical adoption in real-world, human-centric domains has been hindered by the challenge of translating human intentions, preferences, and constraints into formal symbolic models. At the same time, recent advances in LLMs have enabled powerful agentic capabilities, including natural language understanding, flexible reasoning, and interactive problem solving, but these systems lack the formal rigor and guarantees needed for scalable multi-agent coordination. In this paper, we argue that the convergence of these two paradigms offers a timely and transformative opportunity. We articulate several synergistic research directions: leveraging LLMs for translating natural language into DCR specifications, eliciting and refining user preferences, and enhancing inter-agent communication; and conversely, applying DCR models and algorithms to improve coordination, structured reasoning, resource allocation, and communication sensitivity in Agentic LLM systems. Together, these threads point toward hybrid neurosymbolic systems that combine the adaptability of LLMs with the mathematical rigor of DCR.

In human-AI interaction, effective communication relies on aligning the AI agent’s model with the human user’s mental model, a process known as model reconciliation. However, existing model reconciliation approaches predominantly assume deterministic models, overlooking the fact that human knowledge is often uncertain or probabilistic. To bridge this gap, we present a probabilistic model reconciliation framework that resolves inconsistencies in MPE outcome probabilities between an agent’s and a user’s models. Our approach is built on probabilistic logic programming (PLP) using ProbLog, where explanations are generated as cost-optimal model updates that reconcile these probabilistic differences. We develop two search algorithms – a generic baseline and an optimized version. The latter is guided by theoretical insights and further extended with greedy and weighted variants to enhance scalability and efficiency. Our approach is validated through a user study on explanation types and computational experiments showing that the optimized version consistently outperforms the generic baseline.

The introduction of social media websites touted the idea of global communication — exposing users to a worldwide audience and a diverse range of experiences, opinions, and debates. Unfortunately, studies have shown that social networks have instead contributed to growing levels of polarization in society across a wide variety of issues. Social media websites employ algorithmic filtering strategies to drive engagement, which can lead to the formation of filter bubbles and increased levels of polarization. In this paper, we introduce features of affective polarization — feelings towards one’s in-group and out-group — into an opinion dynamics model. Specifically, we show that incorporating a negative out-group stereotype into the opinion dynamics model (1) affects the level of polarization present among agents in the network; (2) changes the effectiveness of algorithmic filtering strategies; and (3) is exacerbated by the presence of extremists in the network. Hence, the inclusion of an affective group mechanism in opinion dynamics modeling provides novel insights into the effects of algorithmic filtering strategies on the extremity of opinions in social networks.