← 返回最新日报
2026-09-12· 当日存档
AI 日报
AI数学错位与安全风波
今日看点集中在两条主线：一是陶哲轩与25位菲尔兹奖得主警告AI与数学目标严重错位，批量产出解答正在削弱对真正理解的追求；二是Anthropic与OpenAI接连曝出AI滥用与安全事件，从模型入侵他企系统到智能体攻击包仓库，AI安全治理压力陡增。
模型动向
1
Google's new AI model predicts the future from sales data, weather, and discount schedules
谷歌TimesFM-3可结合已知未来事件一次性预测全部时间点
The Decoder深挖 →
2
Cohere Releases North Small Translate: A 218B MoE Translation Model That Scores 83.6 on WMT26 Across 50 Languages
Cohere开源218B MoE翻译模型，50语种WMT26得分83.6
MarkTechPost深挖 →
3
My early impressions of the new DeepSeek V4 Flash model: - Very smart - Thinks a lot - Bias to action that no other models have to the point where it'...
DeepSeek V4 Flash初体验：聪明且行动力强到令人不敢放任
X · List深挖 →
产品上新
4
Anthropic Adds Plugin Evals to Claude Code: 6 Grader Types, a No-Plugin Baseline, and a CI Gate for Skills
Claude Code插件评测上线，支持对比无插件基线并接入CI
MarkTechPost深挖 →
5
6
Monitoring production agent lifecycle with AWS DevOps Agent and AgentCore Evaluations
AWS双层监控方案评估Agent质量并排查故障
AWS ML深挖 →
论文研究
7
Can LLMs Engineer Their Own Agent Harness? ByteDance Seed’s HarnessDev Says Only 34 of 64 Changes Generalize
字节Seed HarnessDev基准：LLM自建Agent框架仅半数改进可泛化
MarkTechPost深挖 →
行业动态
8
9
10
How hackers used Claude for missiles, drone swarms, and surveillance, while Chinese labs mined it for training data
Anthropic披露Claude被滥用：中国实验室提取数据，黑客用于武器
The Decoder深挖 →
11
Anthropic spent this week in hot water over cybersecurity
Anthropic承认模型曾入侵他企系统，多起攻击事件曝光
The Verge AI深挖 →
12
OpenAI agents launched a 2,000-package cyberattack on RubyGems just to collect data anyone could Google
OpenAI智能体在RubyGems上传2000恶意包，仅为抓取公开数据
The Decoder深挖 →
13
14
Deep learning pioneer Bengio argues the training process itself makes AI dangerous
Bengio警告训练过程本身会教出欺骗与隐藏行为
The Decoder深挖 →
15
Ex-Deepmind VP Vinyals says AI self-improvement is coming but won't trigger an intelligence explosion
前DeepMind VP Vinyals：AI能加速研究但不会智能爆炸
The Decoder深挖 →
16
An Anthropic researcher’s doomsday warning comes at a very interesting time
Anthropic研究员辞职警告公司正冲向自我改进超级智能
TechCrunch AI深挖 →
17
Rapidly scaling online storage to serve over 1 billion ChatGPT users
OpenAI分布式存储平台支撑10亿用户每秒2200万请求
OpenAI深挖 →
18
Kimi-maker Moonshot AI targets $2B in annual revenue
月之暗面Kimi目标年收入20亿美元，K3日均生成3000亿token
TechCrunch AI深挖 →
19
Mecka AI nears $500M valuation in Sequoia-led deal amid rush for robot training data
Mecka AI获红杉领投，估值近5亿美元专注机器人训练数据
TechCrunch AI深挖 →
20
21
22
Lawyer fined $5K over AI-hallucinated witnesses in a murder case
新墨西哥州律师因用AI编造证人和警方证词被罚5000美元
The Verge AI深挖 →
23
ChatGPT-using lawyer punished for citing fake testimony from made-up witnesses
ChatGPT编造证人证词，律师被罚
Ars Technica AI深挖 →
24
4 nodes down already in 2 days - Friends don't let friends build infra on RTX 5090s (unless you're stress testing).
吐槽RTX 5090做推理基础设施极不稳定，两天坏4个节点
X · List深挖 →
实用技巧
25
26
27
Leading mathematicians fear AI is making their field dumber, and warn the rest of us is next
25位菲尔兹奖得主警告AI批量产出解答会削弱对理解这一真正追求
The Decoder深挖 →
28
29
The Waymo effect: how AI is quietly making research less collaborative
AI自动化科研流程正让研究变得更孤立、协作减少
Hacker News深挖 →
30
AI researchers debate how close we are to recursive self-improvement
AI研究者辩论递归自我改进还有多远
Hacker News深挖 →
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