StanBlog

Vibe coding是人工智能研究员 Andrej Karpathy 于 2025 年推广的一种软件开发方法开发人员使用自然语言指导大型语言模型 (LLM) 生成、优化和调试代码，将重点从手动编码转移到更具对话性、更高级的问题解决和应用程

2025-09-05 Stan ke

Running large language models like Gemma 3 on a Mac mini M4 with only 16GB of RAM quickly forced me to confront the hard

2025-08-02 Stan ke

LLM

why you need to learn 并发与多线程? 最直观的服务器处理多客户端访问的方式，就是为每个连接创建一个线程，或者更早期操作系统里甚至是一个进程。虽然线程比进程更轻量，切换成本也更低，但每连接一个线程在并发量大时仍会面临两个

2025-07-28 计算机科学

计算机基础多线程

A new world order In the world of crypto, few figures have shaped its philosophical and technical trajectory as deeply a

2025-06-08 Stan ke

blockchain

From a young age, many of us are told: “If you work hard and prove yourself, you will be rewarded.” We are steeped in th

2025-04-13 Stan ke

feelings personal growth

Recently, I applied for a project in Tor organization, which aims to reimplement the** metrics-lib **in Rust, maintainin

2025-04-08 Stan ke

编程

进步的幻象生产力的进步，通常被视为人类社会发展的核心动力。它意味着用更少的资源和时间创造更多的价值，理应带来更高效的生产、更富裕的生活和更多的闲暇。然而，现实却常常呈现出一种令人困惑甚至绝望的景象：随着技术的飞速发展和生产效率的指数级提高

2025-04-07 Stan ke

marxism

对于 20B以上的模型 Mac mini 16GB 还是不太够。芯片虽然 CPU 单核强，NPU 功耗比也不错，但真要跑 LLM，最大瓶颈不是 CPU，而是内存带宽和总 RAM。你提到的内存带宽差异，才是关键： M4：带宽在 120GB/

2025-03-19 Stan ke

LLM

Lately, I’ve been reflecting on how deeply our upbringing, culture, and self-perception influence our present. Many of o

2025-03-11 感想

tossed no books, yet lost my pages, Watched my past slip through the ages. A younger me once held them tight, But let th

2025-03-05 感想