# konakona

> A bilingual personal blog by konakona (Samuel He), covering programming, AI, and research notes.

Chinese articles live under /zh/ and English articles under /en/. The root uses browser language detection; start at a localized archive instead.
Each article has a static Markdown representation at its HTML URL followed by index.md. Metadata identifies the actual content language, canonical HTML URL, and available translation.
English article routes may show a Chinese original when no English translation exists. The lists below include each published language version once.

## Navigation

- [Chinese archive](https://blog.konakona.dev/zh/posts/): All Chinese articles.
- [English archive](https://blog.konakona.dev/en/posts/): English articles, with Chinese fallbacks.
- [Chinese RSS](https://blog.konakona.dev/rss.xml)
- [English RSS](https://blog.konakona.dev/en/rss.xml)
- [Sitemap](https://blog.konakona.dev/sitemap-index.xml)

## Chinese articles

- [Claude 根据我的笔记做的两个页面，以及对 Opus 5.5 的一些感想](https://blog.konakona.dev/zh/posts/notes-insights-with-claude/index.md): Opus 5.5 基于我的笔记库 One-Shot 出的两个交互页面，以及对它的 taste、交流体验和 Claude Code 的感想
- [让 Tailscale Exit Node 的流量经过 Mihomo](https://blog.konakona.dev/zh/posts/tailscale-exit-node-through-mihomo/index.md): 救援通道、最小范围的 TUN 接管、DNS 污染排查、最终实现通过 Exit Node 使用 Mihomo 代理
- [Qwen3.8-27B Terminal-Bench 2.1 复现：环境、完整流程与踩坑记录](https://blog.konakona.dev/zh/posts/qwen38-terminal-bench-21-reproduction/index.md): 使用 SGLang、Harbor 0.21.0 和 Terminus-2，在 16 张 A100 40GB GPU 上复现 Qwen3.8-27B Terminal-Bench 2.1 的完整过程、问题排查与结果审计。
- [NITP: Next Implicit Token Prediction for LLM Pre-training](https://blog.konakona.dev/zh/posts/nitp/index.md): 在输出空间和隐空间中同时预测下一个 token。
- [从线性注意力到 Test-Time Training](https://blog.konakona.dev/zh/posts/from-linear-attention-to-test-time-training/index.md): 线性注意力将标准注意力的平方复杂度降低为线性，这很好。但我们能做得更好吗？
- [一次 OpenClaw web_fetch 故障排查](https://blog.konakona.dev/zh/posts/openclaw-webfetch-debug/index.md): proxychains4 的 DNS 劫持和 OpenClaw 的 SSRF 保护撞在一起，导致 web_fetch 彻底罢工。
- [主成分分析（PCA）直观理解](https://blog.konakona.dev/zh/posts/pca-intuitive/index.md): 用方差最大化、协方差矩阵以及谱定理，直观理解主成分分析（PCA）。
- [读图灵《Computing Machinery and Intelligence》的几点感想](https://blog.konakona.dev/zh/posts/thoughts-turing-can-machines-think/index.md): 对图灵 1950 年经典论文《Computing Machinery and Intelligence》的若干思考：模仿游戏、人与机器逻辑差异、离散与连续状态、以及机器智能实现路径。
- [编写 Swift API 客户端库（二）](https://blog.konakona.dev/zh/posts/swift-api-client-part-2/index.md): 学习如何利用泛型、Swift Concurrency 以及 Codable 协议，优雅高效地编写 Web API 客户端。
- [编写 Swift API 客户端库（一）](https://blog.konakona.dev/zh/posts/swift-api-client-part-1/index.md): 学习如何利用泛型、Swift Concurrency 以及 Codable 协议，优雅高效地编写 Web API 客户端。

