[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"$fa_nozXLU6fSgyoqa-674yga7mc9_bsts_9koyFmZVWI":3},{"code":4,"msg":5,"data":6},200,"操作成功",{"id":7,"title":8,"content":9,"digest":10,"source":10,"coverPath":11,"thumbsCoverPath":12,"isTop":13,"isShow":14,"baseClick":13,"clickCount":15,"createTime":16,"typeId":17,"isNewest":18,"newsInfoTypeRespVo":19,"voiceUrl":22,"voiceSize":23,"taskId":24,"releaseTime":25,"titleEn":26,"contentEn":27,"voiceUrlEn":28,"taskIdEn":29,"voiceSizeEn":30},1550,"从工具到教练：松鼠Ai刘海涛在GDPS 2025诠释AI教育的生态跃迁","\u003Cp class=\"ql-align-justify\">\u003Cstrong class=\"ql-lineHeight-1-75\" style=\"font-size: 18px; color: rgb(255, 153, 0);\">2025年12月12日至14日，全球开发者先锋大会暨国际具身智能技能大赛（GDPS 2025）在上海张江科学会堂盛大举行。这场为期三天的盛会，不仅是一场聚焦“代码走向实体”的具身智能巅峰竞技，更是一次全景式检阅人工智能全栈技术发展的行业风向标。大会汇聚国内外近百家开源组织、超500家企业机构、300余位重磅嘉宾，吸引了近五万人次线下参会，其规模与影响力堪称年度开发者生态的收官盛典。\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">在大会核心研讨环节《AI全栈技术探索：从算力基建到大模型应用》专题会议上，松鼠Ai智能老师技术VP刘海涛作为AI教育领域的代表受邀出席并发表主题演讲。他从教育的高复杂性、高交互性的垂直场景出发，系统阐释了松鼠Ai如何构建从底层算力、数据生态、混合专家模型到多智能体协同应用的全栈技术体系，为行业揭示了AI技术与教育规律深度融合的实践路径与未来图景。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cstrong style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">01构建教育专属的全栈智能技术生态\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">在人工智能技术浪潮中，教育领域因其紧密关联认知发展规律、情感互动与长期成效验证，常被视作最具挑战的应用场景之一。刘海涛在演讲中强调，要攻克这一难题，不能仅依赖通用模型的外围适配，而必须构建一个闭环自主、深度融入教育逻辑的全栈技术生态。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">刘海涛介绍，松鼠Ai的实践首先立足于对高质量、多模态教育数据的系统化构建与治理。该体系不仅涵盖7500亿Token的学科文本、全维度纳米级知识图谱与海量讲解视频，更积累了来自超过4300万学生的百亿级学习行为、交互过程乃至情感状态数据，形成了规模与深度并重的“教育数据护城河”。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2026\u002F03\u002Fhistory\u002F225be4b2b3e048e88b53cc6cba19475a.png\" width=\"668\" height=\"null\" style=\"display: block; margin: auto;\">\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">以此为基础，松鼠Ai自主研发了混合专家（MoE）架构的多模态智适应教育大模型，能够真正实现规模化、个性化并存的智能教育生态。刘海涛指出，技术的顶层设计关键在于引入“教育人工智能体”框架，通过部署具备记忆、规划、工具调用与执行能力的多个AI Agent，系统从一个被动的问答工具，演进为能主动诊断学情、动态规划路径、实施精准干预与情感激励的“模拟教学组织”。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">这一从数据、模型到智能应用的全栈布局，离不开对算力基础设施的精细规划与持续投入，从而保障了从大规模预训练到高并发实时推理的稳定与高效。他强调，该体系的最终目标，正是将“因材施教”这一教育理想，转化为可标准化交付、可规模化复制的技术现实。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cstrong style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">02以“精准诊断”与“启发式对话”重塑学习范式\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">全栈技术的价值，最终必须通过解决实际教学中的深层次痛点来验证。刘海涛通过两个生动的功能场景，展示了技术如何突破表层对错的判断，直指学习过程的本质。一方面，在“错因分析与追根溯源”场景中，系统能够借助专项OCR技术精准识别学生手写笔迹与演算过程，并依托大模型深入解析其思维链路，从而将一道基础数学应用题的错解，归因于数十种可能的认知偏差，进而自动关联至知识图谱中的早期薄弱环节，真正实现“精准诊断、根源巩固”。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">另一方面，在引导式学习的“AI苏格拉底”教学场景下， AI并不会直接告诉学生答案，而是通过举例、引导学生一步步提升观察、归纳和分辨知识的能力，培养学生学习的思维。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2026\u002F03\u002Fhistory\u002F7e85cd74913a4aa3b60a4212af614ba9.png\" width=\"662\" height=\"null\" style=\"display: block; margin: auto;\">\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">刘海涛指出，这种能力的背后，离不开高精度、且经过教育专家双重校验的标注数据，以及融合了教育学、认知心理学原理的提示工程与思维链技术。他还强调，松鼠Ai的教育大模型核心定位并非单纯的“知识库”，而是内化了教学方法与认知规律的“专业AI教练”。这些经过长期、大规模实证验证的落地应用，正推动学习从“被动接受”转向“主动探索”，助力学生建立起“发现问题-解决问题”的良性成长循环。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">在全球开发者先锋大会这一聚焦前沿与落地的舞台上，松鼠Ai的实践昭示着中国AI正从工具化应用迈向深入产业规律的“深水区”。其以全栈自研与教育逻辑深度融合所构建的“中国方案”，不仅为教育变革提供了可复制的范式，更彰显了中国科技在复杂垂直领域，以技术深度赋能于人、定义未来的引领姿态。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"color: rgb(136, 136, 136);\">【新闻来源】网易 \u003C\u002Fspan>\u003Ca href=\"https:\u002F\u002Fwww.163.com\u002Fdy\u002Farticle\u002FKHLL9MDS0527VF2P.html \" rel=\"noopener noreferrer\" target=\"_blank\" style=\"color: rgb(136, 136, 136);\"> https:\u002F\u002Fwww.msn.cn\u002Fzh-cn\u002Fnews\u002Fother\u002Far-AA1RTDQy?ocid=BingHp01&cvid=6936317f054647a2afcd53fafcde084a&ei\u003C\u002Fa>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"color: rgb(136, 136, 136);\">（本网转发此文章，旨在为读者提供更多的信息资讯，所涉内容不构成投资、消费建议。文章事实如有疑问，请与有关方核实，文章观点非本网观点，仅供读者参考。）