估值百亿，产品几无，这便是当下最疯狂的AI资本叙事。 最近，前OpenAI CTO米拉·穆拉蒂创立的Thinking Machines Lab，正以高达500亿美元的估值，筹划新一轮40亿至50亿美元的融资。而它拿得出手的，仅是 一个名为Tinker、界面尚待考证的API接口。 相比传统商业逻辑，这无异于一场豪赌：资本押注的，早已不是产品，而是人，是那串贴在创始人身上的"OpenAI开国元老"的黄金标签。 穆拉蒂的实验室，正是席卷硅谷的neolab（新生代实验室）风潮中最耀眼的一朵浪花。据国外知名VC menlo合伙人Deedy推文，AI领域新兴的10家neolab 中，9家在种子轮阶段就斩获10亿美元估值。 一批从OpenAI、DeepMind等巨头出走的顶尖研究员，正以反叛者的姿态，用全新的范式重建AI研究的逻辑。他们不谈营收，不论商业化，只谈那些看似 天方夜谭的方向：情感智能、AI社会、自动化科学家等。 资本的反应则更为直接与狂热：Humans&成立数月估值40亿，SSI瞄准超智能安全估值320亿，Periodic Labs一出手种子轮便是3亿……这些在传统视角 下"啥也没有"的实验室，正以令人咋舌的估 ...

Core Insights - DeepSeek has released two official model versions: DeepSeek-V3.2 and DeepSeek-V3.2-Speciale [1][4] Group 1: Model Features - DeepSeek-V3.2 aims to balance reasoning capability and output length, making it suitable for everyday use. In benchmark tests, it reached the level of GPT-5, slightly below Gemini-3.0-Pro, while significantly reducing output length compared to Kimi-K2-Thinking, thus decreasing computational overhead and user wait time [3] - DeepSeek-V3.2-Speciale is designed to push the reasoning capabilities of open-source models to the extreme, exploring the boundaries of model capabilities. It is an enhanced version of DeepSeek-V3.2 with long-thinking capabilities and incorporates theorem-proving abilities from DeepSeek-Math-V2 [3] Group 2: Achievements - The V3.2-Speciale model has successfully won gold medals at several prestigious competitions, including IMO 2025, CMO 2025, ICPC World Finals 2025, and IOI 2025 [3] Group 3: Availability - The official web version, app, and API have all been updated to the official version of DeepSeek-V3.2. The Speciale version is currently available only as a temporary API service for community evaluation and research [4]

新浪科技讯 12月1日晚间消息，继两个月前发布实验性的 DeepSeek-V3.2-Exp后，DeepSeek今日宣布同 时发布两个正式版模型：DeepSeek-V3.2 和 DeepSeek-V3.2-Speciale。 在高度复杂任务上，Speciale 模型大幅优于标准版本，但消耗的 Tokens 也显著更多，成本更高。目 前，DeepSeek-V3.2-Speciale 仅供研究使用，不支持工具调用，暂未针对日常对话与写作任务进行专项 优化。（文猛） 责任编辑：何俊熹 据悉，DeepSeek-V3.2 的目标是平衡推理能力与输出长度，适合日常使用，例如问答场景和通用 Agent 任务场景。在公开的推理类 Benchmark 测试中，DeepSeek-V3.2 达到了 GPT-5 的水平，仅略低于 Gemini-3.0-Pro；相比 Kimi-K2-Thinking，V3.2 的输出长度大幅降低，显著减少了计算开销与用户等待 时间。 DeepSeek-V3.2-Speciale 的目标是将开源模型的推理能力推向极致，探索模型能力的边界。V3.2- Speciale 是 DeepSeek-V3.2 的长思 ...

Core Insights - DeepSeek has released two official model versions: DeepSeek-V3.2 and DeepSeek-V3.2-Speciale [1] - The main goal of DeepSeek-V3.2 is to balance reasoning capability with output length, making it suitable for everyday use cases such as Q&A and general agent tasks [1] - The DeepSeek-V3.2-Speciale version aims to push the reasoning capabilities of open-source models to the extreme, exploring the boundaries of model capabilities [1] Summary by Categories - **Product Launch** - DeepSeek has updated its official website, app, and API to the official version of DeepSeek-V3.2 [1] - The Speciale version is currently available only as a temporary API service for community evaluation and research [1] - **Model Objectives** - DeepSeek-V3.2 is designed for daily applications, focusing on practical scenarios like Q&A and general agent tasks [1] - DeepSeek-V3.2-Speciale is focused on maximizing the reasoning capabilities of the model, aiming to explore its limits [1]

风险提示及免责条款 市场有风险，投资需谨慎。本文不构成个人投资建议，也未考虑到个别用户特殊的投资目标、财务状况或需要。用户应考虑本文中的任何 意见、观点或结论是否符合其特定状况。据此投资，责任自负。 DeepSeek宣布同时发布两个正式版模型：DeepSeek-V3.2 和 DeepSeek-V3.2-Speciale。DeepSeek-V3.2 的 目标是平衡推理能力与输出长度，适合日常使用，例如问答场景和通用 Agent 任务场景。DeepSeek- V3.2-Speciale 的目标是将开源模型的推理能力推向极致，探索模型能力的边界。 ...

