Cherish's Blog

meta prompts * skill creator: * copilot skill creator: https://github.com/github/awesome-copilot/blob/main/skills/make-skill-template/SKILL.md * claude skill creator: https://skillsmp.com/zh/skills/openclaw-openclaw-skills-skill-creator-skill-md * agent skills: https://github.com/github/awesome-copilot/blob/main/instructions/agent-skills.instructions.md * from coding agent: https://github.com/obra/superpowers/blob/main/skills/writing-skills/SKILL.md * copilot-instruction: * https://github.com/github/awesome-copilot/blob/main/prompts/copilot-instructions-blueprint-generat

Links * claude plugin: https://awesomeclaude.ai/ralph-wiggum * ralph wiggum loop from Geoffrey: https://github.com/ghuntley/how-to-ralph-wiggum/tree/main * video from Geoffrey: https://www.youtube.com/watch?v=4Nna09dG_c0

how to build a chatgpt pretraining -> base model internet text prediction/inference training. —> internet document simulator (just like “exposition in book”) post training (supervised fine tuning) -> SFT model conversation training. —> an assistent (just like “example problem & solution in book”) Same algorithm as pretraining. Just different datasets. We can add different kind of conversions for LLM to learn and simulate later. These conversions can be used to * know how to answer questions (simulate the conversation vibe) * solve hallucination, * LLM self recogntion, * etc. reinfo

MCP (Model Context Protocol) Summary * model: means machine learning models * model context protocol: the standard to adding more context ability for ai model. It’s an open-source standard for connecting AI applications to external systems. * the protocol contains: * specification (schema) of client/server implementation requirements: * JSON-RPC: as the main communication media type * lifecyce management * authorization * transports: how client & server communicate

Abstract We can implement copilot extensions so that it can use copilot llm abilities and do special things for us. There are several ways to implement the extensions: 1. client side extensions: Now only vscode support to add copilot extensions. It’s running locally in your machine, and thus use your local resource and can access all your local workspace and utilize your local installed tools. Ideal for tasks deeply embedded in developer workflows, like editing, real-time code generation, or using local context 2. server side extensions: it’s executed on your backend server. Since

JVM is the heart of Java programming language. When we execute a Java program, JVM is responsible for converting the byte code to the machine-specific code. JVM is also platform-dependent and provides core java functions such as memory management, garbage collection, security, etc. Garbage Collection In Java, a developer does not need to explicitly allocate and deallocate memory – the JVM and more specifically the Garbage Collector – has the duty of handling memory allocation so that the developer doesn’t have to. For most collectors GC related pauses are proportional to size of their heaps

Core Java Tutorial | DigitalOcean Thread Process: A process is a self contained execution environment and it can be seen as a program or application. Thread: lightweight. * a process can have multiple threads running, and at least one main thread. * all thread share parent process code and data * thread creation, context switching, and intercommunication are less expensive * Multithreading: In multi-core system, more than one thread can run at the exactly same time. In single-core system, more than one thread can run parallel using OS feature time slicing to shar

react 提倡组件化开发，以促进复用。react native 也一样是组件化开发思想。不同的是，react 中是使用原生的 html 组件作为基本组件（div、a…），而 react native 使用的是另一些原生组件。用户的自定义组件也是基于这些原生组件。 所以要用 react native，必须了解这些原生组件（就跟学 html 组件差不多） general props 一些所有组件或大部分组件都有的属性。 style 所有的核心组件都接受名为 style 的属性（类似 html 标签中的 html）。这些样式名基本上是遵循了 web 上的 CSS 的命名，只是按照 JS 的语法要求使用了驼峰命名法，例如将 background-color 改为backgroundColor。 实际开发中组件的样式会越来越复杂，我们建议使用StyleSheet.create来集中定义组件的样式。常见的做法是按顺序声明和使用 style 属性，以借鉴 CSS 中的“层叠”做法（即后声明的属性会覆盖先声明的同名属性）。比如像下面这样： 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 import React, { Component } from 'react';

为什么 java stream parallel 有时比 sequential 执行还慢？ 场景 考虑下边的代码，并行执行不一定比顺序执行快，甚至很多时候都是更慢的。 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 @Test public void should_not_sure_if_without_warm_up() { String[] array = new String[1000000]; Arrays.fill(array, "AbabagalamagA"); System.out.println("Benchmark..."); for (int i = 0; i < 5; ++i) { System.out.printf("Run %d: sequential %s - parallel %s\n", i, test(() -> sequential(array)), test(() -> parallel(array))); }

设计 收集一些在设计时可能会用到的比较好的网站 * icons from the fontawesome: 各种各样的图标，很漂亮 * ant design：里边有 UI 设计价值观及图标资源等，还有前端组件库 前端组件库 组件库大合集 组件库大合集2 mac 强迫症的Mac设置指南 terminal [10 Must know terminal commands and tips for productivity](10 Must know terminal commands and tips for productivity) 介绍了一些，简单罗列下： 1. iterm2：是一种 terminal，将其作为默认 terminal，可以方便的分屏等 2. oh my zsh：管理 zsh 配置的，使用 ~/.zshrc，利用 source ~/.zshrc 可以是配置立马生效 1. oh my zsh 插件 3. cat：不打开文件的情况下查看内容 4. imgcat：同上，不过查看的是图片 5. less [filename]：同 cat，不过如果长文件，可以用这个，仅显示一部分 6. pbcopy < [filename]：复制文件内容到 clipboard 7. touch：创建各种各样的文件，可以同时创建