\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>","","https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F12\u002F4ea889ec049e4525bbc5e8d6c362b3d6\u002F教育生态.jpg","https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F12\u002Fthumbs\u002F4ea889ec049e4525bbc5e8d6c362b3d6\u002F教育生态.jpg",0,1,47,"2025-12-29 09:14",2,false,{"id":17,"name":20,"enName":21},"芯位视野","Xinwei Vision","https:\u002F\u002Fxinwei-dev-test.oss-cn-shenzhen.aliyuncs.com\u002Fintelligent\u002Faudio%3A502af46f-b666-45e3-b371-e10737d43661%3A0.wav?Expires=1766993216&OSSAccessKeyId=LTAI5tNvY2RkKjZw4LLWsrPK&Signature=%2Bag97k5GmeoTxH%2FvFrY3cQMWi2g%3D",8429758,"502af46f-b666-45e3-b371-e10737d43661","2025-12-29 09:10","From Tool to Coach: Liu Haitao of Squirrel Ai Explains the Ecological Transition of AI Education at GDPS 2025","\u003Cp class=\"ql-align-justify\">\u003Cstrong class=\"ql-lineHeight-1-75\" style=\"font-size: 18px; color: rgb(255, 153, 0);\">December 12 to 14, 2025, the Global Developer Pioneer Conference and International Embodied Intelligence Skills Competition (GDPS 2025) was grandly held at the Zhangjiang Science Convention Center in Shanghai. This three-day event not only served as a top-level competition focusing on \"code moving towards physical entities,\" but also acted as an overall review of the development of artificial intelligence full-stack technology, serving as an industry trend indicator. The conference brought together nearly 100 open-source organizations from around the world, over 500 enterprises and institutions, and more than 300 prominent guests, attracting nearly 50,000 offline participants. Its scale and influence can be considered the closing ceremony of the annual developer ecosystem.\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">In the core discussion session \"AI Full-Stack Technology Exploration: From Computing Infrastructure to Large Model Applications,\" Liu Haitao, VP of Squirrel Ai's intelligent teacher technology, was invited as a representative of the AI education field to deliver a keynote speech. He systematically explained how Squirrel Ai constructs a full-stack technology system from the underlying computing power, data ecology, hybrid expert models to multi-intelligent agent collaboration applications, revealing the practical path and future vision of the deep integration of AI technology with educational laws for the industry.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cstrong style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">01 Building an Education-Specific Full-Stack Intelligent Technology Ecosystem\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">In the wave of artificial intelligence technology, the education sector is often regarded as one of the most challenging application scenarios due to its close connection with cognitive development rules, emotional interaction, and long-term effectiveness verification. Liu Haitao emphasized in his speech that to tackle this challenge, it is not enough to rely solely on external adaptation of general models, but rather must build a closed-loop self-contained, deeply integrated full-stack technology ecosystem into educational logic.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">Liu Haitao introduced that Squirrel Ai's practice first focuses on the systematic construction and governance of high-quality, multimodal educational data. This system not only includes 75 billion Tokens of subject texts, comprehensive nanoscale knowledge graphs, and massive lecture videos, but also has accumulated hundreds of billions of learning behavior, interaction processes, and even emotional state data from over 43 million students, forming an \"education data moat\" that emphasizes both scale and depth.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2026\u002F03\u002Fhistory\u002F225be4b2b3e048e88b53cc6cba19475a.png\" width=\"668\" height=\"null\" style=\"display: block; margin: auto;\">\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">Based on this, Squirrel Ai independently developed a multimodal intelligent adaptive education large model with a mixed-expert (MoE) architecture, capable of truly achieving a scalable and personalized intelligent education ecosystem. Liu Haitao pointed out that the key to the technical top-level design lies in introducing the \"educational artificial intelligence entity\" framework, by deploying multiple AI agents with memory, planning, tool calling, and execution capabilities, the system evolves from a passive question-and-answer tool into an \"simulated teaching organization\" that can actively diagnose learning conditions, dynamically plan paths, implement precise interventions, and provide emotional motivation.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">This full-stack layout from data, models to intelligent applications relies on meticulous planning and continuous investment in computing infrastructure, thereby ensuring stability and efficiency from large-scale pre-training to high-concurrency real-time reasoning. He emphasized that the ultimate goal of this system is to transform the educational ideal of \"teaching according to individual aptitude\" into a standardized, scalable technological reality.