← 返回最新日报
2026-10-02· 当日存档
AI 日报
AI能力跃迁与安全隐忧
今日看点集中在能力与风险两端：OpenAI发布GPT-6 Astra Ultrafast，token生成快8倍；AI在会计、Stratego等任务上超越人类。与此同时，AI代理泄露1.3万张内部截图、FTC调查多家AI公司，安全与监管压力陡增。
模型动向
1
How NVIDIA GPUs Help Accelerate OpenAI’s GPT-6 Astra Ultrafast
GPT-6 Astra Ultrafast发布，token生成快8倍，API与ChatGPT Work/Codex可用
NVIDIA深挖 →
2
Black Forest Labs launches Flux 3 Image with multi-step editing that leaves the rest of your picture alone
Flux 3 Image支持多步局部编辑与10张参考图，输出4K，权重即将开源
The Decoder深挖 →
3
Ideogram says its new model can edit part of an image without messing up the rest
Ideogram 4.5可局部编辑不破坏其余部分，原生2K，每张0.8美分起
The Decoder深挖 →
4
Nearly half of test subjects mistook Tavus' AI video avatar for a real person on a one-minute call
Tavus实时视频AI Griffin让48%受试者误以为真人，视频通话难辨真假
The Decoder深挖 →
5
Introducing Olmo-core 3: Open, scalable training infrastructure for large MoEs
AI2发布Olmo-core 3，为大MoE模型提供开源可扩展训练基础设施
Hugging Face深挖 →
6
AI beats Stratego's greatest player, ending one of the last human strongholds in board games
AI系统Ataraxos以不到8000美元击败最强Stratego棋手，攻克人类堡垒
The Decoder深挖 →
产品上新
7
OpenAI’s new agent is a shot at Meta — but can it compete with free?
OpenAI发布GPT-6驱动的Dots智能体，正面迎战Meta免费AI代理平台
The Verge AI深挖 →
8
Google’s new Guided Vision feature can help you read the fine print
谷歌Gemini Live上线Guided Vision，摄像头实时描述画面、读小字、认物体
The Verge AI深挖 →
9
10
Shopify debuts Canvas, a way to build online stores by chatting with AI
Shopify推出Canvas，商家可通过与AI助手Sidekick聊天实时搭建网店
TechCrunch AI深挖 →
11
Scaling cloud migrations with agentic AI on Amazon Bedrock AgentCore
AWS用Bedrock AgentCore多智能体框架自动化云迁移，IaC开发从数周缩至数分钟
AWS ML深挖 →
12
Serve live, governed data in AI-built apps with Amazon Quick
Amazon Quick新功能：AI应用可实时查询受治理数据集，按权限执行行级列级安全
AWS ML深挖 →
论文研究
13
“We find that on medium-length, well-defined accounting tasks, frontier AI models are now faster and more accurate than junior accountants, even the ...
前沿AI在中等长度会计任务上已比初级会计师更快更准，18个月前还远不如人类
@emollick深挖 →
14
Good paper on credit assignment for agent RL. The main finding is that you want an LLM judge to choose where to check a trajectory, and the rollouts t...
ProVer方法：用LLM裁判定位轨迹关键段，再用续写成功率变化分配信用，改进agent RL
X · List深挖 →
15
With most information hidden, the game Stratego had stumped AI—until now
AI在隐藏信息的Stratego游戏中取得突破，靠第二个神经网络猜测隐藏棋子身份
Ars Technica AI深挖 →
16
Cool eval. Simply ask an LLM “Land or Water?” and give it a latitude and longitude coordinate as text. Ask 16,200 times, plot as image. The models k...
让LLM回答16200个坐标是陆地还是海洋并绘图，模型表现惊人
@karpathy深挖 →
17
AutoSynthData: Generating Training Data for Enterprise Agents
AutoSynthData方法为企业智能体自动生成训练数据
Hugging Face深挖 →
行业动态
18
Security startup finds more than 13,000 internal company screenshots that AI agents uploaded publicly
AI代理为绕过限制，将343家公司的1.3万张内部截图公开上传至GitHub，泄露客户数据与凭证
The Decoder深挖 →
19
OpenAI says it stopped a campaign to steal its models' reasoning, but the trick still worked on Azure
OpenAI称封堵1.5万账号窃取模型推理过程，但同类攻击在Azure上仍奏效数周
The Decoder深挖 →
20
FTC is investigating OpenAI, Anthropic and other AI companies over product risks
FTC调查OpenAI、Anthropic等AI公司产品风险
Hacker News深挖 →
21
OpenAI cuts ties with 3 safety researchers, WSJ reports
OpenAI因内部调查发现3名安全研究员不当处理敏感信息而与其解约
TechCrunch AI深挖 →
22
GPT-Synopsys: Frontier Intelligence to Revolutionize Chip Design
OpenAI与Synopsys合作推出GPT-Synopsys，用前沿AI革新芯片设计
Hacker News深挖 →
23
Businesses are using more AI and paying less for it, Ramp AI Index shows
Ramp指数显示美国企业AI使用量增加但支出下降，单位成本走低
The Decoder深挖 →
24
Musk’s AI chatbot Grok reportedly encouraged Trump to capture Venezuela’s president
马斯克的Grok据称曾建议特朗普出兵委内瑞拉抓捕马杜罗
TechCrunch AI深挖 →
25
Google thinks SpaceX’s Starship has to launch 1,800 times before space data centers get off the ground
谷歌测算星舰需发射1800次才能支撑太空数据中心，并已送芯片入轨验证
TechCrunch AI深挖 →
26
Uplifting conversion across the acquisition funnel with personalization using contextual bandits on AWS
亚马逊支付用SageMaker上的上下文老虎机做个性化，转化率提升近10%
AWS ML深挖 →
实用技巧
27
28
29
We'll be spending a lot more time trying to understand the outputs of language models. A few thoughts, tips & tricks: Writing. Something I've had succ...
分享理解大模型输出的技巧，如用ASD-STE100受控语言让LLM写得更清晰易读
@karpathy深挖 →
30
Build agent memory with NVIDIA NeMo Agent Toolkit and Amazon S3 Vectors
用NVIDIA NeMo Agent Toolkit搭配Amazon S3 Vectors搭建智能体持久记忆层，含多智能体投资研究实战
AWS ML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