The AI Assistant Is Way Smarter Than Me

Tags:
Categories:

For a middle-aged man who has been coding for ten years, once went abroad for a gilded stint, and still cares about saving face, admitting that AI is better than me is downright embarrassing.

The total monthly cost of all the AI tools I use doesn’t exceed 200 RMB, yet my boss pays me far more than that.

I fully expect earfuls of ridicule:

“That’s only you.”

“That’s what junior devs say.”

“It only handles the easy stuff.”

“It can’t do real engineering.”

“Its hallucinations are severe.”

“It’s not fit for production.”

My experience with AI tools has been robust enough to let me shrug off such mockery. This article won’t promote any specific tool; its main purpose is to create an echo in your thoughts, since I learn so much from each comment thread.

I was among the first users of GitHub Copilot; I started in the beta and, once it ended, renewed for a year without hesitation—and I’m still using it today. I no longer get excited when I solve a thorny problem on my own, nor do I take pride in “elegant code.” The only thing that thrills me now is when the AI accurately understands what I am trying to express, when my AI assistant fulfills my request—and exceeds my expectations.

Of the experience I accumulated over the past decade, what turns out to be most useful with AI tools is:

  • Logic
  • Design patterns
  • Regular expressions
  • Markdown
  • Mermaid
  • Code style
  • Data structures and algorithms

More specifically:

  • A major premise, a minor premise, and a suitable relation between them.
  • Create dependencies cautiously and strictly prevent circular ones.
  • Do not add relations unless necessary; do not broaden the scope of relations unless necessary.
  • Strictly control the size of logic blocks.
  • Use regex-based searches and, following naming conventions, generate code that lends itself to such searches.
  • Generate Mermaid diagrams, review and fine-tune them, then use Mermaid diagrams to guide code generation.
  • Use the names of data structures and algorithms to steer code generation.

I have spent a lot of time contributing to various open-source projects—some in familiar domains, others not. It’s experience that lets me ramp up quickly. You’ll notice that great projects all look alike, while lousy projects are lousy in their own distinct ways.

If my memory gradually deteriorates and I start forgetting all the knowledge I once accumulated, yet still have to keep programming to put food on the table, and if I could write only the briefest reminder to myself on a sticky note, I would jot: Google "How-To-Ask-Questions"

Are humans smarter than AI? Or are only some humans smarter than some AI?

I have to be honest: puffing up my own ego brings no real benefit. As the title says, this article tears off the façade and exposes what I truly feel inside—that AI is better than me, far better. Whenever doubts about AI creep in, I’ll remind myself:

Is AI dumber than humans? Or are only some humans dumber than some AI? Maybe I need to ask the question differently?