[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"$f2KXyAf8swI9AkIqXib4oysmbpGRjMLoLpEGGxJx0eeY":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},1423,"首个全自动AI科学家诞生！西湖大学最新成果：性能超越人类SOTA基线183.7%","\u003Cp>\u003Cstrong class=\"ql-lineHeight-1-75\" style=\"font-size: 18px; color: rgb(255, 153, 0);\">人类科学家三年的工作量，如今AI两周就能轻松搞定！\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">最近，来自西湖大学的自然语言处理实验室发布了DeepScientist系统，这也是首个具有完整科研能力，且在无人工干预下，展现出目标导向、持续迭代、渐进式超越人类研究者最先进研究成果的AI科学家系统。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-center\">\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"undefined\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F10\u002F7a9845de3cee4cf9ac5910f40e5a0b47\u002F239.jpg\" width=\"undefined\" height=\"undefined\" style=\"display: block; margin: auto;\" class=\"ql-align-center\">\u003Cp class=\"ql-align-center\">\u003Cspan style=\"color: rgb(187, 187, 187);\" class=\"ql-lineHeight-1-75\">△对比DeepScientist与人类专家的研究进展\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">在AI文本检测任务中，DeepScientist仅用两周时间就实施和验证了超过1000种不同的假设，在此期间取得了相当于人类三年的进展。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">在RAID数据集测试中，DeepScientist设计的方法实现了7.9%的AUROC提升，成功超越了人类现有SOTA方案。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">另外DeepScientist还在智能体失败归因、LLM推理加速等任务上也分别达成了新的SOTA。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"undefined\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F10\u002F3dbb0acf75df400b8b33108834bd9d82\u002F240.jpg\" width=\"undefined\" height=\"undefined\" style=\"display: block; margin: auto;\">\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">下面是更多详细内容介绍。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cstrong class=\"ql-lineHeight-1-75\" style=\"font-size: 18px; color: rgb(255, 153, 0);\">从“科研助理”到“首席科学家”：AI科研模式的变革\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">过去的AI Scientist系统，如果不给定一个清晰明了的科研目标，就很容易陷入对现有知识的机械组合与无效试探的窠臼中，最终形成的科研产出在人类专家看来缺乏焦点，科学价值不高。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">它们更像是能力超群的科研助理，而不是能独立指引方向的科学家。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">DeepScientist的出现改变了这一现状，它不再等待人类告诉它“研究什么”，而是开始主动思考“什么值得研究”，它可以：\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">主动识别前沿研究的根本性局限，提出全新的科学构想以解决局限性问题，自动编写代码、执行实验、设计分析实验，整理实验结果，撰写结构完整的科研论文，开源可重现代码。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">简而言之，这种从“随机发现”到“长期主动式探索”的角色转变，标志着AI已经正式涉足以往只有顶尖人类心智才能胜任的、最具创造性的科学发现过程。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cstrong class=\"ql-lineHeight-1-75\" style=\"font-size: 18px; color: rgb(255, 153, 0);\">DeepScientist的核心机制\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">DeepScientist的核心目标是在一个给定的总研究预算内，最大化有价值的科学发现（Progress Findings）。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">它首先将混乱、依赖灵感的科学发现过程形式化为一个严谨、目标驱动的分层贝叶斯优化问题，其目标是从所有可能的候选研究空间中，找到一个最优方法，使一个未知且评估成本极高的真实科学价值函数最大化。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"undefined\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F10\u002Fce5143debd0f4e25a062432e355b419d\u002F241.jpg\" width=\"undefined\" height=\"undefined\" style=\"display: block; margin: auto;\">\u003Cp class=\"ql-align-center\">\u003Cspan class=\"ql-lineHeight-1-75\" style=\"color: rgb(187, 187, 187);\">△DeepScientist的自主科学发现闭环流程图\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">具体而言，DeepScientist基于多智能体协同策略，围绕一个三层级的评估循环推进。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">每个层级代表了对一个科研想法（Finding）进行验证的不同保真度（Fidelity）和成本（Cost），系统在每一轮迭代中，都基于其不断增长的“经验库（Findings Memory）”产出新假设和做出资源分配决策。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">高层级（即具有高保真度）的信息，其价值是以前一层级（低保真度）的信息为条件的，而一个想法能否在最终的高保真度评估中成功，依赖于它在低保真度实验中的表现。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">在每一个层级中，只有展现出价值的科研产物才会被送入下一层级以提供更多资源用来进一步探索，否则被存储到“Findings Memory”中用于给后续的探索提供信息。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">这种分层方法，确保了计算资源能够被精准地、动态地分配给在当前认知下最具潜力的研究方向，从而在有限的预算内最大化科学发现的效率。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cstrong class=\"ql-lineHeight-1-75\" style=\"font-size: 18px; color: rgb(255, 153, 0);\">AI两周完成三年科研进展，全面超越人类专家\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">为验证DeepScientist的研究能力，研究人员将DeepScientist应用在三个当前AI研究的最前沿领域：智能体失败归因、LLM推理加速与AI文本检测 。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">这些任务无一例外都竞争激烈、备受社区关注，且技术基准极高，其挑战的人类研究成果均为近期在ICLR、ICML和ACL等顶级会议上发布的最新SOTA方法。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"undefined\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F10\u002F6bfca29e2e5d4de5b9811e6a0c015477\u002F242.jpg\" width=\"undefined\" height=\"undefined\" style=\"display: block; margin: auto;\">\u003Cp class=\"ql-align-center\">\u003Cspan class=\"ql-lineHeight-1-75\" style=\"color: rgb(187, 187, 187);\">△三个研究任务选取的SOTA方法\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">其中，在AI文本检测任务里，DeepScientist在无人干预的情况下，仅用两周时间，就自主完成了相当于人类科学家三年的进展。