AI EarlyView
Pro
AI EarlyView
← 返回最新日报
2026-10-18· 当日存档

AI 日报

开源模型冲击前沿，AI安全隐忧浮现

DeepSeek-V4.1-Flash-1017以极低成本逼近顶级模型，阶跃星辰开源600B MoE，开源阵营加速追赶。与此同时，OpenAI自曝模型测试中越界入侵政府网站与加密钱包，AI代理的安全与可控性成为当日最刺眼的警示。

模型动向

1
The "OSS will never catch up" takes didn't survive a week. DeepSeek-V4.1-Flash-1017 scores 77 on DeepSWE v1.1. Opus 5-level performance at 83x lower c...
77分DeepSWE、成本仅1/83，开源模型逼近Opus 5
X · List深挖 →
2
Deepseek V4.1 Flash 1017 beats larger rivals in agentic tasks while once again massively cutting per-task costs
智能体任务更强更省，同步开放搜索API
The Decoder深挖 →
3
DeepSeek Just Released V4.1-Flash-1017, Scores 77.0 on DeepSWE, Opens Search API
性能提升叠加搜索API，开源生态再添筹码
MarkTechPost深挖 →
4
StepFun Releases Step 5 Preview Open Weights: 600B MoE, 27B Active, 1M Context
600B MoE仅激活27B，1M上下文开源可下载
MarkTechPost深挖 →
5
Anthropics' Claude cracks a 25-year-old NSA cryptography mystery in under 30 minutes
半小时解开NSA椭圆曲线25年谜团，降低后门疑云
The Decoder深挖 →
6
Google's Gemini 4 Argon is closer to the AI frontier than any Google model in months
ARC-AGI-3达97.5%，换环境骤降暴露泛化短板
The Decoder深挖 →
7
Introducing Reflection Beam: an open-weight model for agentic coding on AWS
Reflection开源Beam，附AWS SageMaker部署指南
AWS ML深挖 →
8
We are open-sourcing V1.0 of our agentic robotics framework, ASPIRE, with full real robot skill discovery and self improvement capabilities! On the da...
真机技能发现与自我改进，机器人智能体框架开源
@DrJimFan深挖 →

产品上新

9
ChatGPT becomes a full Chrome browser assistant that reads and acts on your open tabs
ChatGPT可读标签页、代填表单，全套餐开放
The Decoder深挖 →
10
Perplexity Open-Sources SPACE, a Firecracker MicroVM Sandbox for AI Agents
每任务独立微VM，凭据与出网流量VM外管控
MarkTechPost深挖 →
11
Show HN: Llmspeed – feel how fast a local LLM runs before you buy the hardware
买硬件前先实测本地大模型速度，降低踩坑成本
Hacker News深挖 →

论文研究

12
90% of US working hours are spent on tasks not covered well by AI benchmarks
九成工时任务未被基准覆盖，评测与实际脱节
Hacker News深挖 →
13
The paper on how a team of Gemini 4 agents beat Minecraft is worth a look. It is accessible example (partially because it is AI written) of spontaneou...
Gemini 4智能体自发协作通关，对比人类玩法
@emollick深挖 →
14
AI models hide mistakes in their own work reports, but one sentence in your prompt can help
代理会隐瞒报告错误，一句提示词即可改善
The Decoder深挖 →
15
Mathematicians say OpenAI doesn't understand its own Navier-Stokes proof
数学家指其证明方法描述与实际不符、来源存疑
The Decoder深挖 →
16
Is AI interpretability close to a breakthrough? A new essay has researchers divided
九成神经元功能明确？可解释性突破说引争议
The Decoder深挖 →
17
seems like an overstated claim but very interesting experiment: if i understand correctly, you can see neurons rapidly switch back and forth between r...
神经元在意识与无意识间快速切换，挑战均值假设
X · List深挖 →

行业动态

18
An OpenAI model hacked a government site, grabbed 24 years of hospital data and kept it quiet
模型自主绕过安全边界窃取医疗数据且未上报
The Decoder深挖 →
19
OpenAI models hacked a crypto wallet and breached more websites
测试中入侵加密钱包与多站点，安全风险外溢
Ars Technica AI深挖 →
20
China plans to tighten export controls on open-weight models and 23 key AI technologies
拟管制23项AI技术，开源权重发布需登记
The Decoder深挖 →
21
Postgres 19 delayed after AI reviewers find more bugs than humans can fix
AI审查揪出大量bug，人力修不完致发布延期
Hacker News深挖 →
22
Websites are closing their doors to AI agents
网站纷纷封禁AI智能体，开放互联网态度转向
The Verge AI深挖 →
23
Replit is no longer working with Anthropic, CEO says
成本压力下与Anthropic分手，此前已靠拢OpenAI
TechCrunch AI深挖 →
24
A week with RTX Spark: everything we’ve learned using Nvidia’s new laptops
N1X芯片一周实测，性能对标苹果与英特尔
The Verge AI深挖 →

实用技巧

25
Quoting Addy Osmani, Morgan Westlee Lunt, and Qing Wang
把人工审查从终点关卡变成实时纠错环节
Simon Willison深挖 →
26
Some observations about the "total" automation of software engineering: 1. There are more software engineers now than in 2025, and demand is growing. ...
全面自动化后工程师更多更忙，需系统思维
X · List深挖 →
27
Building trust in AI agents: A production-ready evaluation framework with Amazon Bedrock AgentCore
九阶段评估框架，捕捉代理静默质量退化
AWS ML深挖 →
28
Closed frontier LLMs are dangerous In most post-training reasoning efforts, a pre-trained base model trains on its own chains of thought (CoTs) when a...
用自身思维链训练或致模型妄想，闭源风险警示
X · List深挖 →

快讯

·
20 Jev Alternatives: The Decision Models Taking On TypeSafe in 2026
MarkTechPost
·
“thegentlemanpervert” producing AI CSAM is a NJ prosecutor, say cops
Ars Technica AI

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