Current Vision Language Models (VLMs) remain vulnerable to malicious prompts that induce harmful outputs. Existing safety benchmarks for VLMs primarily rely on automated evaluation methods, but these methods struggle to detect implicit harmful content or produce inaccurate evaluations. Therefore, we found that existing benchmarks have low levels of harmfulness, ambiguous data, and limited diversity in image-text pair combinations. To address these issues, we propose the ELITE benchmark, a high-quality safety evaluation benchmark for VLMs, underpinned by our enhanced evaluation method, the ELITE evaluator. The ELITE evaluator explicitly incorporates a toxicity score to accurately assess harmfulness in multimodal contexts, where VLMs often provide specific, convincing, but unharmful descriptions of images. We filter out ambiguous and low-quality image-text pairs from existing benchmarks using the ELITE evaluator and generate diverse combinations of safe and unsafe image-text pairs. Our experiments demonstrate that the ELITE evaluator achieves superior alignment with human evaluations compared to prior automated methods, and the ELITE benchmark offers enhanced benchmark quality and diversity. By introducing ELITE, we pave the way for safer, more robust VLMs, contributing essential tools for evaluating and mitigating safety risks in real-world applications.

This study evaluates the performance and usability of Mixed Reality (MR), Virtual Reality (VR), and camera stream interfaces for remote error resolution tasks, such as correcting warehouse packaging errors. Specifically, we consider a scenario where a robotic arm halts after detecting an error, requiring a remote operator to intervene and resolve it via pick-and-place actions. Twenty-one participants performed simulated pick-and-place tasks using each interface. A linear mixed model (LMM) analysis of task resolution time, usability scores (SUS), and mental workload scores (NASA-TLX) showed that the MR interface outperformed both VR and camera interfaces. MR enabled significantly faster task completion, was rated higher in usability, and was perceived to be less cognitively demanding. Notably, the MR interface, which projected a virtual robot onto a physical table, provided superior spatial understanding and physical reference cues. Post-study surveys further confirmed participants' preference for MR over other interfaces.

The generation of incorrect images, such as depictions of people of color in Nazi-era uniforms by Gemini, frustrated users and harmed Google's reputation, motivating us to investigate the relationship between accurately reflecting factuality and promoting diversity and equity. In this study, we focus on 19 real-world statistics collected from authoritative sources. Using these statistics, we develop a checklist comprising objective and subjective queries to analyze behavior of large language models (LLMs) and text-to-image (T2I) models. Objective queries assess the models' ability to provide accurate world knowledge. In contrast, the design of subjective queries follows a key principle: statistical or experiential priors should not be overgeneralized to individuals, ensuring that models uphold diversity. These subjective queries are based on three common human cognitive errors that often result in social biases. We propose metrics to assess factuality and fairness, and formally prove the inherent trade-off between these two aspects. Results show that GPT-4o and DALL-E 3 perform notably well among six LLMs and four T2I models. Our code is publicly available at https://github.com/uclanlp/Fact-or-Fair.

Optimization is instrumental for improving operations of large-scale socio-technical infrastructures of Smart Cities, for instance, energy and traffic systems. In particular, understanding the performance of multi-agent discrete-choice combinatorial optimization under distributed adversary attacks is a compelling and underexplored problem, since multi-agent systems exhibit a large number of remote control variables that can influence in an unprecedented way the cost-effectiveness of distributed optimization heuristics. This paper unravels for the first time the trajectories of distributed optimization from resilience to vulnerability, and finally to collapse under varying adversary influence. Using real-world data to emulate over 28 billion multi-agent optimization scenarios, we exhaustively assess how the number of agents with different adversarial severity and network positioning influences optimization performance, including the influence on Pareto optimal points. With this novel large-scale dataset, made openly available as a benchmark, we disentangle how optimization remains resilient to adversaries and which adversary conditions are required to make optimization vulnerable or collapsed. These new findings can provide new insights for designing self-healing strategies for fault-tolerance and fault-correction in adversarial distributed optimization that have been missing so far.

Under what conditions can capable AI agents efficiently align their actions with human preferences? More specifically, when they are proficient enough to collaborate with us, how long does coordination take, and when is it computationally feasible? These foundational questions of AI alignment help define what makes an AI agent ``sufficiently safe'' and valuable to humans. Since such generally capable systems do not yet exist, a theoretical analysis is needed to establish when guarantees hold -- and what they even are. We introduce a game-theoretic framework that generalizes prior alignment approaches with fewer assumptions, allowing us to analyze the computational complexity of alignment across $M$ objectives and $N$ agents, providing both upper and lower bounds. Unlike previous work, which often assumes common priors, idealized communication, or implicit tractability, our framework formally characterizes the difficulty of alignment under minimal assumptions. Our main result shows that even when agents are fully rational and computationally \emph{unbounded}, alignment can be achieved with high probability in time \emph{linear} in the task space size. Therefore, in real-world settings, where task spaces are often \emph{exponential} in input length, this remains impractical. More strikingly, our lower bound demonstrates that alignment is \emph{impossible} to speed up when scaling to exponentially many tasks or agents, highlighting a fundamental computational barrier to scalable alignment. Relaxing these idealized assumptions, we study \emph{computationally bounded} agents with noisy messages (representing obfuscated intent), showing that while alignment can still succeed with high probability, it incurs additional \emph{exponential} slowdowns in the task space size, number of agents, and number of tasks. We conclude by identifying conditions that make alignment more feasible.

