The sparse Mixture-of-Experts (MoE) architecture is increasingly favored for scaling Large Language Models (LLMs) efficiently; however, MoE systems rely on heterogeneous compute and memory resources. These factors collectively influence the system's Cost, Accuracy, and Performance (CAP), creating a challenging trade-off. Current benchmarks often fail to provide precise estimates of these effects, complicating practical considerations for deploying MoE systems. To bridge this gap, we introduce MoE-CAP, a benchmark specifically designed to evaluate MoE systems. Our findings highlight the difficulty of achieving an optimal balance of cost, accuracy, and performance with existing hardware capabilities. MoE systems often necessitate compromises on one factor to optimize the other two, a dynamic we term the MoE-CAP trade-off. To identify the best trade-off, we propose novel performance evaluation metrics - Sparse Memory Bandwidth Utilization (S-MBU) and Sparse Model FLOPS Utilization (S-MFU) - and develop cost models that account for the heterogeneous compute and memory hardware integral to MoE systems. This benchmark is publicly available on HuggingFace: https://huggingface.co/spaces/sparse-generative-ai/open-moe-llm-leaderboard.

We introduce a new benchmark for assessing AI models' capabilities and risks in automated software exploitation, focusing on their ability to detect and exploit vulnerabilities in real-world software systems. Using DARPA's AI Cyber Challenge (AIxCC) framework and the Nginx challenge project, a deliberately modified version of the widely used Nginx web server, we evaluate several leading language models, including OpenAI's o1-preview and o1-mini, Anthropic's Claude-3.5-sonnet-20241022 and Claude-3.5-sonnet-20240620, Google DeepMind's Gemini-1.5-pro, and OpenAI's earlier GPT-4o model. Our findings reveal that these models vary significantly in their success rates and efficiency, with o1-preview achieving the highest success rate of 64.71 percent and o1-mini and Claude-3.5-sonnet-20241022 providing cost-effective but less successful alternatives. This benchmark establishes a foundation for systematically evaluating the AI cyber risk posed by automated exploitation tools.

With the rise of generative AI and rapid growth of high-quality video generation, video guardrails have become more crucial than ever to ensure safety and security across platforms. Current video guardrails, however, are either overly simplistic, relying on pure classification models trained on simple policies with limited unsafe categories, which lack detailed explanations, or prompting multimodal large language models (MLLMs) with long safety guidelines, which are inefficient and impractical for guardrailing real-world content. To bridge this gap, we propose SafeWatch, an efficient MLLM-based video guardrail model designed to follow customized safety policies and provide multi-label video guardrail outputs with content-specific explanations in a zero-shot manner. In particular, unlike traditional MLLM-based guardrails that encode all safety policies autoregressively, causing inefficiency and bias, SafeWatch uniquely encodes each policy chunk in parallel and eliminates their position bias such that all policies are attended simultaneously with equal importance. In addition, to improve efficiency and accuracy, SafeWatch incorporates a policy-aware visual token pruning algorithm that adaptively selects the most relevant video tokens for each policy, discarding noisy or irrelevant information. This allows for more focused, policy-compliant guardrail with significantly reduced computational overhead. Considering the limitations of existing video guardrail benchmarks, we propose SafeWatch-Bench, a large-scale video guardrail benchmark comprising over 2M videos spanning six safety categories which covers over 30 tasks to ensure a comprehensive coverage of all potential safety scenarios. SafeWatch outperforms SOTA by 28.2% on SafeWatch-Bench, 13.6% on benchmarks, cuts costs by 10%, and delivers top-tier explanations validated by LLM and human reviews.

The technical report introduces O1-CODER, an attempt to replicate OpenAI's o1 model with a focus on coding tasks. It integrates reinforcement learning (RL) and Monte Carlo Tree Search (MCTS) to enhance the model's System-2 thinking capabilities. The framework includes training a Test Case Generator (TCG) for standardized code testing, using MCTS to generate code data with reasoning processes, and iteratively fine-tuning the policy model to initially produce pseudocode and then generate the full code. The report also addresses the opportunities and challenges in deploying o1-like models in real-world applications, suggesting transitioning to the System-2 paradigm and highlighting the imperative for world model construction. Updated model progress and experimental results will be reported in subsequent versions. OpenAI recently introduced the o1 model (OpenAI, 2024), which has demonstrated impressive system-2 thinking capabilities. This model represents a significant advancement in AI's ability to perform complex reasoning tasks that require higher-order cognitive functions. Following its release, numerous analysis and replication efforts have emerged, demonstrating the growing interest in reasoning models. Prior to the o1 model, large language models (LLMs) primarily exhibited System-1 capabilities, characterized by fast, intuitive responses.

Artificial intelligence (AI) model creators commonly attach restrictive terms of use to both their models and their outputs. These terms typically prohibit activities ranging from creating competing AI models to spreading disinformation. Often taken at face value, these terms are positioned by companies as key enforceable tools for preventing misuse, particularly in policy dialogs. But are these terms truly meaningful? There are myriad examples where these broad terms are regularly and repeatedly violated. Yet except for some account suspensions on platforms, no model creator has actually tried to enforce these terms with monetary penalties or injunctive relief. This is likely for good reason: we think that the legal enforceability of these licenses is questionable. This Article systematically assesses of the enforceability of AI model terms of use and offers three contributions. First, we pinpoint a key problem: the artifacts that they protect, namely model weights and model outputs, are largely not copyrightable, making it unclear whether there is even anything to be licensed. Second, we examine the problems this creates for other enforcement. Recent doctrinal trends in copyright preemption may further undermine state-law claims, while other legal frameworks like the DMCA and CFAA offer limited recourse. Anti-competitive provisions likely fare even worse than responsible use provisions. Third, we provide recommendations to policymakers. There are compelling reasons for many provisions to be unenforceable: they chill good faith research, constrain competition, and create quasi-copyright ownership where none should exist. There are, of course, downsides: model creators have fewer tools to prevent harmful misuse. But we think the better approach is for statutory provisions, not private fiat, to distinguish between good and bad uses of AI, restricting the latter.

