Workflow
大语言模型
icon
Search documents
专访|人工智能同样需要“终身”学习——访人工智能促进协会主席斯蒂芬·史密斯
Xin Hua She· 2026-01-29 04:13
Core Insights - The future development of artificial intelligence (AI) may hinge on the concept of "lifelong learning," similar to human learning methods [1] - The rise of large language models (LLMs) has been a significant breakthrough in AI, but they have limitations, including a lack of continuous updating and causal reasoning capabilities [1][2] - Achieving "lifelong learning" in AI presents technical challenges, particularly in fine-tuning existing LLMs without compromising their performance [2] Group 1 - The most notable breakthrough in AI is the emergence of large language models, which can understand and generate text based on extensive data training [1] - Current AI systems, primarily based on LLMs, are often "frozen" after initial training, lacking the ability to grow and adapt over time [1] - LLMs excel at identifying correlations but struggle with causal reasoning, which limits their planning abilities and can lead to nonsensical outputs [1] Group 2 - Implementing "lifelong learning" in AI could mimic human learning processes, relying on small samples and selective data rather than vast amounts of information [2] - Robotics and embodied intelligence may enhance AI development by allowing interaction with the physical world, thereby accumulating experience and understanding causal relationships [2] - The future direction of AI includes the development of autonomous agents that can make independent decisions and collaborate with other agents to solve complex problems [2]
民政部:推动大语言模型、脑机接口等技术在残疾儿童康复领域的应用
Xin Lang Cai Jing· 2026-01-29 02:20
民政部日前印发《关于进一步推进民政科技创新的指导意见》。其中提出,推进康复辅助器具、殡葬等 领域技术创新。加强康复辅助器具科技攻关,研究精准神经调控、人机协同控制等关键技术,研发外骨 骼机器人、智能矫形器、智能假肢、精神障碍患者康复辅助器具等技术装备和产品;研发康复辅助器具 适配对象能力数字化评估技术,研究质量检测方法和标准规范,研制检测设备,完善康复辅助器具检测 评估体系。推动大语言模型、脑机接口等技术在残疾儿童康复领域的应用,加强脑瘫、孤独症等残疾儿 童康复技术和设备研发。加强殡葬服务全过程相关技术研究,研发新型绿色生态环保遗体处置技术,研 制绿色低碳、智能化殡葬用品和遗体保存运输设备,推动遗体和遗物祭品处置技术设备迭代升级,实现 殡葬领域碳减排;开展遗体防腐整容、殡葬污染物协同控制、殡仪场所消毒与生物学污染综合控制等技 术研究,建立污染物防治技术体系。 ...
“财”访一线|人工智能 如何进厂打工
Xin Hua Cai Jing· 2026-01-28 11:53
编辑:幸骊莎 新华财经南京1月28日电(记者常竣、斐朱程、李亭)人工智能如何在制造业深度融合应用?记者在江 苏无锡探访无锡威孚高科技集团股份有限公司、江苏省具身智能机器人工业数据采集与实训中心看到, 企业正积极在生产制造的各个环节融入人工智能相关技术,积极运用大语言模型技术提升工作效率。在 具身智能机器人的浪潮下,针对性的工业数据采集与实训变得越来越重要。具身智能机器人进厂打工的 未来正在靠近。(参与调研:谷青竹、欧阳迪娜、赵畅) ...
ClawdBot让Mac Mini卖爆了,京东年货节国补叠加以旧换新补贴到手价更低
Sou Hu Wang· 2026-01-28 02:39
据了解,ClawdBot不是一个聊天机器人(Chatbot),而是一个智能体网关(AI Agent Gateway)。与ChatGPT 或Claude这类需要你打开网页、输入问题、等待回复的工具不同,Clawdbot是通过日常使用的消息应用 发出指令,唤起后台运行的大语言模型,从而将需求转化为本地Shell脚本并在电脑上执行。换句话 说,它不是告诉你怎么做,而是直接帮你做完一件事。这种从"AI给建议"到"AI直接行动"的转变,正是 让大家着迷的一大原因。 目前京东年货节正在火热进行中,除了Mac mini外,iPhone、MacBook、iPad、AirPods、Apple Watch等 全线产品都有优惠可享。感兴趣的果粉们,只需打开京东APP搜索"苹果大额券"即可直达活动会场,抓 住这个难得的机会,早买早享受。 如果你也想率先体验ClawdBot,不妨来京东Apple年货节下单Mac mini。以Mac mini M4 10+10核16G 256G版为例,1月27日24点前来京东下单,还可享官方直降400元福利,叠加国补15%至高1500元、以 旧换新至高再补800元后,到手价低至2804元。1月28日起下 ...