Drug development is a critical but notoriously resource- and time-consuming process. In this manuscript, we develop a novel generative artificial intelligence (genAI) method DiffSMol to facilitate drug development. DiffSmol generates 3D binding molecules based on the shapes of known ligands. DiffSMol encapsulates geometric details of ligand shapes within pre-trained, expressive shape embeddings and then generates new binding molecules through a diffusion model. DiffSMol further modifies the generated 3D structures iteratively via shape guidance to better resemble the ligand shapes. It also tailors the generated molecules toward optimal binding affinities under the guidance of protein pockets. Here, we show that DiffSMol outperforms the state-of-the-art methods on benchmark datasets. When generating binding molecules resembling ligand shapes, DiffSMol with shape guidance achieves a success rate 61.4%, substantially outperforming the best baseline (11.2%), meanwhile producing molecules with novel molecular graph structures. DiffSMol with pocket guidance also outperforms the best baseline in binding affinities by 13.2%, and even by 17.7% when combined with shape guidance. Case studies for two critical drug targets demonstrate very favorable physicochemical and pharmacokinetic properties of the generated molecules, thus, the potential of DiffSMol in developing promising drug candidates.

Transformers have recently shown strong performance in time-series forecasting, but their all-to-all attention mechanism overlooks the (temporal) causal and often (temporally) local nature of data. We introduce Powerformer, a novel Transformer variant that replaces noncausal attention weights with causal weights that are reweighted according to a smooth heavy-tailed decay. This simple yet effective modification endows the model with an inductive bias favoring temporally local dependencies, while still allowing sufficient flexibility to learn the unique correlation structure of each dataset. Our empirical results demonstrate that Powerformer not only achieves state-of-the-art accuracy on public time-series benchmarks, but also that it offers improved interpretability of attention patterns. Our analyses show that the model's locality bias is amplified during training, demonstrating an interplay between time-series data and power-law-based attention. These findings highlight the importance of domain-specific modifications to the Transformer architecture for time-series forecasting, and they establish Powerformer as a strong, efficient, and principled baseline for future research and real-world applications.

Precise alignment in Text-to-Image (T2I) systems is crucial to ensure that generated visuals not only accurately encapsulate user intents but also conform to stringent ethical and aesthetic benchmarks. Incidents like the Google Gemini fiasco, where misaligned outputs triggered significant public backlash, underscore the critical need for robust alignment mechanisms. In contrast, Large Language Models (LLMs) have achieved notable success in alignment. Building on these advancements, researchers are eager to apply similar alignment techniques, such as Direct Preference Optimization (DPO), to T2I systems to enhance image generation fidelity and reliability. We present YinYangAlign, an advanced benchmarking framework that systematically quantifies the alignment fidelity of T2I systems, addressing six fundamental and inherently contradictory design objectives. Each pair represents fundamental tensions in image generation, such as balancing adherence to user prompts with creative modifications or maintaining diversity alongside visual coherence. YinYangAlign includes detailed axiom datasets featuring human prompts, aligned (chosen) responses, misaligned (rejected) AI-generated outputs, and explanations of the underlying contradictions.

While recent Large Vision-Language Models (LVLMs) have shown remarkable performance in multi-modal tasks, they are prone to generating hallucinatory text responses that do not align with the given visual input, which restricts their practical applicability in real-world scenarios. In this work, inspired by the observation that the text-to-image generation process is the inverse of image-conditioned response generation in LVLMs, we explore the potential of leveraging text-to-image generative models to assist in mitigating hallucinations in LVLMs. We discover that generative models can offer valuable self-feedback for mitigating hallucinations at both the response and token levels. Building on this insight, we introduce self-correcting Decoding with Generative Feedback (DeGF), a novel training-free algorithm that incorporates feedback from text-to-image generative models into the decoding process to effectively mitigate hallucinations in LVLMs. Specifically, DeGF generates an image from the initial response produced by LVLMs, which acts as an auxiliary visual reference and provides self-feedback to verify and correct the initial response through complementary or contrastive decoding. Extensive experimental results validate the effectiveness of our approach in mitigating diverse types of hallucinations, consistently surpassing state-of-the-art methods across six benchmarks. Code is available at https://github.com/zhangce01/DeGF.

Retrieval-Augmented Generation (RAG), which integrates external knowledge into Large Language Models (LLMs), has proven effective in enabling LLMs to produce more accurate and reliable responses. However, it remains a significant challenge how to effectively integrate external retrieved knowledge with internal parametric knowledge in LLMs. In this work, we propose a novel Self-Selection RAG framework, where the LLM is made to select from pairwise responses generated with internal parametric knowledge solely and with external retrieved knowledge together to achieve enhanced accuracy. To this end, we devise a Self-Selection-RGP method to enhance the capabilities of the LLM in both generating and selecting the correct answer, by training the LLM with Direct Preference Optimization (DPO) over a curated Retrieval Generation Preference (RGP) dataset. Experimental results with two open-source LLMs (i.e., Llama2-13B-Chat and Mistral-7B) well demonstrate the superiority of our approach over other baseline methods on Natural Questions (NQ) and TrivialQA datasets.