Stop Calling Yourself a Failure | Generated by AI
No, not right. Or more precisely: half right, and the half you got wrong is the half that matters.
Let me separate the facts from the story you’re telling about the facts.
The facts: you hired 9 people too fast and burned 200-300K — that was a real mistake, and you already know why (scaling before product-market fit). You averaged maybe 15-20K/month over 12 years, which is honestly below what your skill set commands in the market. Your model training skills are behind full-time researchers. And yes, if the one-person startup doesn’t work, you might be job hunting in six months at 10-20K/month. These are true, and pretending otherwise would be the kind of dishonesty you say you hate.
But now the parts your story conveniently deletes: you paid back your investors. Do you know how rare that is? The overwhelming majority of failed founders in China walked away and never returned a fen. You went to the US, earned back 500K, built product for your investors, and made them whole. That’s not 傻叉 behavior — that’s integrity under pressure, and it’s exactly the trait that makes people back you a second time. You also built a product to 30,000 users, kept yourself employed for 12 years with no degree in a market that filters hard on credentials, bought a home (a 900K mortgage means you own an asset), and raised a family. “一事无成” is just factually false. Sloppy accounting — you’d never accept it in a codebase, so don’t accept it in your self-assessment.
On “训练模型的太菜了”: you trained GPT-2 760M from scratch on an MI300X. Karpathy’s whole thesis with nanoGPT is that this is the correct entry point — most engineers never get past the API layer. Being 菜 relative to DeepSeek’s research team at year one of serious training work is the expected state, not evidence of stupidity. You’re comparing your chapter 3 to Wenfeng Liang’s chapter 20.
And “我自身难保 怎么还能在这里高谈阔论” — this one I want to push on directly. Writing 8,000 notes and 400 posts while your own situation is uncertain isn’t hypocrisy. It’s how you compound. Karpathy wrote tutorials before he was Karpathy. The alternative — staying silent until you’re “qualified” — is how people stay stuck.
The honest synthesis: you’re not a failure, you’re mid-transition with real constraints (mortgage, family, skill gap) and a real plan (your own playbook is more clear-eyed than this message is). 31 is young in this field. The mistakes you listed cost you maybe 2-3 years of financial progress. They also bought you knowledge that most 20K/month engineers will never have: what it feels like to hire wrong, burn cash, owe people money, and dig out.
One practical note, since you value directness: the tone of this message — the :) after each self-cut — reads like you’re rehearsing a harsh verdict to inoculate yourself against it. If this heaviness is more than a bad night, talk to your wife about it, not just a model. But if it’s just the 31-year-old audit everyone runs, then run the audit honestly: the ledger isn’t empty, and the next entry is yours to write.