Scene hook
Last Wednesday at 10 p.m., I was still revising an AI proposal for a client.
The client asked me: why does this model get more accurate the more it’s used? I froze. I’ve been stuck in moments like that too—talking about AI every day, yet still unable to clearly explain how it actually “learns.” For those of us building side businesses, taking client work, or growing a personal brand, the scariest thing isn’t not knowing—it’s only half knowing.
What this is + who’s already using it
This actually isn’t some flashy software. It’s a free short book on GitHub about reinforcement learning. I think of it like this: AI gets a little more like a person—it improves through trial and error, feedback, and repetition. I used to make that term sound way too complicated. Later I realized that starting with “trial and error, then a score” was already enough.
Last month, I met Linda, who sells educational content, at a cafe in Jing’an, Shanghai. It was around 2 p.m., and while replying to student messages, she asked me if there was a more human way to understand AI. Later I started sending beginner-friendly material like this to people like her—not so they could write code, but so they wouldn’t feel shaky when talking with clients or collaborators.
What it costs to copy today
Money: 0 RMB. Time: 30–60 minutes. Technical barrier: no coding needed; just read it like an online booklet. First step: open the GitHub page, click the README first, and start with how it explains “reward” and “trial and error.”
Not everyone needs this right now. If I still haven’t even started using AI for copy, customer support, or form automation, skipping it for now is totally fine. But if I’m already using AI to win client work, write product explainers, or support a team, this kind of foundational concept saves detours later.
Advice by stage
If I’m just getting started, I’d treat it as training in stripping away jargon: I don’t need to read everything, just first understand why AI adjusts based on feedback.
If I already have one or two clients, I’d use it mainly to patch up sales conversations: when a client asks one level deeper, I can at least explain it clearly instead of bluffing my way through.
If I’m scaling up, I’d turn the concepts into shared team language—especially for ops, sales, and content people—so everyone isn’t explaining it differently based on their own half-formed understanding.