## English articles

- [Two pages Claude built from my notes, and some thoughts on Opus 5.5](https://blog.konakona.dev/en/posts/notes-insights-with-claude/index.md): Two interactive pages Opus 5.5 one-shotted from my notes repo, plus my impressions of its taste, how it communicates, and Claude Code
- [Routing Tailscale exit node traffic through Mihomo](https://blog.konakona.dev/en/posts/tailscale-exit-node-through-mihomo/index.md): A rescue tunnel, narrowly scoped TUN interception, and a DNS poisoning investigation that made a Tailscale exit node work with Mihomo
- [Reproducing Qwen3.8-27B's Terminal-Bench 2.1 Results: Setup, Full Run, and Pitfalls](https://blog.konakona.dev/en/posts/qwen38-terminal-bench-21-reproduction/index.md): A complete reproduction of Qwen3.8-27B on Terminal-Bench 2.1 using SGLang, Harbor 0.21.0, and Terminus-2 across 16 A100 40GB GPUs, including setup, debugging, and result auditing.
- [NITP: Next Implicit Token Prediction for LLM Pre-training](https://blog.konakona.dev/en/posts/nitp/index.md): Predicting the next token in both the output space and the hidden space.
- [From Linear Attention to Test-Time Training](https://blog.konakona.dev/en/posts/from-linear-attention-to-test-time-training/index.md): Linear attention reduces the quadratic cost of standard attention to linear, which is great. But can we do better?
- [Debugging an OpenClaw web_fetch Failure](https://blog.konakona.dev/en/posts/openclaw-webfetch-debug/index.md): proxychains4 DNS hijacking collided with OpenClaw's SSRF protection, leaving web_fetch completely broken.
- [Principal Component Analysis, Explained Intuitively](https://blog.konakona.dev/en/posts/pca-intuitive/index.md): An intuitive explanation of PCA for dimensionality reduction, covering variance maximization, covariance matrices, and the spectral theorem.
- [Lecture 6: Language](https://blog.konakona.dev/en/posts/cs50-ai-notes/6-language/index.md): CS50 AI lecture notes on NLP: context-free grammars, n-grams, bag-of-words Naive Bayes, embeddings, word2vec, sequence models, attention, and transformers.
- [Lecture 5: Neural Networks](https://blog.konakona.dev/en/posts/cs50-ai-notes/5-neural-networks/index.md): CS50 AI lecture notes on neural networks, activation functions, backpropagation, and convolutional architectures.
- [Lecture 4: Learning](https://blog.konakona.dev/en/posts/cs50-ai-notes/4-learning/index.md): CS50 AI lecture notes on nearest neighbors, perceptrons, SVMs, reinforcement learning, and clustering.
- [Lecture 3: Optimization](https://blog.konakona.dev/en/posts/cs50-ai-notes/3-optimization/index.md): CS50 AI lecture notes on local search, simulated annealing, linear programming, and CSPs.
- [Crossword Generator](https://blog.konakona.dev/en/posts/cs50-ai-notes/crossword-generator/index.md): CS50 AI project notes on solving crosswords as a CSP with heuristics and AC-3.
- [Lecture 2: Uncertainty](https://blog.konakona.dev/en/posts/cs50-ai-notes/2-uncertainty/index.md): CS50 AI lecture notes on probabilistic reasoning, Bayes' rule, Bayesian networks, and HMMs.
- [Lecture 1: Knowledge](https://blog.konakona.dev/en/posts/cs50-ai-notes/1-knowledge/index.md): CS50 AI lecture notes on propositional and first-order logic, entailment, and resolution.
- [Minesweeper AI](https://blog.konakona.dev/en/posts/cs50-ai-notes/minesweeper-ai/index.md): CS50 AI project notes on knowledge-based inference for Minesweeper.
- [Thoughts on Computing Machinery and Intelligence by Alan Turing](https://blog.konakona.dev/en/posts/thoughts-turing-can-machines-think/index.md): Reflections on Turing's seminal 1950 paper discussing machine intelligence, the Imitation Game, and the nature of human vs. machine logic.
- [Lecture 0: Search](https://blog.konakona.dev/en/posts/cs50-ai-notes/0-search/index.md): CS50 AI lecture notes on search algorithms, from BFS/DFS to A* and minimax.
- [Building a Swift API Client Library: Part 2](https://blog.konakona.dev/en/posts/swift-api-client-part-2/index.md): Learn how to build a clean, layered API client in Swift using powerful features like Generics, Codable, and Swift Concurrency.
- [Building a Swift API Client Library: Part 1](https://blog.konakona.dev/en/posts/swift-api-client-part-1/index.md): Learn how to build a clean, layered API client in Swift using powerful features like Generics, Codable, and Swift Concurrency.
- [Security & Cryptography](https://blog.konakona.dev/en/posts/missing-semester-notes/security-and-cryptography/index.md): MIT Missing Semester Notes on Security & Cryptography
- [Shell & Command-Line Tools](https://blog.konakona.dev/en/posts/missing-semester-notes/shell-and-command-line-tools/index.md): MIT Missing Semester Notes on Shell & Command-Line Tools