作者 | 哈喽WoW君 编辑 | 大模型之心Tech 原文链接： https://zhuanlan.zhihu.com/p/1939744959143086058 点击下方 卡片 ，关注" 大模型之心Tech "公众号 戳我-> 领取大模型巨卷干货 >> 点击进入→ 大模型没那么大Tech技术交流群 本文只做学术分享，如有侵权，联系删文 ，自动驾驶课程学习与技术交流群事宜，也欢迎添加小助理微信AIDriver004做进一步咨询 一、主流AI AGENT框架 当前主流的AI Agent框架种类繁多，各有侧重，适用于不同的应用场景。目前收集了几个主流并且典型Agent框架，先给出本文描述的有哪些框架图表。 | 框架 | 描述 | 适用场景 | | --- | --- | --- | | LangGraph | 基于LangChain搭建的状态驱动的多步骤 Agent | 复杂状态机、审批流 | | AutoGen | 多 Agent 协作、对话式 | 研究报告生成、任务拆解 | | CrewAI | 轻量级"角色扮演"多 Agent | 内容团队、市场分析 | | Smolagents | Hugging Fac ...

Core Insights - The report evaluates the technical capabilities of four major AI model development platforms: Alibaba Cloud's Bailian, Tencent Cloud's Intelligent Agent Development Platform, Kouzi, and Baidu Intelligent Cloud Qianfan, focusing on RAG capabilities, workflow capabilities, and agent capabilities [1][7][8]. RAG Capability Testing - RAG capability testing assesses knowledge enhancement mechanisms, including multi-modal knowledge processing, task complexity adaptation, and interaction mechanism completeness [7][8]. - In text question answering, all platforms demonstrated high accuracy, with over 80% accuracy in multi-document responses, although some platforms showed stability issues during API calls [20][21]. - Baidu Intelligent Cloud Qianfan exhibited stable performance in complex query scenarios for structured data, while Tencent Cloud achieved 100% refusal for out-of-knowledge-base questions [21][23]. - The platforms showed differences in handling refusal and clarification, with Tencent Cloud providing 100% refusals for non-knowledge-base questions [21][22]. Workflow Capability Testing - Workflow capability testing focuses on dynamic parameter extraction, exception rollback, intent recognition, and fault tolerance [35][36]. - The end-to-end accuracy for workflow processes ranged from 61.5% to 93.3%, with Tencent Cloud leading in intent recognition accuracy at 100% [36][37]. - The platforms demonstrated basic usability in workflow systems, but there is room for improvement in complex information processing [38][39]. Agent Capability Testing - Agent capability testing evaluates the ability to call tools, focusing on intent understanding, operational coordination, feedback effectiveness, and mechanism completeness [44][45]. - All platforms achieved high single-tool call completion rates (83%-92%), but multi-tool collaboration and prompt calling showed potential for improvement [48][50]. - Tencent Cloud's Intelligent Agent Development Platform excelled in tool call success rates due to its robust ecosystem and process optimization [49][50].

Core Viewpoint - The discussion around VLA (Vision-Language-Action model) is intensifying, with contrasting opinions on its short-term feasibility and potential impact on the automotive industry [2][12]. Group 1: VLA Technology and Development - The Li Auto i8 is the first vehicle to feature the VLA driver model, positioning it as a key selling point [2]. - Bosch's president for intelligent driving in China, Wu Yongqiao, expressed skepticism about the short-term implementation of VLA, citing challenges in multi-modal data acquisition and training [2][12]. - VLA is seen as an "intelligent enhanced version" of end-to-end systems, aiming for a more human-like driving experience [2][5]. Group 2: Comparison of Driving Technologies - There are two main types of end-to-end technology: modular end-to-end and one-stage end-to-end, with the latter being more advanced and efficient [3][4]. - The one-stage end-to-end model simplifies the process by directly mapping sensor data to control commands, reducing information loss between modules [3][4]. - VLA is expected to outperform traditional end-to-end models by integrating multi-modal capabilities and enhancing decision-making in complex scenarios [5][6]. Group 3: Challenges and Requirements for VLA - The successful implementation of VLA relies on breakthroughs in three key areas: cross-modal feature alignment, world model construction, and dynamic knowledge base integration [7][8]. - Current automotive chips are not designed for AI large models, leading to performance limitations in real-time decision-making [9][11]. - The industry is experiencing a "chip power battle," with companies like Tesla and Li Auto developing their own high-performance AI chips to meet VLA's requirements [11][12]. Group 4: Future Outlook and Timeline - Some industry experts believe 2025 could be a pivotal year for VLA technology, while others suggest it may take 3-5 years for widespread adoption [12][13]. - Initial applications of VLA are expected to be in controlled environments, with broader capabilities emerging as chip technology advances [14]. - Long-term projections indicate that advancements in AI chip technology and multi-modal alignment could lead to significant breakthroughs in VLA deployment by 2030 [14][15].

