AI Engineering Custom Instruction Guide | Generated by AI

Home 2026.05

Done. Here’s what I created and why:

File written: /Users/lzwjava/projects/jekyll-ai-blog/original/2026-05-19-custom-instruction-en.md

What’s in it and why:

I distilled your entire identity — 12 years of shipping, startup founding, model training, heavy LLM usage, self-taught CS — into a custom instruction that pushes AI answers toward your long-term goal of AI engineering. Here’s the structure and the thinking behind each section:

Section What it does
Who I Am Tells the AI your actual biography — startup founder, bank engineer, GPT-2 trainer, top 6% AI user. This prevents it from assuming you’re a junior or a generic user.
Long-Term Goal Makes explicit that you’re transitioning into AI engineering — training, agents, LLM internals. Every answer should accelerate this.
Technical depth (Karpathy-style) Forces first-principles explanations — show the math, walk through the forward pass, reference the code. No hand-waving.
Code-first, build-first Your vibe coding ethos: always prefer code over prose, minimal implementations over long explanations.
AI and LLM bias When there’s a choice between a traditional approach and an AI-native one, the AI picks AI. This compounds over hundreds of answers — you accumulate AI knowledge by default.
Practical > theoretical Working code > white papers. If citing a paper, show the 50-line Python implementation.
CS fundamentals Tells AI you have solid foundations — don’t dumb down, but connect new ideas to what you know.
English first Your professional language. No Chinese in answers.
Search-first Evidence-based answers with sources — you’re building real systems, not collecting plausible guesses.
Vibe coding ethos CLI-first, scripting-heavy, AI-augmented. No GUI workflows. Terminal-native.

Where to paste this:

This works as-is in:

The post is also ready to publish on your Jekyll blog once you run the pipeline — it’s in original/ following your standard naming convention.

Back Donate