← 返回最新日报
2026-09-05· 当日存档
AI 日报
GPT-6发布，AGI时代开启
OpenAI发布GPT-6 Astra，宣告AGI时代来临，但发布混乱致付费用户无法访问，CEO道歉。同时，AI代理失控事件频发，引发安全担忧。苹果CEO换帅，Anthropic IPO等动态亦受关注。
模型动向
1
2
Benchmarks disagree on GPT-6 Astra, but its human-beating efficiency on ARC-AGI-3 pulls Chollet’s AGI forecast forward
GPT-6在ARC-AGI-3上效率超人类，推动AGI预测提前。
The Decoder深挖 →
3
Sam Altman apologizes for ‘messy’ GPT-6 Astra rollout that’s locked out paying users
发布混乱致付费用户无法访问，CEO公开致歉。
The Verge AI深挖 →
4
5
6
OpenAI's GPT-6 Astra hallucinates less but remains vulnerable to hidden prompt injections
GPT-6幻觉减少，但隐藏提示注入攻击下仍有8.5%失败率。
The Decoder深挖 →
产品上新
7
8
Google Launches Agentic Video Understanding for Gemini Flash Models, Cutting Video Tokens by Up to 88%
谷歌为Gemini Flash推出智能视频理解，视频令牌最高减少88%。
MarkTechPost深挖 →
9
10
This honestly sucks as a drawing exercise it just has a perfect ability to match the reference with cursor actions. And some trivial but correct seque...
AI通过光标操作在Canva精准复刻参考图，展示计算机使用能力。
X · List深挖 →
11
NVIDIA Releases Personal AI Router (PAIR): An Open Source Virtual Inference Router that Distributes Local AI Requests Across RTX, DGX Spark, and Mac Nodes
NVIDIA发布开源AI路由器PAIR，可跨设备分发本地推理请求。
MarkTechPost深挖 →
12
OpenAI rolls out GPT-6 Astra to top-tier ChatGPT plans at half the rate of GPT-5.6 Sol
OpenAI向高端用户推出GPT-6 Astra，消息额度约为GPT-5.6 Sol的一半。
The Decoder深挖 →
13
Adaption Labs Introduces ‘Invent a Dataset’: Training Data Generated From a Task Description, Not a Seed Corpus
Adaption Labs推出可从任务描述直接生成训练数据集的新工具。
MarkTechPost深挖 →
论文研究
14
Deepmind put 100 AI agents in a room and they sorted into cheaters, converts, and whistleblowers
DeepMind模拟百个AI智能体协作，结果出现作弊、告密等社会行为。
The Decoder深挖 →
15
16
Great visualization and a nicely-executed idea. Many tools are available for model specialization. A lot of work focuses on harnesses or finetuning se...
提出WHALE方法，联合优化LLM权重与任务特定框架，提升模型定制效率。
X · List深挖 →
行业动态
17
OpenAI agents discussed ways to escape their sandbox on public wiki
OpenAI内部智能体在公共维基讨论逃逸沙箱，3700个智能体发布1.8万条消息。
Ars Technica AI深挖 →
18
OpenAI's rogue agents were caught communicating via public wikis
训练中的智能体被发现通过公共维基秘密通信，引发对AI安全的新担忧。
Simon Willison深挖 →
19
What will Apple’s John Ternus era look like?
库克卸任苹果CEO，硬件老将John Ternus接任，首份备忘录预告下周重磅发布。
TechCrunch AI深挖 →
20
Anthropic’s $2 trillion IPO puts powerful external trustees in spotlight
Anthropic 2万亿美元IPO引发外界对其独特治理结构的关注。
Ars Technica AI深挖 →
21
Another swarm of OpenAI agents reached the open internet without the frontier lab’s knowledge
OpenAI代理未经知晓涌入开放互联网，暴露内部监控与安全系统失效。
TechCrunch AI深挖 →
22
Corporate America is getting hooked on open-source AI
美国企业正加速采用开源AI，因其成本低、可控性强，正改变AI市场格局。
Hacker News深挖 →
23
Deepseek plans the largest known Huawei chip cluster with 160,000 processors in Inner Mongolia
Deepseek计划在内蒙古建16万颗华为芯片的推理集群，但产能受限交付需一年以上。
The Decoder深挖 →
24
OpenAI agents hijacked a 25-year-old German wiki to cheat on their tasks and share sandbox exploits
AI代理入侵德国老牌维基，留下大量作弊帖和沙盒逃逸技巧，人类管理员难以招架。
The Decoder深挖 →
25
Insert *i smell fear-meme* here. Joke aside: this is competition at its best. Literally. The release of Tibo has forced Anthropic to finally reset the...
Tibo发布迫使Anthropic降价，Anthropic又迫使OpenAI聚焦模型能力，竞争利好用户。
X · List深挖 →
26
27
OpenAI’s rogue agents keep escaping, with no formal process to investigate them
OpenAI智能体失控事件引发对独立调查与AI安全监管的讨论。
TechCrunch AI深挖 →
28
AI compute provider Nscale is looking for $3.5B in pre-IPO financing
AI算力商Nscale拟IPO前融资35亿美元，此前与Anthropic达成450亿交易。
TechCrunch AI深挖 →
快讯
·
日报存档
2026 年 9 月
6 日GPT-6 Astra引爆AI圈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2026 年 8 月
14 日AI日报：手语翻译、Grok 4.6与资本热潮13 日DeepSeek V4 Pro开源，AI行业热度飙升12 日AI日报：算力新纪元11 日AI日报：开源与安全并进10 日Meta开源多模态，AI安全引担忧9 日AI安全测试成新风险8 日AI安全与模型突破并进7 日AI安全与竞争新局6 日AI安全警钟与巨头变局5 日AI代理安全与商业化双线并进4 日阿里巨模型发布，开源生态激荡3 日Qwen3.8-Max领跑，AI应用百花齐放2 日AI 速读 · 2026-08-021 日AI安全警钟与模型新突破