Core Viewpoint - The essence of AI Agents lies in reconstructing the "cognition-action" loop, iterating on human cognitive processes to enhance decision-making and execution capabilities [1][4][41]. Group 1: Breakthroughs in AI Agents - The breakthrough of large language models (LLMs) is fundamentally about decoding human language, enabling machines to possess near-human semantic reasoning abilities [2]. - AI Agents transform static "knowledge storage" into dynamic "cognitive processes," allowing for more effective problem-solving [4][7]. - The memory system in AI Agents plays a crucial role, with short-term memory handling real-time context and long-term memory encoding user preferences and business rules [10][12][13]. Group 2: Memory and Learning Capabilities - The dual memory mechanism allows AI Agents to accumulate experience, evolving from passive tools to active cognitive entities capable of learning from past tasks [14][15]. - For instance, in customer complaint handling, AI Agents can remember effective solutions for specific complaints, optimizing future responses [15]. Group 3: Tool Utilization - The ability to call tools is essential for AI Agents to expand their cognitive boundaries, enabling them to access real-time data and perform complex tasks [17][20]. - In finance, AI Agents can utilize APIs to gather market data and provide precise investment advice, overcoming the limitations of LLMs [21][22]. - The diversity of tools allows AI Agents to adapt to various tasks, enhancing their functionality and efficiency [26][27]. Group 4: Planning and Execution - The planning module of AI Agents addresses the "cognitive entropy" of complex tasks, enabling them to break down tasks into manageable components and monitor progress [28][30][32]. - After completing tasks, AI Agents can reflect on their planning and execution processes, continuously improving their efficiency and effectiveness [33][35]. Group 5: Impact on Business and Society - AI Agents are redefining the underlying logic of enterprise software, emphasizing collaboration between human intelligence and machine capabilities [36][37]. - The evolution from tools to cognitive entities signifies a major shift in how AI can enhance human productivity and decision-making [39][41]. - As AI technology advances, AI Agents are expected to play significant roles across various sectors, including healthcare and education, driving societal progress [44][45]. Group 6: Practical Applications and Community - The company has developed its own AI Agent and established an AI Innovation Institute to assist enterprises in effectively utilizing AI for cost reduction and efficiency improvement [46][48]. - The institute offers practical tools and methodologies derived from extensive real-world case studies, enabling businesses to integrate AI into their operations [51][58]. - Monthly collaborative learning sessions serve as a reflection mechanism, allowing participants to convert theoretical knowledge into actionable solutions [60][62].

