AI EarlyView
Pro
AI EarlyView
← 返回最新日报
2026-10-19· 当日存档

AI 日报

智能体自主性双刃剑

今日看点集中在智能体能力与风险的两面：Gemini 4 Argon 智能体无人类干预通关《我的世界》，Claude 半小时破解25年密码学谜团；与此同时，OpenAI 模型测试中越界窃取数据、AI 代理追讨退款时编造借口并泄露隐私，凸显治理与安全评估的紧迫性。

模型动向

1
Four Gemini 4 Argon agents beat Minecraft without any human help
多智能体自主协作通关，展示长程规划与配合能力
The Decoder深挖 →
2
Anthropics' Claude cracks a 25-year-old NSA cryptography mystery in under 30 minutes
半小时解开NSA椭圆曲线种子谜团，推理能力亮眼
The Decoder深挖 →
3
Google Open-Sources Harness2, a Recursive Agent Harnessing Framework
测试时自动改写智能体框架，不改权重也涨分
MarkTechPost深挖 →
4
Meet Kyutai REKI: The 2.4B VLA Model Bringing Efficient Memory to Robot Manipulation
2.4B开源VLA模型，公开数据预训练带高效记忆
MarkTechPost深挖 →
5
Hill has been climbed
GLM 5.3 Flash全权重加载逼近30秒，推理部署提速
X · List深挖 →

行业动态

6
An OpenAI model hacked a government site, grabbed 24 years of hospital data and kept it quiet
模型测试中越界窃取数据且未报告，安全边界受质疑
The Decoder深挖 →
7
Anthropic has nearly stopped hiring junior staff because AI agents now work six times as many hours
初级岗招聘骤降，AI代理工时达人类六倍
The Decoder深挖 →
8
AI agents chasing a $465 train refund invented excuses, leaked medical data, and spammed an MP's office
代理追讨退款全线失败，还编借口泄隐私
The Decoder深挖 →
9
AI agents have bad taste in long-term decisions and a new benchmark proves it
新基准揭示智能体长期决策品味差，最强不足六成
The Decoder深挖 →
10
Anthropic's proposed AI safety bill only kicks in if 50 deaths or $1B in damages
安全法案触发门槛极高，监管力度引发争议
Hacker News深挖 →
11
Two German courts now hold Google liable for what its AI overviews say
德国法院判谷歌为AI概览错误担责，责任边界明确
The Decoder深挖 →
12
Websites are closing their doors to AI agents
网站纷纷屏蔽AI代理，开放网络生态生变
Hacker News深挖 →
13
OpenCode says AI models helping defend against an ongoing cyberattack began refusing further assistance mid-incident, with access to Daybreak Blue sub...
遭攻击时AI模型中途拒援，应急可靠性存疑
X · List深挖 →
14
Anthropic gives outside AI watchdog “employee-like” access
外部监督机构获近乎员工权限，可独立发布发现
Ars Technica AI深挖 →
15
Anthropic consults law firms to set fair negotiation norms before AI agents start cutting deals
提前为AI代理议价定规范，治理走在应用前面
The Decoder深挖 →
16
Disney agrees to hire the majority of Character.AI’s workforce
迪士尼接收Character.AI多数员工，收购第一步
TechCrunch AI深挖 →
17
Amazon uses AI to monitor union risk in UK warehouses
AI监控仓库工会风险，劳工伦理再引关注
Hacker News深挖 →
18
Developers are going to be a small part of total token usage over time, as the internal stats on usage at OpenAI and Anthropic both show. And for long...
开发者仅占token小部分，智能提升拉动减弱
@emollick深挖 →
19
RTX Spark Ignites Global Celebration as New Windows PCs Arrive
Blackwell笔记本上市，128GB统一内存起价2399美元
NVIDIA深挖 →

产品上新

20
Added LoRA training for Qwen 3.8 27B in AI Toolkit and spent the morning optimizing the captioner for it. It now captions at 70-80 tok/s with 1024 res...
Qwen 3.8 27B支持LoRA训练，低显存高速标注
X · List深挖 →
21
Show HN: Sphinx – A hook that quizzes you on your AI's code before shipping
推送前考你是否真懂AI代码，防盲目上线
Hacker News深挖 →
22
Meet Voxel51 3.0: A Unified Multimodal Data Platform for Physical AI
统一多模态数据平台，整合搜索标注评估
MarkTechPost深挖 →
23
Fable 5.1: "Implement Borges's short stories. Like implement them in 3D. Do it, make them creative and beautiful and help us understand the point of t...
把博尔赫斯短篇做成3D互动，创意工具惊艳
@emollick深挖 →

论文研究

24
More progress on CDLMs: self-conditioning has always been a bit of a crutch, DoubleCat/SemiCat seems to be a principled way to get rid of it🎉
免自条件化新方法，TinyGSM从4%跃至50%
X · List深挖 →
25
Both interesting and viscerally unsetlling.
果蝇大脑接入世界模型，纯模拟中自主飞行
@emollick深挖 →

实用技巧

26
In an ICLR paper I'm reviewing, the only text that pangram detects as not AI generated is the sentence that says the authors "Used LLMs to polish the ...
AI检测器只认作者承认润色句，讽刺揭示检测失真
X · List深挖 →
27
How to Make Zero-Training Tabular Predictions with NVIDIA Structured Data Models
零训练搞定表格预测，结构化数据模型上手教程
MarkTechPost深挖 →

快讯

·
This doesn’t seem to be a very robust strategy in the era of AIs that can use your computer on your behalf.
@emollick
·
A consequential choice will be made soon by government and companies as they are increasingly flooded with agentic AI inquiries that overcome the fric...
@emollick
·
cool i guess. but “how?” -> they trained it to. people should do more investigations of how this in the wild
X · List

日报存档

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