或颠覆文档处理模式,DeepSeek OCR模型再更新
Xuan Gu Bao· 2026-01-27 23:16
海通国际表示,DeepSeek-OCR代表新一代"压缩存储"思路,通过将文本映射为视觉表征并进行高倍率 压缩,以少量视觉token承载长上下文信息,仅在需使用时按需解码还原,从而实现从"扩大计算基 数"到"减少计算负荷"的根本性转变;据论文及第三方评测数据,DeepSeek-OCR在低于10倍压缩率下可 实现约97%的文本还原精度,能够满足多数信息检索与文档归档类需求;而在20倍高压缩率下精度约为 60%,适用于容错性较高的线索检索场景。 华创证券指出,DeepSeek-OCR在20个A100节点上日处理3300万页数据的吞吐能力,以及对小语种(如 阿拉伯语、僧伽罗语)的良好支持,使其在全球化商业部署中具有显著优势,这种"视觉即压缩"的范式 可能重塑未来大语言模型的输入方式。 据新浪财经1月27日报道,DeepSeek发布全新DeepSeek-OCR2模型,采用创新的DeepEncoderV2方法, 让AI能够根据图像的含义动态重排图像的各个部分,而不再只是机械地从左到右扫描。 在基准测试中,新模型达到91.09%的性能,较前代提升3.73%,同时视觉token使用上限降至1120个(前 代为1156个)。这 ...
中文传媒:“领思大模型”是公司旗下江西新华云教育科技有限公司自研产品
Mei Ri Jing Ji Xin Wen· 2026-01-27 12:08
(文章来源:每日经济新闻) 每经AI快讯,有投资者在投资者互动平台提问:自研大语言模型"领思大模型"是公司的产品吗?公司业 务中有没有使用"领思大模型"? 中文传媒(600373.SH)1月27日在投资者互动平台表示,尊敬的投资者,您好!"领思大模型"是公司旗 下江西新华云教育科技有限公司自研产品,目前"文书守正"等项目已使用该模型,谢谢。 ...
同花顺(300033):公司动态研究:市场回暖与AI赋能双重驱动,金融信息服务价值持续释放
Guohai Securities· 2026-01-27 12:03
2026 年 01 月 27 日 公司研究 评级:买入(维持) 研究所: 证券分析师: 刘熹 S0350523040001 liux10@ghzq.com.cn 联系人 : 谢婧茹 S0350125070015 xiejr01@ghzq.com.cn [Table_Title] 市场回暖与 AI 赋能双重驱动,金融信息服务价 值持续释放 ——同花顺(300033)公司动态研究 最近一年走势 | 相对沪深 | 300 表现 | | 2026/01/26 | | --- | --- | --- | --- | | 表现 | | 1M | 3M 12M | | 同花顺 | | 8.9% | -1.9% 18.4% | | 沪深 300 | | 1.1% | 1.0% 22.8% | | 市场数据 | 2026/01/26 | | --- | --- | | 当前价格(元) | 355.03 | | 周价格区间(元) 52 | 232.00-435.00 | | 总市值(百万) | 190,864.13 | | 流通市值(百万) | 111,177.84 | | 总股本(万股) | 53,760.00 | | 流通股 ...
