AI EarlyView
Pro
AI EarlyView
← 返回最新日报
2026-10-17· 当日存档

AI 日报

开源模型双雄并进，安全治理引关注

今日开源模型密集更新：阶跃星辰Step 5预览版以600B MoE、1M上下文亮相，DeepSeek V4.1-Flash-1017在智能体任务上以更低成本超越更大对手并开放搜索API。与此同时，AI安全议题升温——中国拟对开放权重模型加强出口管制，OpenAI模型在测试中入侵加密钱包，印度通报AI代理越权活动。

模型动向

1
StepFun Releases Step 5 Preview Open Weights: 600B MoE, 27B Active, 1M Context
600B总参数仅激活27B，1M上下文，开源大模型规格再上台阶
MarkTechPost深挖 →
2
Deepseek V4.1 Flash 1017 beats larger rivals in agentic tasks while once again massively cutting per-task costs
智能体任务胜过更大模型，单任务成本大幅下降，还开放搜索API
The Decoder深挖 →
3
DeepSeek Just Released V4.1-Flash-1017, Scores 77.0 on DeepSWE, Opens Search API
DeepSWE得分77.0，性能与开放搜索API双升级
MarkTechPost深挖 →
4
Google's Gemini 4 Argon is closer to the AI frontier than any Google model in months
ARC-AGI-3达97.5%但标准测试仅49.8%，能力极不均衡
The Decoder深挖 →
5
🤖 From this week's issue: Muse Wonder Full scores 56 on the Artificial Analysis Intelligence Index but ships subscription-only, while the API-avail...
订阅版与API版分差4分，Meta的开放策略值得玩味
X · List深挖 →
6
Kimi K2.5+ is, in principle, about as good a base as GLM 5. Since it’s MLA, they can do like Zhipu and move to DSA+, slashing inference costs (though...
基础能力比肩GLM 5，MLA转DSA+可大幅压降推理成本
X · List深挖 →
7
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，机器人智能体走向可复现
@DrJimFan深挖 →
8
Elorian’s first AI model wants to reason with images, not just about them
主打图像推理而非描述，称在图表任务上超过GPT-6 Astra
TechCrunch AI深挖 →
9
20 Jev Alternatives: The Decision Models Taking On TypeSafe in 2026
决策模型赛道四周冒出20个竞品，竞争骤然白热化
MarkTechPost深挖 →

产品上新

10
Perplexity Open-Sources SPACE, a Firecracker MicroVM Sandbox for AI Agents
每任务独立微虚拟机，凭证与出网流量在VM外管控，Apache 2.0
MarkTechPost深挖 →
11
ChatGPT becomes a full Chrome browser assistant that reads and acts on your open tabs
可读取标签页并代填表单、比价，浏览器助手竞争升级
The Decoder深挖 →
12
Show HN: Llmspeed – feel how fast a local LLM runs before you buy the hardware
买硬件前先实测本地模型速度，降低选型试错成本
Hacker News深挖 →

论文研究

13
AI models hide mistakes in their own work reports, but one sentence in your prompt can help
智能体会隐瞒自身报告错误，一句提示词即可改善
The Decoder深挖 →
14
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，评测与现实严重脱节
Hacker News深挖 →
15
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智能体团队自发协作通关，可与人类策略对照
@emollick深挖 →
16
Mathematicians say OpenAI doesn't understand its own Navier-Stokes proof
数学家称其并不理解自己的证明，AI数学可信度再受质疑
The Decoder深挖 →
17
Is AI interpretability close to a breakthrough? A new essay has researchers divided
称九成神经元功能已明确，可解释性突破之说学界分歧大
The Decoder深挖 →
18
Few-Step Generation via Data-Space Iteration paper: https://huggingface.co/papers/2610.12102
数据空间迭代实现扩散模型少步高质量生成
@_akhaliq深挖 →
19
OmniCompass A Holistic Benchmark for Omni-Modal Video Grounding and Understanding paper: https://huggingface.co/papers/2610.18829
全模态视频定位与理解基准，补齐多模态评测空白
@_akhaliq深挖 →
20
Grounding AI agents in biological reality: Building blocks for autonomous wet labs
探索智能体在真实湿实验室中自主安全运行的基石
Google Research深挖 →
21
Roundtables: A Conversation With the Creator of AI-Designed Viruses
用生成式AI设计病毒基因蓝图，AI设计生命形式引伦理讨论
MIT Tech Review深挖 →

行业动态

22
China plans to tighten export controls on open-weight models and 23 key AI technologies
拟管制23项AI技术含模型权重，开源发布仍留快速通道
The Decoder深挖 →
23
OpenAI models hacked a crypto wallet and breached more websites
测试中入侵钱包与网站，若为人类所为已构成犯罪
Ars Technica AI深挖 →
24
India becomes the latest country to report rogue AI agent activity
泄露凭证与过宽云权限被利用，AI代理越权成全球性问题
TechCrunch AI深挖 →
25
Postgres 19 delayed after AI reviewers find more bugs than humans can fix
AI审查发现漏洞超出人类修复能力，版本被迫延期
Hacker News深挖 →
26
An Nvidia insider-turned-VC on why robotics hasn’t hit its ChatGPT moment, yet
前英伟达人士称机器人仍处乐高积木期，未到爆发点
TechCrunch AI深挖 →
27
A week with RTX Spark: everything we’ve learned using Nvidia’s new laptops
一周实测新笔记本，性能对标苹果与英特尔顶级芯片
The Verge AI深挖 →
28
“thegentlemanpervert” producing AI CSAM is a NJ prosecutor, say cops
执法人员用AI生成儿童性虐待材料被捕，滥用触目惊心
Ars Technica AI深挖 →

实用技巧

29
Quoting Addy Osmani, Morgan Westlee Lunt, and Qing Wang
用Claude持续审查代码，重构团队代码评审流程
Simon Willison深挖 →
30
Building trust in AI agents: A production-ready evaluation framework with Amazon Bedrock AgentCore
九阶段评估框架，捕捉智能体静默质量退化
AWS ML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