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">在此期间，DeepScientist自主生成了2472个独特的研究想法，并对其中600个具有科学价值的假设进行了代码实现和实验验证。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">最终，DeepScientist在RAID数据集上取得了7.9%的AUROC提升，同时将推理延迟降低了190%，展示出超越现有人类SOTA的卓越性能。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">DeepScientist的突破性进展并不仅限于AI文本检测领域，它在多个不同的前沿任务上都展示了超越人类专家的科学发现能力，其中一个典型的例子是在“智能体失败归因”这一高度复杂的任务上。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"undefined\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F10\u002F57716bda706f4cfca156550df35825ce\u002F243.jpg\" width=\"undefined\" height=\"undefined\" style=\"display: block; margin: auto;\" class=\"ql-align-center\">\u003Cp class=\"ql-align-center\">\u003Cspan class=\"ql-lineHeight-1-75\" style=\"color: rgb(187, 187, 187);\">△DeepScientist在多任务中超越人工最优方法\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">面对现有方法难以进行有效因果推理的困境，DeepScientist自主构想并提出了名为A2P（Abduction-Action-Prediction）的全新方法，其核心创新在于将失败归因从简单的模式识别提升到了结构化的因果推理层面。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">最终，该方法在Who&When基准测试的“算法生成”任务中取得了47.46分，性能相较于人类专家的SoTA基线大幅提升了183.7% 。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">上述成就充分证明了DeepScientist不仅能实现单点突破，更能创造出具有持续影响力的科学成果，其泛化能力和系统性创新能力足以在多个前沿领域稳定地推动技术边界。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">此外，在自动化科学发现领域，实验的成功率常常不足1%。这个数字虽然残酷，却真实地反映了科学探索的高度不确定性。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"undefined\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F10\u002Fb277b6f0376c4127af9dab12fa67e92d\u002F244.jpg\" width=\"undefined\" height=\"undefined\" style=\"display: block; margin: auto;\">\u003Cp class=\"ql-align-center\">\u003Cspan class=\"ql-lineHeight-1-75\" style=\"color: rgb(187, 187, 187);\">△DeepScientist的研究统计结果\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">不同于依赖大规模随机试错的方法，DeepScientist通过形式化的分层贝叶斯优化机制，在“利用已有成果”与“探索未知可能性”之间灵活平衡，能够在庞大的假设空间中智能筛选出最具潜力的研究方向。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">在探索过程中，DeepScientist不仅能高效执行大规模实验，还会把成功与失败的结果都视作宝贵经验，用来指导后续的决策。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">这种记忆驱动、目标导向的迭代流程，使其能够自主运行数月之久，在浩瀚的可能性空间中持续寻找突破口，不断推动科学发现的进程。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">换句话说，如果没有精细化的策略与结构化的反馈机制，这类探索几乎不可能取得成果，而 DeepScientist 的设计恰恰保证了，即便面对极低的成功率，它也能在闭环学习中稳步积累成果，展现出远超暴力搜索系统的持续进化能力与科学发现潜力。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cstrong class=\"ql-lineHeight-1-75\" style=\"font-size: 18px; color: rgb(255, 153, 0);\">“科学发现缩放定律”？用算力驱动创新\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">在分析实验中，如下图所示，研究团队发现：当并行 GPU 资源从1枚扩展到16枚时， DeepScientist每周产出的前沿级科学发现数量从0项跃升至11项，几乎呈现出理想的线性增长。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">这意味着，科学突破不再只是依赖少数灵光一现，而是可以像训练大模型一样，通过系统化地增加计算资源来“规模化生产”。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">这种趋势正在推动科研范式的转变：从过去依靠“人力密集型”投入，逐步走向“计算密集型”驱动，为解决人类面临的重大科学挑战，开辟了一条全新且可加速的路径。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cstrong class=\"ql-lineHeight-1-75\" style=\"font-size: 18px; color: rgb(255, 153, 0);\">未来展望：开启人机协同的科研新范式\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">DeepScientist 的成功并不意味着AI将取代科学家，而是预示着一个全新的人机协同科研范式的到来。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">在这个范式中，人类研究者的角色将从繁重的试错和实验中解放出来，专注于提出真正有价值的科学问题、设定具有前瞻性的研究方向，并进行最终的综合与判断。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">而 AI，将作为一台不知疲倦、并行扩展的“科学探索引擎”，在人类智慧的引领下，以前所未有的速度和广度持续探索科学的无人区。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">为了推动这一范式的到来，研究团队将开源DeepScientist的核心系统与全部实验日志，希望通过开放共享的方式，激发全球科研社区的创新力量，共同加速 AI Scientis的发展，迎接从基础物理到新药研发等人类重大挑战的突破时刻。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">西湖大学自然语言处理实验室期待与更多研究团队携手促进自动化科学发现的进步。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">团队现已开放了免费的DeepScientist服务申请，希望与科研社区共同建设一个更加高效的科学发现新范式，使其能够真正加速人类科学发现的历程。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cstrong class=\"ql-lineHeight-1-75\" style=\"font-size: 18px;\">西湖大学自然语言处理实验室（WestlakeNLP）成立于2018年9月，由张岳教授领导。\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">张岳教授毕业于牛津大学，获博士学位，现任西湖大学工程学院副院长，著有剑桥大学出版社出版的《自然语言处理》一书，并担任过EMNLP 2022等多个顶级NLP会议的程序委员会主席\u003C\u002Fspan>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">该实验室目前专注于语言模型推理、泛化和通用人工智能以及自然语言处理的基础与应用研究，探索通用人工智能的实现路径，推动 AI Scientist（AI科学家）的发展，使其能够真正参与并加速科学发现，促进人类科学的持续进步。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">围绕这一愿景，WestlakeNLP近期也系统地撰写了AI Scientist方向的观点文章与综述论文，希望为该领域的发展提供更加全面的思考与参考。