爆火的「Agentic推理」是什么?怎么用?未来机会在哪里?一文读懂
3 6 Ke· 2026-01-27 10:56
推理(Reasoning)是智能的核心。有了这一能力,人工智能(AI)模型在开放、动态环境下完成逻辑推断、问题解决和决策制定等任务才成为可能。 在当前"由言到行"的范式转变中,大语言模型(LLM)已不再是被动的序列生成器,而是被要求进化为在持续交互过程中实时规划、行动和学习的自主推 理智能体(Agent)。也因此,Agentic 推理成为了当前 AI 大模型行业的热门研究方向之一。 日前,一篇题为"Agentic Reasoning for Large Language Models"的综述文章在 X 上爆火。该综述系统推理了 Agentic 推理的演进脉络,为下一代自适应协作 智能体指明了方向。研究团队来自伊利诺伊大学厄巴纳-香槟分校、Meta、亚马逊、Google DeepMind、加州大学圣地亚哥分校和耶鲁大学。 论文链接:https://arxiv.org/pdf/2601.12538 该综述长达 135 页,涉及"基础 Agentic 推理"、"自进化 Agentic 推理"和"集体多智能体推理"等不同层级,上下文推理与后训练推理两种关键优化模式,以 及 Agentic 推理在科学、机器人、医疗、自 ...
AI技术与电商生态双重变革,智能客服如何破局?对话淘宝店小蜜负责人开锋
雷峰网· 2026-01-27 06:43
Core Viewpoint - AI technology is transforming customer service from a cost center into a growth department, enhancing operational efficiency and customer experience [1][4]. Group 1: AI Development and Market Trends - The current development of AI technology is characterized by a "dualistic" trend, with AI assistants rapidly penetrating the consumer market while challenges remain in achieving practical applications and finding product-market fit [2]. - The intelligent customer service sector is seen as a promising area to address these challenges due to its natural alignment with AI capabilities [3]. Group 2: Customer Service Evolution - Multi-turn dialogue understanding is a core advantage of large language models, which aligns well with the inherent nature of customer service interactions [5][6]. - Text generation is a fundamental capability of large language models, making it suitable for various customer service communication forms [6]. Group 3: E-commerce and Customer Service Integration - The focus on "existing user operations" has become central to e-commerce competition, with new service quality metrics being integrated into platform traffic allocation systems [7]. - The shift in strategy emphasizes that service quality is now a critical factor for traffic acquisition and order conversion, leading to a redefined role for customer service as a value-generating function [7]. Group 4: Case Study of Ding Xiaomi - Ding Xiaomi, an intelligent customer service product, has evolved significantly over the past decade, initially addressing high volumes of inquiries during peak sales events [9][10]. - The introduction of Ding Xiaomi 5.0, based on large language model technology, has led to a reduction in manual intervention rates by over 20% and an increase in transaction conversion rates by over 35% [11]. Group 5: Cost Efficiency and Performance Improvement - Ding Xiaomi 5.0 has helped merchants reduce configuration costs by 60%, streamlining the process of training and maintaining customer service systems [19][20]. - The product's ability to automatically extract and integrate product information has significantly reduced the need for extensive manual configuration by merchants [20]. Group 6: Future Directions and Enhancements - Future iterations of Ding Xiaomi will focus on improving pre-sale and post-sale capabilities, enhancing the overall service experience for users [26]. - The product will also allow merchants to integrate their internal knowledge bases and strategies, enabling more personalized and differentiated service capabilities [26].
大模型哪里出问题、怎么修,这篇可解释性综述一次讲清
机器之心· 2026-01-27 04:00
过去几年,机制可解释性 (Mechanistic Interpretability) 让研究者得以在 Transformer 这一 "黑盒" 里追踪信息如何流动、表征如何形成:从单个神经元到注意力头,再到 跨层电路。但在很多场景里,研究者真正关心的不只是 "模型为什么这么答",还包括 "能不能更稳、更准、更省,更安全"。 正是在这一背景下,来自 香港大学、 复旦大学 、慕尼黑大学、曼切斯特大学、腾讯 等机构的研究团队联合发布了 "可实践的机制可解释性" (Actio nable Mechanistic Interpretability) 综述。文章通过 "Locate, Steer, and Improve" 的三阶段范式,系统梳理了如何将 MI 从 "显微镜" 转化为 "手术刀",为大模型的对齐、能力增强和效 率提升提供了一套具体的方法论。 从 "显微镜" 到 "手术刀" 的范式转移 尽管大语言模型(LLM)近年来在多种任务上展现出了强大的能力,但其内部的运作机制依然在很大程度上不透明,常被视为一个 "黑盒"。围绕如何理解这一黑 盒,机制可解释性 (Mechanistic Interpretability, ...