Core Insights - The article discusses the evolution of large models from LLMs to MLLMs and then to Agents, highlighting the increasing capabilities and associated security risks, particularly focusing on jailbreak attacks as a significant threat [2][3][4]. Group 1: Evolution of Large Models - The transition from LLMs to MLLMs and then to Agents represents a significant paradigm shift in AI, with each stage introducing new capabilities and security challenges [7][16]. - LLMs, based on neural network breakthroughs, have limitations in handling multi-modal data, leading to the development of MLLMs that integrate text, image, and audio [8][12]. - MLLMs expand capabilities but also increase attack surfaces, allowing for more sophisticated jailbreak attacks that exploit visual and audio vulnerabilities [13][15]. Group 2: Jailbreak Attack Classification - The article proposes a dual-dimensional classification framework for jailbreak attacks based on "attack impact" and "attacker permissions," providing a comprehensive analysis of attack methods across different model types [25][32]. - Attacks are categorized into training phase and inference phase, with specific techniques such as backdoor attacks and prompt attacks identified [29][30]. - The classification also distinguishes between white-box and black-box attacks, emphasizing the varying levels of access attackers have to model internals [32][36]. Group 3: Data Sets and Evaluation Metrics - The article reviews existing datasets and evaluation metrics for jailbreak research, noting limitations in diversity and coverage, particularly in multi-modal and multi-turn scenarios [37][43]. - It categorizes datasets based on their sources and formats, highlighting the need for improved dynamic datasets that can keep pace with evolving attack strategies [39][41]. - Five main categories of evaluation metrics are discussed, including human evaluation, automated assessments, and custom metrics tailored to specific research needs [44][58].

" 欧米伽未来研究所 " 关注科技未来发展趋势，研究人类向欧米伽点演化过程中面临的重大机遇与挑战。将不定期推荐和发布世界范围重要科技研究进展和未 来趋势研究。（ 点击这里查看欧米伽理论 ） 我们生活在一个由空间构成的世界中。从每天在家居、办公环境或城市街道中的移动，到规划一次跨越山海的旅行，乃至科学家们研究气候变迁的地理模 式、城市扩张的复杂格局，这一切都深刻地依赖于我们对空间的感知、理解和运用能力。这种核心能力，我们称之为"空间智能"。 长久以来，人类凭借自身的感官系统和发达的大脑，不断地探索、适应并改造着周遭的空间环境，演化出了独特的空间认知机制。而今，随着人工智能 （AI）技术的日新月异，特别是大语言模型（LLMs）的横空出世，机器也开始显露出令人瞩目的空间智能潜力。这场由大模型引领的技术浪潮，正以前 所未有的深度和广度，渗透到从微观尺度的机器人导航，到中观尺度的城市规划管理，再到宏观尺度的地球科学研究等诸多领域。 这部报告由清华大学和芬兰赫尔辛基大学共同发布，将带领读者一同深入探究，大模型是如何被赋予"空间感"的？它们在跨越不同尺度的空间智能任务中 扮演着怎样日益重要的角色？以及在迈向更高级空间智能的 ...

Core Viewpoint - The article discusses the evolution of spatial intelligence through the development of large language models (LLMs) and their applications across various scales, from micro-level robotics to macro-level earth sciences, highlighting both opportunities and challenges in this field [1][2][35]. Section Summaries Section 1: The Foundation of Spatial Intelligence - How Large Models "Understand" Space - To enable machines to possess spatial intelligence, they must develop effective spatial memory and flexible abstract spatial reasoning capabilities [2][3]. Section 2: Spatial Memory and Knowledge - The "Cognitive Map" in Large Models - Large models acquire spatial information through "internal absorption" during pre-training and "external invocation" when needing real-time data [4][5]. Section 3: Abstract Spatial Reasoning - Beyond Memorization - Current large models primarily mimic spatial tasks using language modeling rather than possessing deep spatial reasoning akin to human cognition [9]. Section 4: Multi-Scale Spatial Intelligence Applications Driven by Large Models - Large models are increasingly important in various spatial intelligence tasks across different scales, from individual robots to urban environments and global systems [10][11]. Section 5: Embodied Intelligence - Enhancing Robot Spatial Understanding and Action - The development of embodied intelligence focuses on enabling robots to perceive, understand, and navigate physical environments effectively [11][12]. Section 6: Urban Spatial Intelligence - Empowering Smarter, More Livable Cities - Large models are applied in urban settings to enhance spatial understanding, reasoning, and decision-making for better city management [15][16]. Section 7: Earth Spatial Intelligence (ESI) - Insights into Our Planet - ESI leverages AI and large models to analyze vast amounts of earth observation data, addressing global challenges like climate change and resource management [20][21]. Section 8: Challenges and Prospects - The Future of Spatial Intelligence - Despite significant advancements, challenges remain in spatial reasoning, data integration, and model interpretability, necessitating ongoing research and development [29][30].

Core Viewpoint - Despite significant advancements in multimodal large language models (MLLMs), existing models still lack sufficient alignment with human preferences, primarily due to the focus of current alignment research on specific areas such as reducing hallucination issues. The overall impact of aligning with human preferences on enhancing various capabilities of MLLMs remains uncertain [1]. Group 1: Contributions - Introduction of a new dataset containing 120,000 finely annotated preference comparison pairs, significantly improving scale, sample diversity, annotation granularity, and quality compared to existing resources [5]. - Development of an innovative critique-based reward model that provides better interpretability and more informative feedback than traditional scalar reward mechanisms, achieving superior performance with a 7B model compared to existing 72B models [5]. - Implementation of dynamic reward scaling to optimize the use of high-quality comparison pairs, enhancing data utilization efficiency [5]. - Comprehensive evaluation across 10 dimensions and 27 benchmarks, demonstrating significant and consistent performance improvements in various aspects [5][15]. Group 2: Data Sources and Annotation - Data sources include image datasets from LLaVA-OV, VLfeedback, and others, totaling 10 million samples, with video data primarily from SharedGPT-4-video [6]. - Annotation focuses on three dimensions: usefulness, authenticity, and ethics, with detailed scoring and ranking justifications provided by human experts [7]. Group 3: Performance Evaluation - The proposed model framework shows competitive performance against GPT-4o across multiple benchmarks, with notable improvements in custom benchmarks, validating the effectiveness of the training algorithm's reward signals [10]. - The model's conversational abilities and safety features improved significantly, with average enhancements exceeding 10% in conversation benchmarks and a reduction of unsafe behaviors by at least 50% [17]. Group 4: Future Research Directions - The study emphasizes the potential for further exploration of the dataset's value, particularly in utilizing the rich annotation granularity to enhance current alignment algorithms and address specific benchmark limitations [21]. - Future efforts will focus on leveraging detailed information and advanced optimization techniques to improve MLLM alignment and establish a more universal multimodal learning framework [22].