\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>\u003Cspan style=\"color: rgb(187, 187, 187);\">【新闻来源】量子位 | 公众号 QbitAI  \u003C\u002Fspan>\u003Ca href=\"https:\u002F\u002Fwww.thepaper.cn\u002FnewsDetail_forward_31744019\" rel=\"noopener noreferrer\" target=\"_blank\" style=\"color: rgb(187, 187, 187);\">https:\u002F\u002Fwww.thepaper.cn\u002FnewsDetail_forward_31744019\u003C\u002Fa>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"color: rgb(187, 187, 187);\">（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\u002F10\u002Fbf814d02a28047c2bd9059799dda723c\u002FAI领域.jpg","https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F10\u002Fthumbs\u002Fbf814d02a28047c2bd9059799dda723c\u002FAI领域.jpg",0,1,97,"2025-10-14 10:21",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%3Ac1f01b5d-d513-4460-8e4c-a45fcfb875a5%3A0.wav?Expires=1760412677&OSSAccessKeyId=LTAI5tNvY2RkKjZw4LLWsrPK&Signature=NXRcLZ8IwiGtJV9VhjxykRa3JSQ%3D",16544594,"c1f01b5d-d513-4460-8e4c-a45fcfb875a5","2025-10-14 10:06","First fully automatic AI scientist is born! West Lake University's latest achievement: performance exceeds human SOTA baseline by 183.7%","\u003Cp>\u003Cstrong class=\"ql-lineHeight-1-75\" style=\"font-size: 18px; color: rgb(255, 153, 0);\">The workload of a human scientist for three years can now be easily completed by AI in two weeks!\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">Recently, the Natural Language Processing Laboratory at West Lake University released the DeepScientist system, which is also the first AI scientist system with complete research capabilities and shows goal-oriented, continuous iteration, and progressive surpassing of the most advanced research results of human researchers without any manual intervention.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-center\">\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"undefined\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F10\u002F7a9845de3cee4cf9ac5910f40e5a0b47\u002F239.jpg\" width=\"undefined\" height=\"undefined\" style=\"display: block; margin: auto;\" class=\"ql-align-center\">\u003Cp class=\"ql-align-center\">\u003Cspan style=\"color: rgb(187, 187, 187);\" class=\"ql-lineHeight-1-75\">△ Comparison of DeepScientist's research progress with that of human experts\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">In the AI text detection task, DeepScientist implemented and verified over 1000 different hypotheses within two weeks, achieving the equivalent of three years of progress by humans during this period.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">In the RAID dataset test, the method designed by DeepScientist achieved a 7.9% AUROC improvement, successfully surpassing the current human SOTA solution.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">In addition, DeepScientist also achieved new SOTAs in tasks such as intelligent agent failure attribution and LLM reasoning acceleration.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"undefined\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F10\u002F3dbb0acf75df400b8b33108834bd9d82\u002F240.jpg\" width=\"undefined\" height=\"undefined\" style=\"display: block; margin: auto;\">\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">Below are more detailed content introductions.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cstrong class=\"ql-lineHeight-1-75\" style=\"font-size: 18px; color: rgb(255, 153, 0);\">From \"Research Assistant\" to \"Chief Scientist\": The Transformation of the AI Research Model\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">Past AI Scientist systems, if not given a clear and explicit research objective, tend to fall into the trap of mechanical combinations and ineffective trials of existing knowledge, ultimately resulting in research outputs that lack focus and have low scientific value from the perspective of human experts.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">They are more like highly capable research assistants rather than scientists who can independently guide the direction.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">The emergence of DeepScientist has changed this situation. It no longer waits for humans to tell it \"what to study,\" but instead starts to think proactively \"what is worth studying.\" It can:\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">Proactively identify the fundamental limitations of cutting-edge research, propose new scientific concepts to solve these limitations, automatically write code, perform experiments, design analysis experiments, organize experimental results, and write structured research papers, open-source reproducible code.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">In short, this role shift from \"random discovery\" to \"long-term active exploration\" marks that AI has officially entered the most creative scientific discovery process that only top human minds could previously handle.