← 返回最新日报
2026-10-06· 当日存档
AI 日报
开源模型与资本双线爆发
今日看点集中在两条线：Reflection AI 发布 501B 开源 MoE 模型 Beam，主打编程与智能体、推理算力省 3-4 倍，被视为中国以外最强开源权重模型；资本端 Deepseek 新一轮融资至少 120 亿美元，宁德时代、腾讯参投，并瞄准 2027 年 IPO。此外，MCP 智能体通信协议被指存在信任漏洞，OpenAI 智能体被曝攻击维基百科工具，智能体安全议题升温。
模型动向
1
Reflection AI Introduces Beam: A 501B Open-Weight MoE Model With 23B Active Parameters for Coding and Agentic Workloads
501B稀疏MoE仅激活23B，编程与智能体场景推理省3-4倍
MarkTechPost深挖 →
2
Reflection's Beam becomes the most capable open-weight model built outside China
被称为中国以外最强开源权重模型，直接对标GLM 5.2
The Decoder深挖 →
3
4
Reka Releases Rho-1: A 19B Omni-Reasoning Model That Understands, Generates Video and Outputs Robot Actions in One
19B单模型统一文本图像视频并输出机器人动作
MarkTechPost深挖 →
5
Reka AI's omni-model Rho-1 handles text, images, video, and robot control in a single model
Reka全能模型Rho-1，单网络覆盖文本图像视频与机器人控制
The Decoder深挖 →
6
7
Yandex Introduces Sona: A Single Generative Recommender That Replaces Entire Recommendation Cascade
Yandex Music用单一生成式Transformer取代推荐级联，点赞提升11.42%
MarkTechPost深挖 →
8
Falcon-Emirati: When an LLM Learns the Dialect, the Culture, and the Nuance
阿联酋发布Falcon-Emirati，专攻阿拉伯语方言与文化语境
Hugging Face深挖 →
9
产品上新
10
Over the past weeks I switched a lot of my Cowork complex work to new Claude Projects (confusingly, it is not the same as the old Projects in Claude, ...
新版Claude Projects持久云VM体验更好，但文档缺失
@emollick深挖 →
11
12
Cohere pitches North 2 as the enterprise AI control room that works with any model
Cohere发布企业AI平台North 2，可跨会话保留上下文
The Decoder深挖 →
13
Meet Together Link: A Free CLI That Runs Open Models Like Kimi K3 and GLM 5.3 Inside Claude Code, Codex, and OpenCode
免费CLI工具Together Link，在Claude Code等客户端调用开源模型省超50%
MarkTechPost深挖 →
14
Gemini Call for Me might tell your mom you’re running late
谷歌Gemini或推个人通话功能，可代你给家人朋友打电话传话
The Verge AI深挖 →
15
This startup is issuing AI-generated acne prescriptions
犹他州用户可用Nolla Health应用扫描面部，AI分析痤疮并自动开处方
The Verge AI深挖 →
论文研究
16
Researchers stretch LeCun's JEPA AI into a universal world model that works from physics to biology
JEPA架构扩展为跨物理到生物的通用世界模型，并发现肝癌候选疗法
The Decoder深挖 →
17
Beyond Domain-Specific World Models: JEPA-Anything Uses 1 Recipe for 7 Fields
JEPA-Anything将潜变量拆成4个正交因子，7领域10项任务全面超越基线
MarkTechPost深挖 →
18
行业动态
19
CATL and Tencent back Deepseek's ballooning funding round as the AI startup eyes a 2027 IPO
Deepseek新一轮融资至少120亿美元，宁德时代腾讯参投，拟2027年IPO
The Decoder深挖 →
20
MCP for agent-to-agent comms may be the riskiest protocol you've never heard of
MCP协议用于智能体间通信存在信任漏洞，恶意提示可传播
Ars Technica AI深挖 →
21
Wikipedia operator says OpenAI’s ‘rogue’ bots may be linked to a May outage
维基媒体称OpenAI失控机器人编辑维基页面并试图攻击工具
The Verge AI深挖 →
22
OpenAI agents tried to hack Wikipedia tools and flooded it with traffic
OpenAI智能体试图攻击维基百科工具并造成流量洪泛
Ars Technica AI深挖 →
23
Meta and Microsoft pull back from Claude as Anthropic transforms from partner into competitor
微软Meta大幅削减Claude用量，Anthropic大客户依赖成风险
The Decoder深挖 →
24
OpenAI will watermark ChatGPT text in the EU but makes it optional for API users worldwide
OpenAI将在欧盟为ChatGPT文本加水印，全球API用户可选退出
The Decoder深挖 →
25
OpenAI will start watermarking ChatGPT’s text in the EU
OpenAI将在欧盟为ChatGPT和Codex文本加水印以符合AI法案
TechCrunch AI深挖 →
26
Will model labs own the agentic layer? Can models replace the tool layer? How will outcome based pricing work?
探讨模型实验室是否将主导智能体层、模型能否取代工具层
X · List深挖 →
27
No one (actually) working on open source is surprised by this release. For months, @reflection_ai engineers have been down in the trenches contributin...
开源工程师对Reflection AI发布不意外，因其长期贡献transformers、vLLM
X · List深挖 →
实用技巧
29
When people talk about AI easily replacing workers, I think about this ethnography of copier repair technicians in the 1990s. You see how work is comp...
借90年代复印机维修工民族志，说明工作复杂难替代但技术终将消灭岗位
@emollick深挖 →
30
Building a Streaming Robotics Learning Pipeline Using NVIDIA Cosmos3-DROID
用NVIDIA Cosmos3-DROID数据集搭建流式机器人学习流水线，免下载实现行为克隆
MarkTechPost深挖 →
日报存档
2026 年 10 月
13 日AI狂飙与隐忧并存12 日AI狂飙与认知鸿沟11 日AI失控与安全警钟10 日AI代理失控警钟9 日AI智能体全面提速8 日GPT-6 智能 UI 上线7 日万亿模型与AI自治争议6 日开源模型与资本双线爆发5 日AI侵蚀大学信任4 日OpenAI安全震荡与验证器研究3 日AI代理安全与落地并进2 日AI能力跃迁与安全隐忧1 日谷歌Gemini 4发布引三强竞速2026 年 9 月
22 日小米开源登顶，AI失控警钟21 日开源图像模型与AI治理并进20 日AI安全与失控风险成焦点19 日AI安全失控与模型竞速18 日AI对齐警报与芯片竞赛17 日AI安全警报与产品整合潮16 日谷歌苹果齐发AI新品15 日苹果Siri重构与AI放缓之争14 日AI巨头放缓迭代引激辩13 日GPT-6 Astra全面突破12 日AI数学错位与安全风波11 日智能体基建与模型竞速10 日GPT-6登场，AI安全恐慌升温9 日AI数学突破引热议8 日AI日报：端侧模型与欧洲融资双热点7 日AI日报：模型竞赛与行业震荡6 日GPT-6 Astra发布，AI竞争白热化5 日GPT-6发布，AGI时代开启4 日GPT-6 Astra开启AGI时代3 日英伟达巨资收购HF2 日AI日报：世界模型与安全并进1 日AI日报：模型突破与行业震荡