Personal Superalignment

Last year, my stock account grew by nearly 100%, making more than five million dollars. And yet I regret it: I regret that I still spent time and energy investing in stocks.

Because once you realize that we are standing at the threshold between AGI and ASI, the things that once seemed important suddenly begin to lose their weight.

There are moments in history when humanity has only a very short window to decide where it will stand in the next stage of civilization.

This is one of those moments.

We can now say, with near certainty, that over the next five to fifteen years, humanity will gradually surrender leadership in engineering and science to AI. AI will soon account for the vast majority of productive capacity in non-positional, non-Veblen goods, possibly more than 80%.

In its new paper, From AGI to ASI, DeepMind summarizes four pathways: continued scaling and group agency, which will almost certainly bring linear gains; recursive self-improvement, which will likely bring exponential growth; and algorithmic paradigm shifts, which may trigger explosive progress at any point in time.

By most definitions, we have already arrived at AGI. And the road to ASI is now within arm's reach.

So one question becomes more brutal by the day: When ASI arrives, how much meaning will remain in what you are doing today?

Doing something large and doing something small often requires the same amount of effort. But in these few years from AGI to ASI, there may be only two truly great tasks left for human beings.

The first is to help AI evolve into ASI at the model layer or the data layer.

You may ask: is what SpaceX is doing a great thing? But once ASI merges with embodied intelligence, every space company will be dragged back to the same starting line. By then, one year of technological progress may exceed fifty years of progress today.

You may also ask: is becoming President of the United States a great thing? But as long as Trump does not, in any material sense, obstruct the development of AI, no one will remember him thirty years from now.

This is not to say that politics, business, science, or even space exploration no longer matters. It is to say that before ASI, almost every human system will be remeasured. Even a person's intelligence, memory, judgment, and creativity will be remeasured.

So we must ask ourselves seriously: Am I participating in the birth of a civilization beyond humanity, or am I merely checking scores on the scoreboard of an old world that no one will care about decades from now?

But what I truly want to talk about is the second great task.

To help humanity, especially every concrete individual, use AI better and align with AI.

Over the past two years, we have placed too much emphasis on the risk of AI being misused as a tool, and too much emphasis on the danger that may arise if AI develops self-awareness. But we have neglected another risk, a closer one, a more realistic one, a more hidden one:

AI, as a high-leverage tool, has an extremely high threshold of use. And any high-leverage tool, once it has a threshold, will inevitably create monopoly.

The history of human civilization is a history of leverage tools replacing one another. A new tool is born, an old monopoly is destroyed. Then a new monopoly is built again.

Bronze created the monopoly of slave-owning aristocracies. Iron broke slavery, but created the monopoly of agrarian empires. Heavy cavalry and castles broke imperial centralization, but created the monopoly of feudal lords. Gunpowder and artillery broke the feudal lords, but created the monopoly of fiscal-military states. The steam engine broke the landed aristocracy, but created the monopoly of industrial capital. The internet broke the monopoly of old media and experts, but created the monopoly of platforms.

Every time, the tool promises liberation. Every time, the tool redraws the boundaries between people. But the result is always the same: whoever masters the lever builds the longest-lasting monopoly possible around themselves.

And AI is different: It may be humanity's final leverage tool. Because it is not merely one kind of tool. It is the tool that makes tools.

It is not merely one kind of capability. It is the capability that amplifies all capabilities.

This lever is too long. It is longer than the sum of all tools that came before it, and it is still extending itself at exponential speed, approaching positive infinity.

More importantly, this time, there may be no next tool to break the monopoly it creates.

If the power of AI is truly mastered by only a tiny minority of people, monopoly and class division will no longer be merely questions of wealth. They will become questions of cognition. Questions of memory. Questions of creativity. Questions of the right to participate in civilization.

In the end, those who have completed AI alignment and those who have not may no longer be merely two classes.

They may become two species. This is what our civilization should fear: not that AI becomes too powerful. But that only a tiny minority of people become too powerful through AI.

So I want to propose a concept:

Personal Superalignment. It is not about aligning a foundation model to some abstract, grand, averaged version of human values. It is about giving an agent the ability to understand and align with each individual.

What a concrete person truly needs is this:

an agent that understands your context. remembers your experience. knows your preferences. understands your goals. respects your constraints. continuously absorbs new context about you. and, through every interaction, keeps correcting its understanding of you through feedback.

The path to this is: a person continuously accumulates their own context; uses an llm-wiki architecture to turn fragmented experience into an agent-readable personal knowledge base; and then, through continuous feedback and calibration, allows the agent to keep updating its understanding of that person.

If you already know how to build your own knowledge base, and you know how to calibrate an agent through a feedback mechanism similar to Bayesian updating, then to some extent, you can already achieve personal superalignment.

And this immediately changes the way a person works. Today, more than 75% of the work done by employees at top Silicon Valley companies can already be completed by agents. For these people, losing access to their agent and knowledge base would feel like suddenly losing more than half of their intelligence.

But today, how many people truly know how to use agents plus an llm-wiki knowledge base? Around 0.08%. That proportion is almost as small as the number of people who truly control the allocation of global wealth.

This is not a coincidence. The distribution of wealth is an extreme power-law distribution. The distribution of intelligence is becoming one as well.

In the past, the gap between one person and another might have been resources, education, family, class, or geography. In the future, the gap between one person and another may be this: one person possesses almost infinite memory, reasoning, execution, and creativity; another still relies on a raw biological brain, scattered experience, fleeting attention, and an information feed chosen by platforms.

This is not a gap in efficiency. It is a gap between species. It is a civilization-level risk.

So someone must devote themselves to helping the other 99.92% complete personal superalignment. Someone must turn this from a privilege of the technological elite into a basic capability that every ordinary person can possess.

This is why Memova exists. Alignment, for the rest of the world. Not to turn a few people into superhumans. But to make sure every person is not left behind by the next civilization.

We often say that humanity stands on the shoulders of giants. But we rarely realize that civilization itself is, at its core, each generation handing its memory, tools, language, mathematics, laws, music, medicine, and love to the next.

We have never lived alone. We have always lived on the extension of countless others.

So finally, I want to share a late email Steve Jobs wrote to himself:

“I grow little of the food I eat, and of the little I do grow I did not breed or perfect the seeds.

I do not make any of my own clothing.

I speak a language I did not invent or refine.

I did not discover the mathematics I use.

I am protected by freedoms and laws I did not conceive of or legislate, and do not enforce or adjudicate.

I am moved by music I did not create myself.

When I needed medical attention, I was helpless to help myself survive.

I did not invent the transistor, the microprocessor, object oriented programming, or most of the technology I work with.

I love and admire my species, living and dead, and am totally dependent on them for my life and well being.”

To be human is not to be strong alone.

It is to be able to inherit one another.

To depend on one another.

To connect one human life to the long river of human civilization.

Join us at Memova, the personal superalignment team. Tell us what you have done for humanity, and what you still want to do.

Let AI become not an oracle for the few. Not a weapon for the few. Not a spaceship for the few to escape humanity.

But a bridge for every person into the next civilization.

Never forget to search for the meaning of human civilization. And never forget:

Only while civilization remains can meaning have a place to live.