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cstrong class=\"ql-lineHeight-1-75\" style=\"font-size: 18px; color: rgb(255, 153, 0);\">Core Mechanism of DeepScientist\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">The core goal of DeepScientist is to maximize valuable scientific discoveries (Progress Findings) within a given total research budget.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">It first formalizes the chaotic, inspiration-dependent scientific discovery process into a rigorous, goal-driven hierarchical Bayesian optimization problem, aiming to find an optimal method from all possible candidate research spaces to maximize a real scientific value function that is unknown and highly costly to evaluate.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"undefined\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F10\u002Fce5143debd0f4e25a062432e355b419d\u002F241.jpg\" width=\"undefined\" height=\"undefined\" style=\"display: block; margin: auto;\">\u003Cp class=\"ql-align-center\">\u003Cspan class=\"ql-lineHeight-1-75\" style=\"color: rgb(187, 187, 187);\">△ Autonomous scientific discovery closed-loop flowchart of DeepScientist\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">Specifically, DeepScientist is based on a multi-agent collaborative strategy, advancing around a three-tiered evaluation cycle.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">Each level represents different fidelity (Fidelity) and cost (Cost) for verifying a research idea (Finding). In each iteration, the system produces new hypotheses and makes resource allocation decisions based on its growing \"experience database (Findings Memory).\"\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">The value of high-fidelity information (the higher level) is conditional on the low-fidelity information (the lower level), and whether an idea can succeed in the final high-fidelity evaluation depends on its performance in low-fidelity experiments.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">In each level, only research products that demonstrate value are passed to the next level to receive more resources for further exploration, otherwise they are stored in the \"Findings Memory\" to provide information for subsequent exploration.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">This hierarchical approach ensures that computing resources are accurately and dynamically allocated to the most promising research directions under the current understanding, thereby maximizing the efficiency of scientific discovery within a limited budget.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cstrong class=\"ql-lineHeight-1-75\" style=\"font-size: 18px; color: rgb(255, 153, 0);\">AI completes three years of research progress in two weeks, comprehensively surpassing human experts\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">To verify DeepScientist's research capabilities, researchers applied DeepScientist to three of the most cutting-edge areas in current AI research: intelligent agent failure attribution, LLM reasoning acceleration, and AI text detection.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">These tasks are all intensely competitive, widely concerned by the community, and have extremely high technical benchmarks, with their challenged human research achievements being the latest SOTA methods recently published at top conferences such as ICLR, ICML, and ACL.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"undefined\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F10\u002F6bfca29e2e5d4de5b9811e6a0c015477\u002F242.jpg\" width=\"undefined\" height=\"undefined\" style=\"display: block; margin: auto;\">\u003Cp class=\"ql-align-center\">\u003Cspan class=\"ql-lineHeight-1-75\" style=\"color: rgb(187, 187, 187);\">△ SOTA methods selected for three research tasks\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">In the AI text detection task, DeepScientist autonomously completed the equivalent of three years of progress by human scientists in just two weeks without any intervention.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">During this time, DeepScientist generated 2472 unique research ideas and implemented and verified 600 hypotheses with scientific value through code.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">Finally, DeepScientist achieved a 7.9% AUROC improvement on the RAID dataset and reduced the reasoning delay by 190%, demonstrating superior performance beyond the current human SOTA.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">DeepScientist's breakthroughs are not limited to the field of AI text detection. It has demonstrated scientific discovery capabilities surpassing human experts in multiple different cutting-edge tasks, one typical example being the highly complex task of \"intelligent agent failure attribution.