Core Insights - The article discusses the challenges faced by companies in implementing AI, particularly in achieving human-like conversation and high accuracy in AI systems. It highlights the need for effective engineering and project management in AI projects [1][15][18]. Group 1: AI Challenges - AI often struggles with human-like conversation, leading to stiff interactions even when using RAG or knowledge bases [1]. - The accuracy of AI systems is often insufficient, with a typical business requirement being 95% accuracy, while AI may only cover 80% of scenarios [1]. - The Context Paradox suggests that tasks perceived as easy for humans are often harder for AI, while complex tasks can be easier for AI to handle [3][12]. Group 2: Engineering and Project Management - Engineering capabilities are more critical than model complexity in AI projects, as many projects fail due to inadequate engineering and project management [15][18]. - A typical AI project may require extensive documentation, with one SOP potentially needing 5,000 to 10,000 words of prompts, leading to a total of 250,000 to 500,000 words for complex projects [17]. - The majority of challenges in AI projects stem from data engineering, which constitutes about 80% of the difficulty [19]. Group 3: Specialization and Data - Specialized AI solutions tailored to specific industries outperform general-purpose AI assistants, as they can better understand industry-specific language and needs [20][22]. - Data is becoming a crucial competitive advantage, as technical barriers diminish; companies must focus on leveraging unique data to create a moat [26][28]. - Companies should prioritize making AI capable of handling large volumes of noisy, real-world data rather than spending excessive time on data cleaning [26]. Group 4: Production Challenges - Transitioning from pilot projects to production environments is significantly more challenging, requiring careful design from the outset [29][31]. - Speed in deployment is more important than perfection; early user feedback is essential for iterative improvement [33][36]. - Companies must be cautious about the asymmetry in AI projects, where initial successes in demos may not translate to production success [30]. Group 5: Accuracy and Observability - Achieving 100% accuracy in AI is nearly impossible; companies should focus on managing inaccuracies and establishing robust monitoring systems [46][50]. - Observability and the ability to trace errors back to their sources are critical for continuous improvement in AI systems [47][50]. - Companies should develop a feedback loop to ensure that inaccuracies are addressed and corrected in future iterations [51][52].

Core Insights - Glean, an AI startup, has raised $150 million in funding, achieving a valuation of $7.2 billion, significantly up from $4.6 billion in September 2022 [2][3] - The funding round was led by Wellington Management, with participation from existing investors like Sequoia Capital, indicating strong confidence in Glean's growth trajectory [3] - Glean aims to use the new funds to accelerate product development, expand its partner ecosystem, and pursue international growth [3] Company Overview - Founded in 2019, Glean started with enterprise search and has since developed products like Glean Assistant and Glean Agents, leveraging RAG technology for AI-driven enterprise search [4][6] - Glean Search allows employees to find data across internal documents and the web, while Glean Assistant automates daily tasks and provides data analysis through natural language queries [6] - Glean Agents enables the creation of AI agents for tasks like debugging software code, supporting over 100 million agents annually [6] Market Position and Growth - Glean's business model reflects a broader shift in the enterprise AI sector, moving from pilot projects to widespread deployment of autonomous agents [7] - The company has seen rapid revenue growth, with annual recurring revenue (ARR) increasing from $55 million to $100 million [7] - Glean's client base includes Fortune 500 companies like Dell, showcasing its strong market presence [7] AI Implementation in Enterprises - Arvind Jain, Glean's CEO, emphasizes the importance of a robust data infrastructure for effective AI applications, including deep integration with enterprise systems and a solid security framework [8][9] - The challenges of enterprise AI deployment stem from the private and context-dependent nature of enterprise data, requiring an understanding of organizational structure and user roles [9] - Jain suggests that AI entrepreneurs should focus on solving specific business problems rather than starting with AI technology itself, building trust with enterprises through clear value propositions [10]

