Author: Systematic Long Short Source: tokendispatch Translation: Shan Oppa, Golden Finance

I find that many current discussions about artificial intelligence are still trapped in two extreme positions. One group believes that AI will take away all jobs overnight; the other group dismisses this wave altogether, claiming that these tools are still unstable and require human supervision. The problem is that neither attitude makes much sense. New technology doesn’t have to be perfect to spark disruptive change; it just needs to be good enough that the risks of ignoring it outweigh the risks of embracing it.

This is when direction matters more than speed. A car moving in the right direction at 1 mile an hour will also go further than a car standing still. The same goes for starting as early as possible: it gives you room to trial and error, learn, and adjust direction, and in the end you can still stay ahead of the hindsighters in the diffusion of innovation theory.

This does not mean that every time a new technology emerges, humans are destined to be eliminated. What innovation really does is raise the threshold of “substitutability”. This discussion comes at a perfect time - we are seeing AI agents starting to build their own economies: paying, contracting, and even hiring humans to help complete tasks.

So what do you do in a world where predictions about the future become obsolete before they can be fully debated?

In today's special column, SysLS will tell you: How to judge the future and make decisions without waiting for "absolute certainty". This article is about: acting before the window closes, identifying which advantages are only temporary, and why first movers often have the right to openly trial and error and adjust direction.

Introduction

The first time I realized we were at an inflection point was in my previous job. Even though everyone around me pretended that nothing would change, I clearly felt that the pace of the industry was slowing down.

At the time, I was managing a team of nearly 20 people at a hedge fund, doing what I had been familiar with for many years. Logically, I could have accomplished more there. But I still gave up the coveted position and started a business from scratch with a minimalist team - a decision that few people understood at the time and was regarded as crazy. And with recent news of large-scale layoffs, people voluntarily quit their jobs to start their own businesses, or quietly using their spare time to start their own businesses, my actions no longer seem so unreasonable.

Many people have asked me where all this will ultimately lead. This article is my answer. Frankly, I’m not sure how big this revolution will be, but quantitative finance has taught me one thing: the right direction is often enough.

The trend has arrived

What really woke me up was ChatGPT o1. Before this, I only called them "large language models" rather than true "artificial intelligence". I still don't believe that they can produce even a hint of real intelligence.

But for the first time, o1 enables large language models to reliably generate usable code based on clearly structured cues. It's still rough, with occasional hallucinations and logical confusion, but the point is: it produces useful code.

My judgment is simple: once AI can write usable code, it will recursively optimize its own logic and accelerate iteration at a speed we can hardly imagine. Whenever I share this point of view, someone always refutes that the code generated by AI still has loopholes and cannot be "online for commercial use." But what they ignore is that code written by humans also has bugs.

We don’t need perfect code to stop writing code. The moment we discover that AI writes fewer bugs and far faster than humans, that’s the moment we give up coding. The threshold for handing over coding work completely to an agent is actually very low. When I saw o1’s capabilities with my own eyes, I knew: the future will be completely changed.

Quantitative Finance and Knowledge Moat

I initially thought that AI would eventually swallow up most of the quantitative finance field. But I thought it would take time - after all, there is very little institutional-level code publicly available for large model training. I once compared software engineering to a pyramid: at the bottom are basic "code porters", at the top are senior developers with architectural thinking, and at the top are specialized talents such as data scientists and quantitative development experts. I think the higher the professional threshold, the safer the career.

I have predicted that within two years, low-level code work will be completely replaced, followed by senior developers; progressively, professional knowledge will also be absorbed by large models, and eventually even high-end positions will not escape.

But I soon discovered that head model manufacturers will eventually hire professionals to inject industry knowledge into the models. Expertise may seem like a moat for years to come, but it can also be slowly eroded.

The remaining moat

In the next five years, there are still several types of businesses that will not be easily disrupted:

1. Private data: Businesses that generate vast amounts of unique data are difficult to disrupt. For example, large institutions such as Millennium Management can collect analyst opinions, in-depth research reports, trading recommendations and real price fluctuations, and use these data to fine-tune head models to create advantages that are difficult to replicate. Any business that generates private data that is not easily accessible to head models will survive longer.

2. Industries with regulatory barriers and artificial approval bottlenecks are difficult to disrupt. If you want to trade in the traditional financial market, you need to open a brokerage account, obtain a license, and sign a contract around the world. Cryptocurrency trading is easy, but it is much more difficult for a non-Chinese company to trade iron ore in China. If any progress requires manual approval, the speed of the industry will always be limited by the cost and efficiency of approval.

3. It is not difficult for authoritative endorsement services to allow AI to draft legal opinions based on comprehensive research and relevant laws, but we are still willing to spend tens of thousands of dollars on lawyers to draft them - because AI's legal opinions are worthless at this stage. The same goes for smart contract auditing: AI may have reached or even surpassed the top level of humans, but most people are still willing to pay for authoritative certification from brand organizations. What you buy is not the opinion itself, but the authority behind the opinion.

4. Physical intelligence lags behind hardware iteration much slower than software, and hardware damage is more difficult to repair. Physical businesses that interact with the real world have a low probability of being disrupted in the short term. But once the hardware catches up, the pyramid logic will also be followed: bottom-level jobs will disappear first, and then professional positions will take their turn.

