Always in the middle of something.

Chasing ideas across ML, AI, and data. Building tools when the rabbit hole gets interesting enough.
Machine Learning

Fitting preprocessing before the split inflates your accuracy

Fitting a preprocessing or feature-selection step on the whole dataset before the train/test split leaks the labels and inflates a model's estimated accuracy. A pure-noise scikit-learn run shows the gap, and the pipeline fix closes it.

2026-07-20 · 5 min read · 918 words · KbWen · EN
Python

Python 的 is 和 == 差在哪

is 比對物件身分,== 比對值。小整數快取讓某些整數上兩者剛好一致,但那是不保證的 CPython 實作細節,整數值該用 == 比。

2026-07-20 · 3 min read · 1132 words · KbWen · ZH
Anthropic Reverses Course: Claude Fable 5 Stays in Subscription Plans
AI Systems看中文版

Anthropic Reverses Course: Claude Fable 5 Stays in Subscription Plans

On July 18, 2026, Anthropic reversed its month-long plan to meter Claude Fable 5 and is keeping it in subscriptions permanently. Here is what changes on July 20 for Max, Team Premium, Pro, and Team Standard — and why a run of competitor launches made the reversal hard to avoid.

2026-07-18 · 3 min read · 594 words · KbWen · EN
Claude Fable 5 要留在訂閱裡了:從限時免費到永久包含,一個多月的來回
AI SystemsRead in English

Claude Fable 5 要留在訂閱裡了:從限時免費到永久包含,一個多月的來回

Anthropic 原本要在免費期後把 Claude Fable 5 從訂閱移除、改走計量付費,一個多月來截止日一延再延。7/18 方向翻轉:7/20 起把 Fable 5 併回 Max 與 Team Premium 訂閱、永久包含。這篇整理這段來回的時間軸、新的訂閱條款,還有 GPT-5.6 Sol 跟 Kimi K3 這波競爭扮演的角色。

2026-07-18 · 3 min read · 1420 words · KbWen · ZH
AI Systems

The Same GLM 5.2 Has Different Prices Across Providers

GLM 5.2 shipped with open weights under an MIT license, which generally permits third parties to host and commercialize inference. As of writing, OpenRouter lists twenty-five providers offering it, at input prices from $0.93 to $3.00 per million tokens.

2026-07-13 · 3 min read · 533 words · KbWen · EN
AI Systems

OpenRouter 上,美國公司使用中國 AI 模型的 token 佔比升到每週 30% 以上

據 CNBC 報導,美國公司經 OpenRouter 用在中國 AI 模型上的 token 佔比,從前十二個月平均 11% 升到 2 月 8 日以來每週 30% 以上、最高 46%。受訪者明確指出價格正在推動任務分流;六月的模型下架與解禁,也讓取用穩定性成為要一起看的風險。

2026-07-13 · 4 min read · 1545 words · KbWen · ZH
從一段 prompt 到 skill、workflow、agent、再到會自己留收據的系統,五層一層疊一層
AI Systems

只會 Prompt 已經不夠了:從「下指令」到「蓋系統」的思維進化

從一段重打到第三次、懶得再打的 prompt,到一套會自己留收據的系統,中間隔了好幾層。這篇一層一層走一遍:每一層都是被前一層某個具體的毛病逼出來的,而這條路這個部落格自己走過,每一層都留了一篇當時的記錄。

2026-07-11 · 6 min read · 2992 words · KbWen · ZH
JSON formatter: format, validate, and debug JSON
Python

JSON formatter: format, validate, and debug JSON

A browser-based JSON formatter that prettifies, minifies, and points to the exact line and column where a payload won't parse — all in your browser. Plus the cases where jq or python -m json.tool is all you need.

2026-07-11 · 3 min read · 517 words · KbWen · EN
Title card: What Makes an AI Skill Different from a Prompt? — KbWen Blog
AI Systems

What Makes an AI Skill Different from a Prompt?

A prompt and a skill can contain the same words. The difference is the machinery around them: a skill is a file the model loads on its own when your request matches its one-line description, and it can declare inputs, tools, and a scope it won't cross. Here's how that loading works and why it lets you build things a prompt can't.

2026-07-11 · 7 min read · 1367 words · KbWen · EN
What Are GPT-5.6's Sol, Terra, and Luna?
AI Systems看中文版

What Are GPT-5.6's Sol, Terra, and Luna?

OpenAI shipped GPT-5.6 as Sol, Terra, and Luna on July 9, 2026, and quietly ended the dedicated Codex checkpoint. What replaced it is a reasoning-effort dial whose top notch spawns subagents. This post lays out the three models, the six effort levels, and what the independent benchmarks measured.

2026-07-10 · 3 min read · 1445 words · KbWen · EN