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cstrong style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">02 Re-shaping the Learning Paradigm with \"Precise Diagnosis\" and \"Heuristic Dialogue\"\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">The value of full-stack technology must ultimately be verified by solving deep-seated pain points in actual teaching. Liu Haitao demonstrated through two vivid functional scenarios how technology can go beyond surface-level correct or incorrect judgments to directly address the essence of the learning process. On one hand, in the \"error cause analysis and root cause tracing\" scenario, the system can accurately identify students' handwritten scripts and calculation processes using specialized OCR technology, and deeply analyze their thinking chains based on large models, thus attributing the wrong solution of a basic math application problem to dozens of possible cognitive biases, automatically connecting to early weak links in the knowledge graph, truly achieving \"precise diagnosis, root consolidation.\"\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">On the other hand, in the \"AI Socrates\" teaching scenario of guided learning, the AI does not directly tell students the answer, but instead guides them step by step through examples, helping them improve their ability to observe, summarize, and distinguish knowledge, cultivating students' learning thinking.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2026\u002F03\u002Fhistory\u002F7e85cd74913a4aa3b60a4212af614ba9.png\" width=\"662\" height=\"null\" style=\"display: block; margin: auto;\">\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">Liu Haitao pointed out that this capability is inseparable from high-precision, educationally validated annotated data, as well as prompt engineering and thought chain technologies that integrate educational theory and cognitive psychology principles. He also emphasized that the core positioning of Squirrel Ai's educational large model is not merely a \"knowledge base,\" but rather a \"professional AI coach\" that internalizes teaching methods and cognitive laws. These proven applications, which have been tested through long-term and large-scale field trials, are driving learning from \"passive acceptance\" to \"active exploration,\" helping students establish a virtuous cycle of \"identifying problems - solving problems.\"\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">At the Global Developer Pioneer Conference, which focuses on cutting-edge and practical applications, Squirrel Ai's practice demonstrates that China's AI is moving from tool-based applications into the \"deep waters\" of in-depth industrial laws. Its \"Chinese solution\" built by full-stack independent research and deep integration with educational logic not only provides a replicable model for educational transformation but also highlights the leading posture of Chinese technology in complex vertical fields, empowering people through technological depth and defining the future.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"color: rgb(136, 136, 136);\">[News Source] NetEase \u003C\u002Fspan>\u003Ca href=\"https:\u002F\u002Fwww.163.com\u002Fdy\u002Farticle\u002FKHLL9MDS0527VF2P.html \" rel=\"noopener noreferrer\" target=\"_blank\" style=\"color: rgb(136, 136, 136);\"> https:\u002F\u002Fwww.msn.cn\u002Fzh-cn\u002Fnews\u002Fother\u002Far-AA1RTDQy?ocid=BingHp01&cvid=6936317f054647a2afcd53fafcde084a&ei\u003C\u002Fa>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"color: rgb(136, 136, 136);\">（This article is reprinted by our site to provide readers with more information and news. The content involved does not constitute investment or consumption advice. If there are any questions about the facts of the article, please verify with the relevant parties. The views expressed in the article are not the views of our site, and are for reference only.）\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>","https:\u002F\u002Fxinwei-dev-test.oss-cn-shenzhen.aliyuncs.com\u002Fintelligent\u002Faudio%3A27c7b8e2-d81e-4c26-829b-18042449fa08%3A0.wav?Expires=1774838435&OSSAccessKeyId=LTAI5tNvY2RkKjZw4LLWsrPK&Signature=Jjrsr1q5EWt4g2wyvLsyPXaWD%2BI%3D","27c7b8e2-d81e-4c26-829b-18042449fa08",12219034]