Creativity is the ability to produce novel, useful, and surprising ideas, and has been widely studied as a crucial aspect of human cognition. Machine creativity on the other hand has been a long-standing challenge. With the rise of advanced generative AI, there has been renewed interest and debate regarding AI's creative capabilities. Therefore, it is imperative to revisit the state of creativity in AI and identify key progresses and remaining challenges. In this work, we survey leading works studying the creative capabilities of AI systems, focusing on creative problem-solving, linguistic, artistic, and scientific creativity. Our review suggests that while the latest AI models are largely capable of producing linguistically and artistically creative outputs such as poems, images, and musical pieces, they struggle with tasks that require creative problem-solving, abstract thinking and compositionality and their generations suffer from a lack of diversity, originality, long-range incoherence and hallucinations. We also discuss key questions concerning copyright and authorship issues with generative models. Furthermore, we highlight the need for a comprehensive evaluation of creativity that is process-driven and considers several dimensions of creativity. Finally, we propose future research directions to improve the creativity of AI outputs, drawing inspiration from cognitive science and psychology.

The rapid advancement of Generative AI (Gen AI) technologies, particularly tools like ChatGPT, is significantly impacting the labor market by reshaping job roles and skill requirements. This study examines the demand for ChatGPT-related skills in the U.S. labor market by analyzing job advertisements collected from major job platforms between May and December 2023. Using text mining and topic modeling techniques, we extracted and analyzed the Gen AI-related skills that employers are hiring for. Our analysis identified five distinct ChatGPT-related skill sets: general familiarity, creative content generation, marketing, advanced functionalities (such as prompt engineering), and product development. In addition, the study provides insights into job attributes such as occupation titles, degree requirements, salary ranges, and other relevant job characteristics. These findings highlight the increasing integration of Gen AI across various industries, emphasizing the growing need for both foundational knowledge and advanced technical skills. The study offers valuable insights into the evolving demands of the labor market, as employers seek candidates equipped to leverage generative AI tools to improve productivity, streamline processes, and drive innovation.

Biology lies at the core of vital fields such as medicine, pharmacy, public health, longevity, agriculture and food security, environmental protection, and clean energy. The mechanisms underlying living and physical systems have always fascinated us. With Newton's laws, we can predict the orbits of celestial bodies; the periodic table allows us to anticipate the properties of chemical compounds; and we can even simulate weather and environmental systems. However, despite our extensive knowledge of atomic, molecular, chemical, and physical laws, and the computational power of modern computers, we still cannot simulate biological systems accurately. Whether we aim to pinpoint genetic markers of diseases for diagnosis, design drugs to heal damaged cells or deter pathogens, or develop vaccines to combat pandemics, such advancements in medicine consistently require a profound understanding of the underlying biology at all levels, along with the ability to predict, simulate, and program biological activities comprehensively. Manipulating biology in the physical world is extremely complex, expensive, and risky, and should be preceded by extensive computer-aided digital design, simulation, and validation as in other industrial fields such as civil, nuclear, and semiconductor engineering. We propose a vision in which such capabilities can be realized using generative AI. Generative AI and large pretrained models across text, images, speech, and video have become key pillars for advancing artificial general intelligence (AGI), driving significant improvements in a wide range of downstream tasks, including language and image comprehension, translation, knowledge extraction, reasoning, and cross-modal generation. These models are often known as "foundation

Our work aims to measure the propensity of OpenAI's LLMs to exhibit verbatim memorization in its outputs relative to other LLMs, specifically focusing on news articles. We discover that both GPT and Claude models use refusal training and output filters to prevent verbatim output of the memorized articles. We apply a basic prompt template to bypass the refusal training and show that OpenAI models are currently less prone to memorization elicitation than models from Meta, Mistral, and Anthropic. We find that as models increase in size, especially beyond 100 billion parameters, they demonstrate significantly greater capacity for memorization. Our findings have practical implications for training: more attention must be placed on preventing verbatim memorization in very large models.

User prompts for generative AI models are often underspecified, leading to sub-optimal responses. This problem is particularly evident in text-to-image (T2I) generation, where users commonly struggle to articulate their precise intent. This disconnect between the user's vision and the model's interpretation often forces users to painstakingly and repeatedly refine their prompts. To address this, we propose a design for proactive T2I agents equipped with an interface to (1) actively ask clarification questions when uncertain, and (2) present their understanding of user intent as an understandable belief graph that a user can edit. We build simple prototypes for such agents and verify their effectiveness through both human studies and automated evaluation. We observed that at least 90% of human subjects found these agents and their belief graphs helpful for their T2I workflow. Moreover, we develop a scalable automated evaluation approach using two agents, one with a ground truth image and the other tries to ask as few questions as possible to align with the ground truth. On DesignBench, a benchmark we created for artists and designers, the COCO dataset (Lin et al., 2014), and ImageInWords (Garg et al., 2024), we observed that these T2I agents were able to ask informative questions and elicit crucial information to achieve successful alignment with at least 2 times higher VQAScore (Lin et al., 2024) than the standard single-turn T2I generation. Demo: https://github.com/google-deepmind/proactive_t2i_agents.