中游算法领域，江北企业自研的 "数据炼钢"元冶·钢铁大模型、可以"一键出片"的KINO视频大模型、农业全场景数字化转型的智稷农业大模型等一批垂类 大模型，正在工业、消费、文娱等领域广泛应用。 此前，江北新区、浦口区已与迈富时进行多轮洽谈，本次签约亮相，意味着这个百亿级人工智能企业项目正式落地大江北岸。 作为行业巨头，迈富时的产业链上下游，齐聚着一众明星企业：华为、腾讯、字节、百度、蚂蚁、沐曦、OpenText、HCL等。当以迈富时为核心的人工 智能产业集群，在江北逐渐壮大，更多"人工智能+" 的示范场景、发展路径也将逐渐明晰。一个百亿级项目，为江北带来的是更多产业发展的新可能。 如今放眼江北，上游算力领域建设有总算力达10000P的中国移动智算中心、服务客户300余家的南京人工智能计算中心。 12月1日，迈富时全球研发总部签约落户活动在南京举行，项目总投资100亿元。 迈富时成立于2009年，在香港主板上市，是全球领先的AI应用平台，也是国内收入规模最大、增速最快的营销及销售SaaS公司。去年全年营业收入超过 80亿元人民币。 企业全力发展，江北鼎力支持——近年来，新区不仅组建了"1+2+1+N"基金矩阵，支 ...

Core Viewpoint - The former CEO of Google, Eric Schmidt, suggests that while the U.S. and China may never be "best friends," they can learn to cooperate despite their differing systems [1][4]. Group 1: Technological Advancements - Schmidt highlights that 70% of global AI patents and 75% of clean energy technology patent applications come from China, indicating a significant shift in technological capabilities [4]. - Chinese companies like DeepSeek and Kimi are now competing with OpenAI's GPT, utilizing an open-source approach that contrasts with the closed-source model of U.S. firms, allowing for faster innovation [4]. - A Chinese humanoid robot company has developed a product that can assemble precision parts at one-third the cost of similar U.S. products, showcasing a new competitive strategy [4]. Group 2: Industry Reactions - Schmidt's recent comments have sparked discussions in the tech community, with some recalling his previous statements from 2010 that claimed China was 20 years behind the U.S. in technology [6]. - Engineers and industry professionals are acknowledging the shift in perception, with some expressing that the previous views of China as merely a copycat have changed due to China's active open-source community [6]. - The sentiment in the tech community reflects a pragmatic approach, recognizing that both countries have their strengths and that complete decoupling would harm both sides' supply chains [6]. Group 3: Future Implications - Schmidt's remarks serve as a reminder that as perceptions of China change, the competition will become more intense and focused on hard capabilities [6]. - The need for cooperation in areas like counter-terrorism, stability, and AI safety is emphasized, suggesting that collaboration may be necessary despite underlying tensions [6].

KYIV, Ukraine, Dec. 01, 2025 (GLOBE NEWSWIRE) -- Kyivstar (Nasdaq: KYIV; KYIVW), Ukraine’s leading digital operator, and the WINWIN AI Center of Excellence under the Ministry of Digital Transformation of Ukraine have selected Google’s Gemma as the base model for training the large language model (LLM). Gemma, Google’s next-generation open AI model, has been proven effective in both international and domestic projects. Kyivstar is the Ukrainian Government’s strategic partner and operational lead for developi ...

12月1日，AI产业链喜提开门红！科创人工智能ETF汇添富(589560)放量上涨1.74%，全天成交额超2300万元，环比激增106%。 科创人工智能ETF汇添富(589560)标的指数强势上涨1.59%，成分股乐鑫科技上涨8.55%，晶晨股份，复旦微电、 恒玄科技等涨超4%，寒武 纪、云天励飞等涨超2%，澜起科技、金山办公等强势跟涨，芯原股份等回调。 | 序号 | 代码 | 名称 | 涨跌幅 | 估算权重 ▼ | | --- | --- | --- | --- | --- | | 1 | 688256 | 塞武纪-U | 2.36% | 14.15% | | 2 | 688008 | 澜起科技 | 1.65% | 10.71% | | 3 | 688521 | 芯原股份 | -1.76% | 9.96% | | 4 | 688111 | 金山办公 | 1.85% | 9.93% | | 5 | 688099 | 品晨股份 | 4.84% | 5.77% | | 6 | 688169 | 石头科技 | 0.40% | 5.66% | | 7 | 688608 | 恒玄科技 | 4.21% | 4.86% | ...

本研究由史蒂文斯理工学院（Stevens Institute of Technology）的 IntelliSys Lab 团队以及休斯顿大学（University of Houston） 的 ANTS Lab 团队的相关研究人员 共同完成，主要研究方向包括无服务器计算、高性能计算、AI 安全、联邦学习、LLM 代理系统、隐私保护及分布式智能系统等领域。 随着大语言模型（LLM）的商业价值快速提升，其昂贵的训练成本使得模型版权保护（IP Protection）成为业界关注的焦点。然而，现有模型版权验证手 段（如模型指纹）往往忽略一个关键威胁：攻击者一旦直接窃取模型权重，即拥有对模型的完全控制权，能够逆向指纹 / 水印，或通过修改输出内容绕过指 纹验证。 在此背景下，史蒂文斯理工学院（Stevens Institute of Technology）王灏老师团队的第二年级博士生熊子洵同学提出了 iSeal。该工作已被 AAAI 2026 主轨道以 poster 形式录用。 iSeal 是首个面向「端到端」模型窃取场景设计的加密指纹方案。它通过引入加密机制，使得指纹可抵御拥有模型完全控制权的攻击者所发起的「合谋遗忘 ...