The Value Function of a Life

Lately, I’ve been thinking about a question: in an age when AI has become so advanced, how can individuals stay competitive and continue to grow?

Honestly, I was deeply impressed by OpenAI’s Astra launch video. From an MIT Interdata 7/32 drawing a small yellow dot in response to a spoken command in the late 1970s, to Astra creating a complete 3D model of a rocket from scratch today, the pace of AI’s evolution has exceeded what most people imagined possible.

A concept attracting considerable attention recently is recursive self-improvement: AI improving itself to produce an even more capable next generation of AI. Following that idea, I’ve developed some reflections of my own that I’d like to share.

1. We should approach personal growth through the same framework that drives AI’s evolution.

When the concept of an agent first emerged in reinforcement learning, its foundation was a Markov decision process involving state, action, reward, and value.

Take a simple example. Sitting at my desk right now is my state. Writing these reflections is my action. The immediate, short-term feedback is the reward, while value represents my long-term development as an individual.

Perhaps this post will receive very few likes or saves, making the short-term feedback weak. But if a single comment sparks an insight and expands the boundaries of my understanding, it could contribute enormously to my long-term value.

Conversely, if I chose to pursue a national hero ranking badge in Honor of Kings, the short-term reward signal might be strong, while the contribution to my personal value would be negligible—unless my goal was to become a professional player in the first place.

This makes defining your value function especially important. You could describe it in terms of your passion, ambition, or career path. I prefer the concept of a value function because it aligns more closely with the framework of an agent’s development and lends itself more readily to measurement.

Distinguishing between immediate rewards and long-term value is another essential part of improving ourselves. If I enjoy an excellent meal at a Michelin-starred restaurant, for example, I would classify that experience as a reward rather than an addition to my long-term value.

Some might counter this with Steve Jobs’s idea of “connecting the dots”: the effects of our actions may only become apparent with time and opportunity, remaining invisible in the moment. I don’t disagree. But within my framework, a clear distinction between reward and value is necessary for more effective iteration and growth.

2. We should prioritize developing metacognition and let our understanding guide our actions.

In cognitive psychology, metacognition refers to understanding our own cognitive processes. Here, I use the term more broadly to mean understanding the underlying nature of the physical world.

If the interview preparation steps I previously shared are shu—practical techniques—then metacognition is dao: the underlying principles.

Consider system design. What is audio, fundamentally? What is an image? What is text? How are they transformed into signals that AI can understand? And why can AI understand them at all?

These fundamental questions are easily overlooked during interview preparation. Yet interviewers often ask what appear to be unusual or tangential questions. My interpretation is that our difficulty with them often reveals gaps in our metacognition.

When reading foundational work on ResNet or Layer Normalization, have you ever thought, “I could have come up with that”? That, too, can reflect a gap in metacognition.

It is like looking in the rearview mirror at the road a car has already traveled. The route seems obvious, as though it could only have unfolded that way. Yet every inch of that road represents a decision made real in the physical world.

Before those researchers made their decisions, countless paths were available to them. Their metacognition helped them choose the one that now seems self-evident.

As the saying attributed to Einstein goes: “Everything should be made as simple as possible, but not simpler.”

Naturally, the next question is: how, specifically, can we improve our metacognition?

That is a vast question. The moment I give a concrete example, I turn dao into shu—an underlying principle into a practical technique. Still, let me offer one.

Suppose you want to apply AI’s underlying principles to your work and translate them into practical methods. Start by understanding AI at a broader level. Identify 30–50 of its most influential foundational papers. Read them in their historical context: What was the central idea of each paper? How did it influence the research that followed?

If you cannot summarize a paper’s core idea in one or two sentences, you do not yet understand it deeply enough.

There are, of course, many ways to develop that understanding: videos, written explanations, and conversations. Why do I suggest starting with papers? Because papers are the medium closest to the underlying principles. Courses and books often rework the original ideas to create a smoother narrative or meet publishing requirements, which can make their essence harder to discern.

Think of the difference between eating a boiled potato and eating potato chips. The added ingredients may make the latter more appealing, yet they can work against the body’s fundamental need for nourishment and survival.

Once your understanding of dao reaches a certain level, you can translate it naturally into shu across different situations. As the Chinese saying goes, “The great image has no form.” In my view, the pre-AI model of personal growth through accumulating experience in a narrow niche will now be reexamined and reshaped.

I’ll close with a quote from Ivan Zhao:

“Every miracle material required people to stop seeing the world via the rearview mirror and start imagining the new one. Carnegie looked at steel and saw city skylines. Lancashire mill owners looked at steam engines and saw factory floors free from rivers. Now, AI has arrived, the next skyline is there, waiting for us to build it.”