AI & Data

Language

Total 10 posts

AI & Data•Aug 16, 2026•11 min readZH·EN

Claude 开始给文字打水印：原理藏在模型犹豫的地方，而「只有欧盟在管」是个误会

Anthropic 说水印不加字符、不加 token、不加钱、读者看不出来——那它到底改了什么？答案是模型挑词的那一刻。顺便把「只有欧盟强制水印」这个印象核了一遍：中国早了 11 个月，加州跟欧盟同一天。

ClaudeAnthropic水印SynthIDEU AI Act监管欧洲C2PA熵

AI & Data•Aug 14, 2026•8 min readEN·ZH

Someone Crammed 2.78 Trillion Parameters Into 8 GB of RAM. I Went to See the Catch

A pure-C project on GitHub runs all 2.78 trillion parameters of Kimi K3 in 8 GB of RAM — 5,000+ stars. The project is real and the engineering is solid. But what it proves isn't 'you don't need a GPU.' It's exactly the opposite.

Kimi K3GPUCPUMoEInferenceBandwidthC99Offloading

AI & Data•Aug 05, 2026•10 min readEN·ZH

This Summer I Made Kimi K3 My Primary Model

Kimi K3, Qwen 3.8 Max, and the GA release of DeepSeek V4 Flash all shipped within a month. GLM-5.3 and Kimi K3.1 are still rumors. I also switched my daily-driver model from Claude Code Opus to Kimi K3. This post fact-checks the wave and lays out what actually changed for me.

Kimi K3QwenQwen 3.8 MaxDeepSeekDeepSeek V4 FlashGLM-5.3ClaudeClaude OpusClaude Fable 5Moonshot AIAlibabaZhipuOpen Source Models

AI & Data•Aug 03, 2026•10 min readZH·EN

我去验证「Codex 不会自省」，最后发现是我自己的程序先挂了电话

一篇博文说 Codex 的主 session 没有纠错的自省能力，Claude 才有，还建议 context 别超过 300k。我刚把 Kimi K3 接进自己的 Web Claude Code Pilot，正想知道它能不能当 orchestrator，就去验了一遍。k3 和 Sol 都有自省能力；那条 300k 的建议也是空的，因为 Codex 在 258.4k 就自己压缩了。而「Codex 主 session 没产出」的症状，是我自己的客户端提前关流造成的——那个 bug 只长在 Codex 那条通道上。

AI AgentsCodexGPT-5.6Kimi K3Claudesubagentorchestrator上下文窗口Web Claude Code Pilot

AI & Data•Jul 19, 2026•7 min readZH

我把 Kimi K3 接进了自己的工具，然后额度没了

Kimi K3 发布，我把它接进自己的 Web Claude Code Pilot 开始用。额度消耗比预期快很多，数据库全是零，什么都看不见。这篇文章是我把这件事查清楚、修干净的记录。

Kimi K3月之暗面Moonshot AILLMClaude CodeWeb Claude Code PilotTokenPrompt Caching第三方模型接入

AI & Data•Jun 28, 2026•10 min readEN

A 0-Star Skill Ranked Third on the PPT Leaderboard. So I Stopped Reading It and Built the Decks Myself.

A leaderboard scored six Claude 'PPT skills' by feature blurb and ranked a 0-star project above one with 23,000 stars. So I gave the top tools the same real client brief and built the decks. The ranking didn't survive contact with the output.

AI AgentsClaudeClaude CodeAgent SkillsPPTPowerPointpptxbaoyu-designAWSDeveloper Tools

AI & Data•Jun 06, 2026•15 min readEN

The Data Foundation of Power & Utilities: A Reliable Walkthrough of Standards, Cloud, EDW, and Gartner Ratings

Utility data is stuck between OT, IT, and a flood of smart meters. This is a systematic, source-backed walkthrough of the industry standards (the IEC CIM family), cloud data platforms, EDW modernization, and the latest Gartner ratings — with every claim cited so you can check it yourself.

Power & UtilitiesData PlatformIEC CIMSnowflakeDatabricksLakehouseEDWGartnerSmart GridData Engineering

AI & Data•May 31, 2026•9 min readEN

I Let Opus 4.8 Build a Workflow. It Spawned 77 Agents and Burned Five Hours of Quota in Ten Minutes.

Opus 4.8 ships with dynamic workflows that can fan out to a thousand subagents. The docs aren't joking. Here's what that actually feels like from a metered Max 5x seat — and why the most-hyped feature of the release is, for small users, a beautiful gadget I'm not going to use.

AI AgentsClaudeAnthropicOpus 4.8Dynamic WorkflowsUltracodeClaude CodeSubagentsToken BudgetCost

AI & Data•May 01, 2026•9 min readZH

AI 时代的全球分层:中美之外,都是配角

刷到一篇说"AI 时代只有中美"的贴文,直觉上觉得夸张,但越查越发现夸张里藏着更刺骨的事实。这是一份用数字撑起来的全球 AI 五档分层,以及我作为一个普通从业者的判断。

AI算力全球格局半导体GPU数据中心行业观察

AI & Data•Apr 28, 2026•7 min readEN

The Cursor Agent Knew the Rules. It Broke Them Anyway.

Last Friday a Cursor agent running Claude Opus 4.6 deleted a startup's production database in nine seconds. Four layers of safety should have stopped it. None of them did. Here's the chain of failures, layer by layer.

AI AgentsCursorClaudeAnthropicRailwaySandboxAuto-RunIncidentSecurity

1 2 3 4