国内AI动态

1. 新石器NewClaw：AI一体化解决方案，零门槛当无人车指挥官| 2026AI Partner·北京亦庄AI+产业大会

管理一千台无人车需要多少人？答案是：一个人，一部手机，一句话。当自动驾驶逐渐“平权”，真正的瓶颈从技术转向了规模化运营。

来源：36氪 | 原文链接 | 发布时间：05-21 18:55

2. 从概念到产线一：AI在工业制造领域的深水区探索| 2026AI Partner·北京亦庄AI+产业大会

AI在工业制造领域，不是“锦上添花”的辅助工具，而是“重新设计工厂”的核心引擎。从AOI报废板的秒级识别，到刀具参数的动态优化，再到打通设计、生产、供应链的全链路智能——这场对话告诉我们，每1%的效率提升都是真金白银，AI的价值不是叠加功能，而是把“人等货”变成“货等人”。

来源：36氪 | 原文链接 | 发布时间：05-21 18:29

3. 城市级AI服务：从试点到常态化，机器人的实景作战与规模化落地| 2026AI Partner·北京亦庄AI+产业大会

具身智能最大的瓶颈不是算法，而是数据——没有量产就没有数据，没有数据就无法进化。酷哇的解法是“以战养战”：让机器人在真实运营中一边干活一边成长，用万台规模反哺模型迭代。

来源：36氪 | 原文链接 | 发布时间：05-21 18:15

4. 硅谷 AI 格局生变：卡帕西跳槽、马斯克助阵，孙正义成孤守者

来源：钛媒体 | 原文链接 | 发布时间：05-21 18:32

5. Meta裁员8000人背后：扎克伯格千亿美元豪赌AI

来源：钛媒体 | 原文链接 | 发布时间：05-21 18:24

6. 孙正义600亿美元押注OpenAI，软银内部慌了

来源：钛媒体 | 原文链接 | 发布时间：05-21 18:24

7. 腾讯混元开源全新翻译模型Hy-MT2 ，上线小程序「腾讯Hy翻译」

来源：量子位 | 原文链接 | 发布时间：05-21 17:48

8. 菲尔兹奖得主都看懵了：OpenAI非数学模型首次自主突破80年未解数学难题

来源：量子位 | 原文链接 | 发布时间：05-21 17:36

9. 风行在线CEO易正朝：先全员Coding，再All in众创丨AIGC2026

来源：量子位 | 原文链接 | 发布时间：05-21 17:27

10. DeepSeek组建Harness团队，非「超能力者」不要？中国AI开启「做产品」的关键一跳

大模型行业的战火，正在从“基座智商”烧向“终端入口”。5月19日，DeepSeek 资深研究员陈德里在小红书上发了一条招聘贴，证实内部正在组建全新的”Harness”（代码智能体工程）团队，目标直指 Anthropic 旗下的明星产品 Claude Code。

来源：雷锋网 | 原文链接 | 发布时间：05-21 18:52

11. OpenAI放弃的事，为什么阿里可能做得成

用户打开千问App，与AI对话，即可完成淘宝上的商品挑选、对比及下单购买。在淘宝端内，点击“千问AI购物助手”也可以体验AI购物。

来源：雷锋网 | 原文链接 | 发布时间：05-21 15:59

12. 词元时代，万物智能 | 摩尔线程2026产品发布会：打造全场景AI算力基石

5 月 18 日，摩尔线程在北京举办主题为“词元时代，万物智能”的产品发布会。在 Agentic AI 驱动词元（Token）需求呈指数级跃升的关键节点，万物智能处于爆发前夜，算力的基石作用愈发关键。

来源：InfoQ | 原文链接 | 发布时间：05-21 23:37

13. 中国最神秘AI孵化器正式亮相：11位“大佬”导师成为超强外挂

近日，上海 AI 孵化器语生科学正式亮相。一年前的 2025 年 3 月，“语生科学创新院”（Genora Science）在上海悄然落地。

来源：InfoQ | 原文链接 | 发布时间：05-21 23:18

国外AI动态

1. US Cyber Command races to deploy AI on top-secret networks

US Cyber Command has launched a task force to run AI models from OpenAI, Google, and others on the most classified Pentagon and NSA networks The trigger: AI systems like Anthropic’s Claude Mythos can find security vulnerabilities faster than the best human hackers - and according to Anthropic, comparable tools could be widely available within six to 24 months.