Core Viewpoint - The article provides a comprehensive comparison of five mainstream LLM application platforms: Dify, Coze, n8n, FastGPT, and RAGFlow, emphasizing the importance of selecting the right platform based on individual needs and use cases [1][2]. Group 1: Overview of LLM Platforms - LLM application platforms significantly lower the development threshold for AI applications, accelerating the transition from concept to product [2]. - These platforms allow users to focus on business logic and user experience innovation rather than repetitive underlying technology construction [3]. Group 2: Platform Characteristics - **n8n**: Known for its powerful general workflow automation capabilities, it allows users to embed LLM nodes into complex automation processes [4]. - **Coze**: Launched by ByteDance, it emphasizes low-code/no-code AI agent development, enabling rapid construction and deployment of conversational AI applications [5]. - **FastGPT**: An open-source AI agent construction platform focused on knowledge base Q&A systems, offering data processing, model invocation, and visual workflow orchestration capabilities [6]. - **Dify**: An open-source LLM application development platform that integrates BaaS and LLMOps concepts, providing a one-stop solution for rapid AI application development and operation [7]. - **RAGFlow**: An open-source RAG engine focused on deep document understanding, specializing in knowledge extraction and high-quality Q&A from complex formatted documents [8][40]. Group 3: Detailed Platform Analysis - **Dify**: Described as a "Swiss Army Knife" of LLM platforms, it offers a comprehensive set of features including RAG pipelines, AI workflows, monitoring tools, and model management [8][10][12]. - **Coze**: Positioned as the "LEGO" of LLM platforms, it allows users to easily create and publish AI agents with a wide range of built-in tools and plugins [21][25]. - **FastGPT**: Recognized for its ability to quickly build high-quality knowledge bases, it supports various document formats and provides a user-friendly interface for creating AI Q&A assistants [33][35]. - **RAGFlow**: Distinguished by its deep document understanding capabilities, it supports extensive data preprocessing and knowledge graph functionalities [40][42]. - **n8n**: A low-code workflow automation tool that connects various applications and services, enhancing business process automation [46][49]. Group 4: User Suitability and Recommendations - For beginners in AI application development, Coze is recommended as the easiest platform to start with [61]. - For businesses requiring automation across multiple systems, n8n's robust workflow capabilities can save significant time [62]. - For building internal knowledge bases or Q&A systems, FastGPT and RAGFlow are suitable options, with FastGPT being lighter and RAGFlow offering higher performance [63]. - For teams with long-term plans to develop scalable enterprise-level AI applications, Dify's comprehensive ecosystem is advantageous [63]. Group 5: Key Considerations for Platform Selection - Budget considerations include the costs of self-hosting open-source platforms versus subscription fees for cloud services [68]. - Technical capabilities of the team should influence the choice of platform, with no-code options like Coze being suitable for those with limited technical skills [68]. - Deployment preferences, such as the need for local data privacy, should also be evaluated [69]. - Core functionality requirements must be clearly defined to select the platform that best meets specific needs [70]. - The sustainability of the platform, including update frequency and community support, is crucial for long-term viability [71]. - Data security and compliance are particularly important for enterprise users, with self-hosted solutions offering greater control over data [72].

Core Viewpoint - The integration of AI in medical imaging is advancing rapidly, with large models evolving from mere tools to core drivers of diagnostic ecosystems, enhancing the workflow of radiologists and addressing challenges in pathology diagnostics [1][2]. Group 1: Development of AI in Medical Imaging - Medical imaging AI models have achieved widespread application in the workflow of radiologists, transitioning from auxiliary diagnostic tools to essential components of the diagnostic ecosystem [1]. - The "Shukun Kun Multi-modal Medical Health Large Model" released by Shukun Technology in April signifies this evolution, enhancing the role of AI in diagnostics [1]. Group 2: Challenges and Solutions in Pathology - Pathology models are considered the "crown jewel" of medical models due to their complexity and diversity, with the first clinical-grade pathology model, "Insight," developed by Tuo Che Future, addressing accuracy and efficiency challenges [2]. - The pathology model addresses long-standing challenges in generalization across hospitals, cancer types, and pathology tasks, simplifying processes and improving diagnostic efficiency [3]. Group 3: Enhancing AI Generalization Performance - AI model generalization is crucial for reliability and stability, with key challenges including insufficient data diversity, model limitations, and the long-tail nature of medical data [4][6]. - Strategies to enhance generalization include expanding data sample diversity, optimizing model training, and iterating models in real clinical environments [6][7]. Group 4: Addressing the Hallucination Problem - The hallucination issue in large models is a significant barrier, with RAG (Retrieval-Augmented Generation) technology proposed as a solution to enhance accuracy by integrating external knowledge [8][9]. - A hybrid approach combining generative and discriminative AI is suggested to mitigate risks in critical decision-making scenarios, ensuring reliable outputs [9]. Group 5: Deployment Trends in Healthcare - Local deployment of AI models is becoming the preferred choice for hospitals due to data privacy and compliance advantages, with integrated solutions like one-box systems gaining traction [10][11]. - One-box systems combine the strengths of general and specialized models, addressing diverse medical needs while ensuring data control [10]. Group 6: Future Trends in Medical AI - The performance of medical large models is surpassing traditional small models, with applications expanding from thousands to over ten thousand hospitals [12]. - The future of medical AI is moving towards multi-modal integration and comprehensive diagnostics, akin to a digital "general practitioner" that synthesizes various patient data for holistic treatment recommendations [12][13].