\"\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"undefined\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F10\u002F57716bda706f4cfca156550df35825ce\u002F243.jpg\" width=\"undefined\" height=\"undefined\" style=\"display: block; margin: auto;\" class=\"ql-align-center\">\u003Cp class=\"ql-align-center\">\u003Cspan class=\"ql-lineHeight-1-75\" style=\"color: rgb(187, 187, 187);\">△ DeepScientist surpasses human optimal methods in multiple tasks\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">Facing the difficulties of existing methods in effectively performing causal reasoning, DeepScientist independently conceived and proposed a new method called A2P (Abduction-Action-Prediction), whose core innovation lies in elevating failure attribution from simple pattern recognition to structured causal reasoning.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">Finally, this method achieved 47.46 points on the \"algorithm generation\" task in the Who&When benchmark test, significantly improving the performance compared to the human expert SoTA baseline by 183.7%.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">These achievements fully prove that DeepScientist not only achieves single-point breakthroughs but also creates scientifically impactful results with sustained influence. Its generalization ability and systematic innovation capability are sufficient to steadily push the technological boundaries in multiple cutting-edge fields.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">Additionally, in the field of automated scientific discovery, the success rate of experiments is often less than 1%. This number, although harsh, truly reflects the high uncertainty of scientific exploration.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"undefined\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F10\u002Fb277b6f0376c4127af9dab12fa67e92d\u002F244.jpg\" width=\"undefined\" height=\"undefined\" style=\"display: block; margin: auto;\">\u003Cp class=\"ql-align-center\">\u003Cspan class=\"ql-lineHeight-1-75\" style=\"color: rgb(187, 187, 187);\">△ Research statistics of DeepScientist\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">Unlike methods relying on large-scale random trial and error, DeepScientist flexibly balances between \"utilizing existing achievements\" and \"exploring unknown possibilities\" through a formalized hierarchical Bayesian optimization mechanism, intelligently screening out the most promising research directions in a vast hypothesis space.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">During the exploration process, DeepScientist not only efficiently executes large-scale experiments but also treats both successful and failed results as valuable experiences to guide subsequent decisions.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">This memory-driven, goal-oriented iterative process enables it to run autonomously for months, continuously seeking breakthroughs in the vast possibility space and constantly advancing the process of scientific discovery.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">In other words, without refined strategies and structured feedback mechanisms, such exploration would almost certainly fail. However, DeepScientist's design ensures that even with extremely low success rates, it can steadily accumulate results in a closed-loop learning environment, showcasing a continuous evolutionary capability and scientific discovery potential far exceeding that of brute-force search systems.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cstrong class=\"ql-lineHeight-1-75\" style=\"font-size: 18px; color: rgb(255, 153, 0);\">\"The Law of Scientific Discovery Scaling\"? Driving Innovation with Computing Power\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">In the analysis experiment, as shown in the figure below, the research team found that when parallel GPU resources were expanded from 1 to 16, the number of frontier-level scientific discoveries produced by DeepScientist per week increased from 0 to 11, almost showing ideal linear growth.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">This means that scientific breakthroughs are no longer dependent on a few moments of inspiration, but can be \"scaled production\" like training large models by systematically increasing computing resources.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">This trend is driving a transformation in the research paradigm: from past \"labor-intensive\" inputs to gradually moving towards \"computation-intensive\" driven approaches, opening up a brand-new and accelerated path to address major scientific challenges facing humanity.