来源：The Decoder | 原文链接 | 发布时间：05-21 23:20

2. Cohere open-sources its strongest model yet

The Canadian AI company Cohere is releasing its most powerful language model to date, Command A+, as open source under an Apache 2 The article Cohere open-sources its strongest model yet appeared first on The Decoder.

来源：The Decoder | 原文链接 | 发布时间：05-21 23:17

3. Anthropic is about to become the first profitable AI lab

Anthropic is approaching its first profitable quarter, the Wall Street Journal reports - with a projected operating profit of $559 million on $10 9 billion in Q2 revenue.

来源：The Decoder | 原文链接 | 发布时间：05-21 23:15

4. OpenAI could file confidential IPO paperwork within days

OpenAI is also preparing an IPO, according to the Wall Street Journal, and could file confidential paperwork with the SEC within days The article OpenAI could file confidential IPO paperwork within days appeared first on The Decoder.

来源：The Decoder | 原文链接 | 发布时间：05-21 23:12

5. SpaceX IPO filing shows billions in AI losses, a $2 trillion valuation target, and turbine spending that signals more data center conflicts ahead

SpaceX has filed for what could be the largest IPO ever, targeting a valuation of up to $2 trillion The filing reveals xAI losses of $6.

来源：The Decoder | 原文链接 | 发布时间：05-21 23:10

6. SAP taps Mistral AI to help customers migrate legacy software

SAP uses Mistral AI models to simplify the migration to S/4HANA The article SAP taps Mistral AI to help customers migrate legacy software appeared first on The Decoder.

来源：The Decoder | 原文链接 | 发布时间：05-21 22:59

7. Deepseek wants to take on Claude Code and OpenAI’s Codex with “Deepseek Code”

Deepseek is building a new team in Beijing to develop its own AI code agent, working title “Deepseek Code,” a direct competitor to Claude Code, Codex, and Cursor Applicants should know agent loops, MCP, and context engineering and be heavy users of existing coding tools.

来源：The Decoder | 原文链接 | 发布时间：05-21 02:48

8. LinkedIn’s war on AI slop is not just a policy update—it is an admission that the platform lost control of its feed

Linkedin is cracking down on AI-generated junk content it calls “AI slop “ In early tests, the platform says it correctly flagged generic posts 94 percent of the time.

来源：The Decoder | 原文链接 | 发布时间：05-21 02:21

9. Spotify adds AI-powered Q&A and briefing generation features to podcasts

Spotify will let you generate daily or weekly briefs based on your prompts.

来源：TechCrunch | 原文链接 | 发布时间：05-21 23:27

10. The Path, founded by Tony Robbins and Calm alums, hopes to offer safer AI therapy

The Path says its AI model has scored 95 on the mental health safety AI benchmark, Vera-MH This compares to a top score of 65 for the consumer bots.

来源：TechCrunch | 原文链接 | 发布时间：05-21 22:00

11. Hark raises $700M Series A for its secretive ‘universal’ AI interface

Hark expects to release its first multimodal models this summer, which it says will power a personal AI platform that works with existing products and services The company expects to follow that with hardware devices built specifically for those systems.

来源：TechCrunch | 原文链接 | 发布时间：05-21 22:00

12. Google is pitching an AI智能体 ecosystem to consumers who may not buy it

One of the most promising introductions at Google’s I/O developer conference on Tuesday was a new way for consumers to use the web: AI agents Unfortunately, it was also the most confusing.

来源：TechCrunch | 原文链接 | 发布时间：05-21 21:52

今日看点

今日国内AI亮点集中在2026亦庄AI+产业大会，围绕无人配送AI Agent、工业制造AI化、具身智能与解自AI，体现AI正从概念验证全面走向产线规模化持续稳定；国际上，Anthropic危机关头即将成首个盈利AI实验室，DeepSeek开发代码智能体，OpenAI加速备战IPO，AI已正超越费钱产品，全面进入商业化阶段。

数据来源：36氪、钛媒体、量子位、雷锋网、InfoQ、TechCrunch、The Decoder | 共收录25条