These moats are real, but none are permanent. What they buy is time, not absolute safety.

Judging the future of chaos

When the future is noisy and the pace of change is so fast that most analogies fail, people often fall into two misunderstandings: either waiting for absolute certainty before acting, or copying past experience (such as "This is like the Internet bubble") and using wrong models to guide action. Both are wrong.

When information is incomplete, thinking from first principles is more valuable. You don’t have to accurately predict the outcome, you just need to be in the right direction and make a good layout - so that "shooting early but being wrong" can be tolerated, and "shooting early and being right" can obtain excess returns.

When the future is uncertain, asymmetric returns are everything.

The implementation method is very simple: first think about "what conditions must be met for a certain result to occur", and then look at "whether these conditions are clearly visible." The inflection point we are experiencing was not unforeseen: the signals were already there — code that can be written, models that can be recursively optimized, industry knowledge that can be purchased rather than just accumulated. As long as you are willing to face these signals, you can roughly judge the direction even if you cannot see the specific path.

You can continue to deduce recursively: What the world will look like when intelligent bodies can train themselves, replicate themselves, and are truly autonomous, we can’t even imagine now. An agent that can improve its intelligence by 0.1% through a series of actions may seem insignificant, but as long as the increment is not zero, the next improvement may be even greater, and the cycle repeats. There is a huge power law effect here, and it is worth imagining the future based on it.

By the time the signal becomes visible to everyone, opportunities will already be crowded. In the market, you trade uncertainty for early entry; in careers and entrepreneurship, the same rule applies.

So the real question is not "what will happen" but: what do I already know? Which direction does it point? What is the cost of acting now rather than waiting?

One thing people often overlook is that the action itself creates the message. Action doesn't happen in a vacuum. You take action on the world, and the world responds with information; information drives iteration, and iteration leads to smarter actions. This is the nature of progress.

Standing still without complete information is going backwards. To move toward action is to explore.

Thinking about the next step

I know that if I just want to settle for the status quo, I can still be stable for a few more years. But deep down, I know: if you want to do something, you must start as early as possible. I’ve always wanted to build a career that was truly my own, and that window was closing fast.

To be clear, nothing will happen to the world's top hedge funds. They hold private data and are difficult to replace. Traditional financial markets are also subject to human signatures – both at the regulatory and transaction level. But I believe that these giants will eventually use AI to replace most of their employees, even final positions such as fund managers. Not immediately, but sooner or later it will happen.

I judge: I only have 4-5 years left. Once the leading model manufacturers hire enough professionals, there will be little room for emerging trading companies to survive. In markets such as the U.S. stock market, this pattern has already taken shape. In a few years, the efficiency gap will be unimaginably large.

Soon, "second-rate players" will have no place at all. I could continue to work for the “best,” but it’s more in line with my goals to strike out now and join a market where I have real advantages and hard-to-duplicate knowledge. So I resigned decisively, devoted myself fully, and finally founded @openforage.

The turning point has been reached

Today, the window of opportunity is visibly closing. The pace of change is no longer slow, and most in the industry are starting to realize that optimizations that once took months now take just weeks.

I don’t think jobs will disappear completely in the next few years. Human beings will always have a need to exist - we are social creatures, and as long as humans are in power, we need the presence of our own kind; humans still do not trust AI, so authoritative certification still needs to come from humans. I can imagine a future where there is an AI CEO, but most likely it will still need to be “approved” and certified by a human CEO. This human certification mechanism goes down the pyramid: human managers manage and certify their agents.

But the employment logic will completely change. If it’s easier for a CEO to give instructions to an agent than to you, there’s no need for him to hire you. In the future, shallow code transfer jobs will become increasingly difficult to find.

To become irreplaceable, you must transcend the limitations of the current agent on a time scale: receiving instructions, managing the agent, and collaborating with it over the long term (weeks, months, years). Long-term strategic thinking and policy planning will be the strongest professional moat in the foreseeable future. You also need to go beyond agents in terms of breadth of vision: they are contextually limited, seemingly omniscient, but cannot see how component A affects B, how B affects C, and ultimately the ripple effect on D—they lack global vision.

If you can see farther and broader, absorb information quickly, make long-term decisions, and be popular, you will be able to keep your job, at least for the foreseeable future.

If you plan to remain an employee, you might as well take stock of your job content: Some tasks are unique to humans and irreplaceable; others will be replaced at low cost in the next few years. Do more of the former and less of the latter.

In a truly high-quality company, occupying a truly moatable and irreplaceable position can give you a career buffer period when other positions are swallowed up by big models. You can still use your spare time to give it a try and try to create a meaningful career.

But if you are eager to leave unique value to the world, please think carefully about the track direction you choose. If the window to build a moat is closing, you must act before the market fully digests the coming competition.

Conclusion

As long as you are willing to look at the signals that create turning points, you can see them clearly in advance. Most people don't want to look, or don't act after looking at it, or wait until the signal is obvious, and the opportunity is long gone.

Don’t ignore the changing landscape. Don’t stay in a declining place and tell yourself “wait for better times before you jump.” There is no better opportunity. Opportunity never announces itself. By the time everyone can see clearly, the window has usually closed.

I saw it, I made the bet, and now I'm living the consequences of that bet - good or bad.