📈 Data to start your week｜Kimi-K3 编码、DeepSeek 收入与 AI 安全测试｜英文原文 + 中文翻译

English original

📈 Data to start your week

Kimi codes ⬆ Lithium recovery ⬆ Confidence in tech ⬇

1. Kimi can code. Kimi-K3, released last week, overtakes Fable 5 and GPT-5.6 Sol to take first place on the frontend code arena benchmark. 1
2. Demand unlocked. DeepSeek nears $500 million in annualized revenue with gross margins as high as 70-80% on its V4 model. 1
3. Ask and receive. Tests of leading AI models on prompts based on real terrorist cases show that a third of responses would’ve provided useful help to attackers, and labeling the same prompts as "research" increased compliance from 17% to 42%. 1

Public access boundary

The publicly readable page ends after these three signals. It then asks readers to continue in the Substack app or purchase a paid subscription. The rest of the post is not accessible here and is not translated or supplemented.

中文翻译

📈 开始新一周的数据

Kimi 编码能力上升，锂回收上升，对科技的信心下降

1. Kimi 会写代码。 上周发布的 Kimi-K3 超过 Fable 5 和 GPT-5.6 Sol，在 Frontend Code Arena 基准测试中排名第一。1
2. 需求被打开。 DeepSeek 的 V4 模型年化收入接近 5 亿美元，毛利率最高可达 70% 至 80%。原文保留了「最高可达」这一限定。1
3. 提出请求，就能得到帮助。 对基于真实恐怖主义案件设计的提示词进行测试后，领先 AI 模型约有 三分之一的回答可以为攻击者提供有用帮助；把同一批提示词标注为「research」后，模型的服从率从 17% 上升到 42%。1

公开内容边界

公开页面在以上三条数据之后停止，随后提示读者在 Substack 应用中继续阅读，或购买付费订阅。后续正文在当前可见范围之外，本文不翻译，也不补写。