Five Lessons from Three Years of College: From a Wrong Reward Function to a Private Evaluation System

Three years of college have passed without my noticing. Looking back, I made quite a few mistakes—but to my surprise, those mistakes turned into some of my most valuable assets.

1. The Wrong Reward Function

Overall, I think the environment today’s middle school and college students face is still full of competition. From childhood onward, we live in a competitive environment where we try to push up our gaokao scores, GPAs, comprehensive evaluation scores, and graduate entrance exam scores. These metrics all share one trait: they can be improved through individual effort alone, with no need to develop teamwork skills—you just need to sit and study quietly by yourself.

Let’s reflect: are these metrics really the ultimate goal? Do they really reflect a person’s overall ability? My answer to both is no. They are not the ultimate goal—the ultimate goal is to get things done and done well, and the metrics themselves produce no social value. Nor do they reflect overall ability—they are just approximate measurements of one or two dimensions.

In the language of reinforcement learning: this environment’s reward is misdesigned. Over-optimizing on a wrong reward can never produce an optimal policy. Worse, the policy it produces treats everyone as an opponent and competitor, and treats these metrics as the sole standard for judging whether you are good. And a policy trained this way immediately fails the moment it enters real scenarios like teamwork or business competition.

So what is the right reward? If the social norm is “metrics are the reward,” we can’t change the environment, but we can change how we read the signal: every outside opinion is just a judgment of one particular ability of yours, and does not represent your true, comprehensive reward. In other words, build your own private evaluation system, and pull the reward back to the real objective. Under this evaluation system—as objective as I can make it—the optimal policy I have learned might be:

treat others as api instead of judger or competitor, while play a role of api for others too.

Treating others as APIs means understanding their interfaces, the boundaries of their abilities, and the exchange of value—rather than treating them as judges or rivals. At the same time, it means packaging yourself as a stable, reliable, callable interface.

2. Fulfillment and Happiness

Fulfillment and happiness mean doing what you love, and then providing value to others and to society—that is, realizing your own value in life.

Recently I was chatting with a friend. His passive income is already quite good, yet he still feels bored and empty. We analyzed it together, and the reason turned out to be simple: people adapt very quickly to “getting,” so no matter how much money you have, a few months of playing around and you’re tired of it—but people adapt very slowly to “creating” and “being needed.” So to stay fulfilled over the long run, you must do what you love, realize your own value, and earn money along the way. Note the order: money is the result, not the goal—the moment you make it the goal, you fall back into that wrong reward above.

3. The First Principle of Getting Things Done: Practice, Reflection, Iteration

The first principle of getting things done is practice, reflection, and iteration. Learn by doing, not do after learning.

All three are indispensable: practice without reflection is low-level repetition; reflection without practice is a castle in the air. And the quality of iteration depends on whether each loop attributes the mistake clearly—iteration with “fixing” but no “why” spins in place; once you understand the “why,” mistakes truly turn into assets.

4. Black Vitality

Black vitality matters—it is the kind of resilience that grows out of adversity.

The strength grown in good times is white: bright, sunny, but brittle. Black vitality grows in valleys, setbacks, and failures; it is not a pretty color, but it is extremely tough.

When setbacks hit, do you dare to say out loud: let me do it!

5. Aesthetics

Cultivate your aesthetics.

Aesthetics sits upstream of judgment: only by knowing what is good and what is bad can you choose the right direction at countless forks in the road. As generation becomes cheaper and cheaper, judgment becomes more and more expensive—and this may be the single ability most worth investing in during the AI era. There is no shortcut to cultivating aesthetics: massive high-quality input (papers, source code, outstanding products and works), plus constant nitpicking of your own output.