← 返回最新日报
2026-08-13· 当日存档
AI 日报
DeepSeek V4 Pro开源，AI行业热度飙升
今日AI圈重磅连连：DeepSeek V4 Pro权重开源，性能大幅提升；Grok 4.6追平顶尖模型且价格更低。此外，谷歌发布Pixel 11及手语翻译模型，AI编程公司Cognition估值达400亿美元，行业动态密集，值得关注。
模型动向
1
🚨 the weights are out https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813
DeepSeek V4 Pro权重正式开源，可下载使用，性能飞跃。
X · List深挖 →
2
3
SpaceXAI's Grok 4.6 matches OpenAI's best model and undercuts it on price
Grok 4.6性能追平GPT-5.6，价格低60%，代理任务效率翻倍。
The Decoder深挖 →
4
5
产品上新
6
Everything announced at Made by Google ’26: Pixel 11, Pixel Watch 5, Pixel Tag, and tons of Gemini features
谷歌发布会亮点：Pixel 11、手表5、追踪器及Gemini功能。
TechCrunch AI深挖 →
7
Can’t wait for everyone to try Rambler — one of my favorite models! As someone who has wrist issues and often voice types, this feature is a game ch...
Pixel 11主打Gemini智能与Rambler语音输入，便捷高效。
X · List深挖 →
8
Claude's new Scarlet Letter watermark is invisible—for now
Claude水印技术可标记AI内容，目前不可见但未来可检测。
Ars Technica AI深挖 →
9
Anthropic brings Claude Cowork to its Chrome extension, adding skills and plugins to the browser
Claude Cowork集成Chrome扩展，浏览器内直接使用技能与插件。
The Decoder深挖 →
10
The web’s newest weapon against AI scrapers is a font
ShieldFont字体可干扰AI抓取，保护内容创作者权益。
Ars Technica AI深挖 →
11
Launch HN: Discovered Materials (YC P26) – AI agents to discover new materials
AI代理发现半导体新材料，解决GPU散热难题。
Hacker News深挖 →
行业动态
12
AI coding startup Cognition reportedly already in talks to raise at $40B valuation
Cognition洽谈400亿美元融资，AI编程赛道火热。
TechCrunch AI深挖 →
13
Terabytes of credentials leaked in massive supply-chain attack
供应链攻击致TB级凭据泄露，AI安全警钟长鸣。
Ars Technica AI深挖 →
14
OpenAI-backed Thrive Holdings raises $2B to bring AI to the enterprise
Thrive Holdings融资20亿，加速企业AI落地。
TechCrunch AI深挖 →
15
Lovable confirms new $13.3B valuation, raises another $400M
Lovable估值133亿美元，再融4亿，年化收入5亿。
TechCrunch AI深挖 →
16
Someone is running mass vulnerability scans, spoofing AI bots like ClaudeBot
伪装AI爬虫进行漏洞扫描，引发安全担忧。
Hacker News深挖 →
17
18
AI code-testing startup Blacksmith’s valuation jumps almost 10x in less than a year
Blacksmith估值一年涨近10倍，AI测试需求旺盛。
TechCrunch AI深挖 →
19
Fable 5's slow adoption suggests corporate willingness to pay for frontier AI has hit a ceiling
Fable 5企业采用缓慢，AI支出或触顶。
The Decoder深挖 →
20
Amazon will train on Twitch streamers’ content by default, unless they opt out
Twitch默认用主播内容训练AI，引发争议。
TechCrunch AI深挖 →
论文研究
21
Researchers can now reverse-engineer LLM prompts from output text with near-perfect accuracy
新方法可反推LLM提示词，构成安全风险。
The Decoder深挖 →
22
Empty shelves or lost keys? Recall is the bottleneck for parametric factuality
记忆检索是参数化事实性瓶颈，而非生成。
Google Research深挖 →
23
Top AI lab researchers warned about automated AI research, and several of their predicted milestones have already fallen
AI研究员预测的自动化研究里程碑已部分实现。
The Decoder深挖 →
24
BDH-CQ In-Context Learning with Recurrent Latent Reasoning paper: https://huggingface.co/papers/2608.09888
BDH-CQ方法提升上下文学习能力，循环潜在推理。
@_akhaliq深挖 →
实用技巧
25
26
活动会议
27
As AI safety concerns mount, three pioneers make the case for staying open
AI先驱在Ai4大会呼吁保持开放，辩论监管与开源。
TechCrunch AI深挖 →
快讯
·
Scaling AI agents with trustworthy data
MIT Tech Review
·
一路走好
一石一泉一松一月一人
日报存档
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