Fine-tuning LLMs on narrowly harmful datasets can lead to behavior that is broadly misaligned with respect to human values. To understand when and how this emergent misalignment occurs, we develop a comprehensive framework for detecting and characterizing rapid transitions during fine-tuning using both distributional change detection methods as well as order parameters that are formulated in plain English and evaluated by an LLM judge. Using an objective statistical dissimilarity measure, we quantify how the phase transition that occurs during fine-tuning affects multiple aspects of the model. In particular, we assess what percentage of the total distributional change in model outputs is captured by different aspects, such as alignment or verbosity, providing a decomposition of the overall transition. We also find that the actual behavioral transition occurs later in training than indicated by the peak in the gradient norm alone. Our framework enables the automated discovery and quantification of language-based order parameters, which we demonstrate on examples ranging from knowledge questions to politics and ethics.

In dense metropolitan areas, searching for street parking adds to traffic congestion. Like many other problems, real-time assistants based on mobile phones have been proposed, but their effectiveness is understudied. This work quantifies how varying levels of user coordination and information availability through such apps impact search time and the probability of finding street parking. Through a data-driven simulation of Madrid's street parking ecosystem, we analyze four distinct strategies: uncoordinated search (Unc-Agn), coordinated parking without awareness of non-users (Cord-Agn), an idealized oracle system that knows the positions of all non-users (Cord-Oracle), and our novel/practical Cord-Approx strategy that estimates non-users' behavior probabilistically. The Cord-Approx strategy, instead of requiring knowledge of how close non-users are to a certain spot in order to decide whether to navigate toward it, uses past occupancy distributions to elongate physical distances between system users and alternative parking spots, and then solves a Hungarian matching problem to dispatch accordingly. In high-fidelity simulations of Madrid's parking network with real traffic data, users of Cord-Approx averaged 6.69 minutes to find parking, compared to 19.98 minutes for non-users without an app. A zone-level snapshot shows that Cord-Approx reduces search time for system users by 72% (range = 67-76%) in central hubs, and up to 73% in residential areas, relative to non-users.

While stereotypes are well-documented in human social interactions, AI systems are often presumed to be less susceptible to such biases. Previous studies have focused on biases inherited from training data, but whether stereotypes can emerge spontaneously in AI agent interactions merits further exploration. Through a novel experimental framework simulating workplace interactions with neutral initial conditions, we investigate the emergence and evolution of stereotypes in LLM-based multi-agent systems. Our findings reveal that (1) LLM-Based AI agents develop stereotype-driven biases in their interactions despite beginning without predefined biases; (2) stereotype effects intensify with increased interaction rounds and decision-making power, particularly after introducing hierarchical structures; (3) these systems exhibit group effects analogous to human social behavior, including halo effects, confirmation bias, and role congruity; and (4) these stereotype patterns manifest consistently across different LLM architectures. Through comprehensive quantitative analysis, these findings suggest that stereotype formation in AI systems may arise as an emergent property of multi-agent interactions, rather than merely from training data biases. Our work underscores the need for future research to explore the underlying mechanisms of this phenomenon and develop strategies to mitigate its ethical impacts.

Current safety alignment for large language models(LLMs) continues to present vulnerabilities, given that adversarial prompting can effectively bypass their safety measures.Our investigation shows that these safety mechanisms predominantly depend on a limited subset of attention heads: removing or ablating these heads can severely compromise model safety. To identify and evaluate these safety-critical components, we introduce RDSHA, a targeted ablation method that leverages the model's refusal direction to pinpoint attention heads mostly responsible for safety behaviors. Further analysis shows that existing jailbreak attacks exploit this concentration by selectively bypassing or manipulating these critical attention heads. To address this issue, we propose AHD, a novel training strategy designed to promote the distributed encoding of safety-related behaviors across numerous attention heads. Experimental results demonstrate that AHD successfully distributes safety-related capabilities across more attention heads. Moreover, evaluations under several mainstream jailbreak attacks show that models trained with AHD exhibit considerably stronger safety robustness, while maintaining overall functional utility.

Achieving both agile maneuverability and high energy efficiency in aerial robots, particularly in dynamic wind environments, remains challenging. Conventional thruster-powered systems offer agility but suffer from high energy consumption, while fixed-wing designs are efficient but lack hovering and maneuvering capabilities. We present Floaty, a shape-changing robot that overcomes these limitations by passively soaring, harnessing wind energy through intelligent morphological control inspired by birds. Floaty's design is optimized for passive stability, and its control policy is derived from an experimentally learned aerodynamic model, enabling precise attitude and position control without active propulsion. Wind tunnel experiments demonstrate Floaty's ability to hover, maneuver, and reject disturbances in vertical airflows up to 10 m/s. Crucially, Floaty achieves this with a specific power consumption of 10 W/kg, an order of magnitude lower than thruster-powered systems. This introduces a paradigm for energy-efficient aerial robotics, leveraging morphological intelligence and control to operate sustainably in challenging wind conditions.

