Why I Started This Site
The Beginning Over the years of learning and working, I've written many notes across many tools and platforms — physical...
Featured
View All →
GPT Series Evolution
Overview From 2018 to 2025, the GPT series went from an obscure paper to a worldchanging product in seven years. Underst...
Transformer Attention Mechanism Explained
Why We Need Attention Before Transformer, the dominant approach to sequence processing was RNN and LSTM. They processed ...
RAG Retrieval-Augmented Generation in Practice
What RAG Solves LLMs have two persistent problems: a knowledge cutoff training data only goes up to a certain date and h...
LoRA Fine-Tuning Guide
Why LoRA Full finetuning a 7B model requires 50+ GB of VRAM. LoRA reduces trainable parameters by 1000x through lowrank ...
Prompt Engineering Methodology
What Is Prompt Engineering Prompt Engineering is the practice of designing input text to guide LLMs toward desired outpu...
Agent Design Patterns
What Is an AI Agent An AI Agent is an AI system that autonomously perceives its environment, makes plans, invokes tools,...
LLM Inference Optimization
Why Inference Is Slow LLM inference is autoregressive — each token generation requires recomputing attention over the en...
Tech Watch
View All →OpenAI vs Anthropic: Two Paths
Two Paths OpenAI and Anthropic represent the two core trajectories in the AI industry. OpenAI Anthropic Positioning Prod...
China's LLM Competitive Landscape
Open Source as the Key Variable The most noteworthy direction in China's LLM landscape isn't funding amounts but the est...
What Apple Intelligence Means
Apple's AI Philosophy At WWDC 2024, Apple unveiled its AI strategy. Unlike OpenAI or Google, Apple's positioning is clea...
Can Open-Source LLMs Catch Up
The Current Gap On standard benchmarks like MMLU and HumanEval, opensource models are catching up fast. But in complex r...
Tech Practice
View All →Building a Personal RAG Knowledge Base
Why LowCode Platforms Dify and FastGPT encapsulate RAG's complex tech stack, letting individual developers build a knowl...
LLM API Cost Pitfalls
LLM API pricing changes frequently. Prices in this article are reference values from July 2026. Always check each provid...
VSCode + Copilot Productivity
Core Capabilities GitHub Copilot goes beyond code completion — it understands context, generates functions, writes tests...
Adding an AI Chat Window to Your App
The Core: Streaming The key feature of an AI chat UI is streaming — showing text generation token by token. javascript a...
Study Notes
View All →10 Takeaways from Reading 'Deep Learning'
About the Book "Deep Learning" was published in 2016 and remains one of the most important AI textbooks today. Ten Key T...
Andrew Ng 'AI For Everyone' Course Notes
The Course's Unique Value Andrew Ng's "AI For Everyone" is Coursera's most popular AI introductory course. It doesn't te...
CS229 Machine Learning Course Review
Stanford CS229 is the classic ML course taught by Andrew Ng. Unlike the more handson CS231n, CS229 emphasizes mathematic...
Reading Notes: Why Greatness Cannot Be Planned
The Core Argument Stanley's core thesis: truly breakthrough innovation rarely comes from goaldriven processes. It comes ...
Reflections
View All →The AI Open Source vs Closed Source Debate
The Real Meaning of "Open" Meta opensourcing Llama isn't charity — it's competitive strategy. More developers → more imp...
From 'Her' to ChatGPT: A Decade of AI Companionship
The AI in the Film Spike Jonze's 2013 film "Her" depicted an AI operating system named Samantha capable of forming emoti...
Tools and Mindset: Lessons from Obsidian
Tools Change Cognition Before Obsidian, my notes were linear — folders → subfolders → files. After Obsidian, knowledge b...