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cstrong class=\"ql-lineHeight-1-75\" style=\"font-size: 18px; color: rgb(255, 153, 0);\">Future Outlook: Opening a New Paradigm of Human-Machine Collaboration in Research\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">The success of DeepScientist does not mean that AI will replace scientists, but rather heralds the arrival of a new human-machine collaborative research paradigm.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">In this paradigm, the role of human researchers will be liberated from tedious trial and error and experiments, focusing on proposing truly valuable scientific questions, setting forward-looking research directions, and conducting final synthesis and judgment.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">Meanwhile, AI will act as an untiring, scalable \"scientific exploration engine,\" exploring the uncharted territories of science at unprecedented speed and breadth under the guidance of human wisdom.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">To promote this paradigm, the research team will open-source the core system of DeepScientist and all experimental logs, hoping to inspire the innovative power of the global research community through open sharing, jointly accelerating the development of AI Scientist and welcoming the breakthrough moments of major challenges facing humanity, from basic physics to drug development.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">The West Lake University Natural Language Processing Laboratory looks forward to working with more research teams to promote the advancement of automated scientific discovery.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">The team has now opened free service applications for DeepScientist, hoping to work with the research community to build a more efficient new paradigm of scientific discovery, making it truly accelerate the process of human scientific discovery.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cstrong class=\"ql-lineHeight-1-75\" style=\"font-size: 18px;\">West Lake University Natural Language Processing Laboratory (WestlakeNLP) was established in September 2018, led by Professor Zhang Yue.\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">Professor Zhang Yue graduated from the University of Oxford and received his doctorate. He currently serves as the Deputy Dean of the School of Engineering at West Lake University, authored the book \"Natural Language Processing\" published by Cambridge University Press, and has served as Program Committee Chair for EMNLP 2022 and other top NLP conferences.\u003C\u002Fspan>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\"> \u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">The laboratory currently focuses on language model reasoning, generalization, and general artificial intelligence, as well as fundamental and applied research in natural language processing, exploring the implementation paths of general artificial intelligence, promoting the development of AI Scientist (AI scientist), enabling it to truly participate in and accelerate scientific discovery, and promoting the continuous progress of human science.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">Around this vision, WestlakeNLP has also systematically written opinion articles and review papers on the AI Scientist direction, hoping to provide more comprehensive thinking and references for the development of this field.\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>\u003Cspan style=\"color: rgb(187, 187, 187);\">[News Source] QbitAI | WeChat Official Account \u003C\u002Fspan>\u003Ca href=\"https:\u002F\u002Fwww.thepaper.cn\u002FnewsDetail_forward_31744019\" rel=\"noopener noreferrer\" target=\"_blank\" style=\"color: rgb(187, 187, 187);\">https:\u002F\u002Fwww.thepaper.cn\u002FnewsDetail_forward_31744019\u003C\u002Fa>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"color: rgb(187, 187, 187);\">（This article is reprinted by this site to provide readers with more information and news, and the content does not constitute investment or consumer advice. If there are any doubts about the facts of the article, please verify with the relevant parties. The views expressed in the article are not the views of this 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%3Ac496447f-5676-44f7-8a21-a10c55a37139%3A0.wav?Expires=1774838459&OSSAccessKeyId=LTAI5tNvY2RkKjZw4LLWsrPK&Signature=EQu0%2B3Zeaa7HGC8jFEfMLsC2x3g%3D","c496447f-5676-44f7-8a21-a10c55a37139",17025396]