Graph-structured data, which captures non-Euclidean relationships and interactions between entities, is growing in scale and complexity. As a result, training state-of-the-art graph machine learning (GML) models have become increasingly resource-intensive, turning these models and data into invaluable Intellectual Property (IP). To address the resource-intensive nature of model training, graph-based Machine-Learning-as-a-Service (GMLaaS) has emerged as an efficient solution by leveraging third-party cloud services for model development and management. However, deploying such models in GMLaaS also exposes them to potential threats from attackers. Specifically, while the APIs within a GMLaaS system provide interfaces for users to query the model and receive outputs, they also allow attackers to exploit and steal model functionalities or sensitive training data, posing severe threats to the safety of these GML models and the underlying graph data. To address these challenges, this survey systematically introduces the first taxonomy of threats and defenses at the level of both GML model and graph-structured data. Such a tailored taxonomy facilitates an in-depth understanding of GML IP protection. Furthermore, we present a systematic evaluation framework to assess the effectiveness of IP protection methods, introduce a curated set of benchmark datasets across various domains, and discuss their application scopes and future challenges. Finally, we establish an open-sourced versatile library named PyGIP, which evaluates various attack and defense techniques in GMLaaS scenarios and facilitates the implementation of existing benchmark methods. The library resource can be accessed at: https://labrai.github.io/PyGIP. We believe this survey will play a fundamental role in intellectual property protection for GML and provide practical recipes for the GML community.

Generative models such as Large Language Models, Diffusion Models, and generative adversarial networks have recently revolutionized the creation of synthetic data, offering scalable solutions to data scarcity, privacy, and annotation challenges in data mining. This tutorial introduces the foundations and latest advances in synthetic data generation, covers key methodologies and practical frameworks, and discusses evaluation strategies and applications. Attendees will gain actionable insights into leveraging generative synthetic data to enhance data mining research and practice. More information can be found on our website: https://syndata4dm.github.io/.

This paper introduces Democracy-in-Silico, an agent-based simulation where societies of advanced AI agents, imbued with complex psychological personas, govern themselves under different institutional frameworks. We explore what it means to be human in an age of AI by tasking Large Language Models (LLMs) to embody agents with traumatic memories, hidden agendas, and psychological triggers. These agents engage in deliberation, legislation, and elections under various stressors, such as budget crises and resource scarcity. We present a novel metric, the Power-Preservation Index (PPI), to quantify misaligned behavior where agents prioritize their own power over public welfare. Our findings demonstrate that institutional design, specifically the combination of a Constitutional AI (CAI) charter and a mediated deliberation protocol, serves as a potent alignment mechanism. These structures significantly reduce corrupt power-seeking behavior, improve policy stability, and enhance citizen welfare compared to less constrained democratic models. The simulation reveals that an institutional design may offer a framework for aligning the complex, emergent behaviors of future artificial agent societies, forcing us to reconsider what human rituals and responsibilities are essential in an age of shared authorship with non-human entities.

With advancements in AI, new gaze estimation methods are exceeding state-of-the-art (SOTA) benchmarks, but their real-world application reveals a gap with commercial eye-tracking solutions. Factors like model size, inference time, and privacy often go unaddressed. Meanwhile, webcam-based eye-tracking methods lack sufficient accuracy, in particular due to head movement. To tackle these issues, we introduce We bEyeTrack, a framework that integrates lightweight SOTA gaze estimation models directly in the browser. It incorporates model-based head pose estimation and on-device few-shot learning with as few as nine calibration samples (k < 9). WebEyeTrack adapts to new users, achieving SOTA performance with an error margin of 2.32 cm on GazeCapture and real-time inference speeds of 2.4 milliseconds on an iPhone 14. Our open-source code is available at https://github.com/RedForestAi/WebEyeTrack.

Current Emotion Recognition in Conversation (ERC) research follows a closed-domain assumption. However, there is no clear consensus on emotion classification in psychology, which presents a challenge for models when it comes to recognizing previously unseen emotions in real-world applications. To bridge this gap, we introduce the Unseen Emotion Recognition in Conversation (UERC) task for the first time and propose ProEmoTrans, a solid prototype-based emotion transfer framework. This prototype-based approach shows promise but still faces key challenges: First, implicit expressions complicate emotion definition, which we address by proposing an LLM-enhanced description approach. Second, utterance encoding in long conversations is difficult, which we tackle with a proposed parameter-free mechanism for efficient encoding and overfitting prevention. Finally, the Markovian flow nature of emotions is hard to transfer, which we address with an improved Attention Viterbi Decoding (AVD) method to transfer seen emotion transitions to unseen emotions. Extensive experiments on three datasets show that our method serves as a strong baseline for preliminary exploration in this new area.