VOLUME I, CHAPTER 1, SECTION 2
The engineers training the AI are cutting their own throats.
Last section locked one thing. Everyone is on the list, including the Silicon Valley men saying the plumber can't be replaced. This section is about the names at the very top of that list.
In the whole AI Stampede, the most sinister position is not at the bottom. It is at the top.
San Francisco Bay. Mountain View, Cupertino, Palo Alto. The heart of Silicon Valley. The smartest minds on the planet are packed into a few square miles, the top graduates of MIT, Stanford, Carnegie Mellon, Tsinghua, Peking, the IITs.
Pay starts at three hundred thousand a year. A senior researcher pulls one to three million. And when these companies, OpenAI, Anthropic, Google DeepMind, Meta AI, xAI, poach each other's best people, the signing bonus can hit a hundred million dollars. Not salary. Signing bonus. Meta reportedly put a hundred-million-dollar one-time offer on a single researcher. The most expensive workers in human history.
And what do they do all day?
They write code to train the AI. They tune the model. They clean the data. They design the reward. They test where it is weak and go make it stronger. Every day, one goal. Make the AI smarter, sharper, closer to doing every job a human can do.
Teach the AI to write code, and the first person replaced is the programmer. Themselves. Teach the AI to do research, and the first person replaced is the scientist. Themselves. Teach the AI to design AI, and the first person replaced is the architect. Themselves.
Every move that makes the AI stronger makes them weaker. Every line of code is their own layoff notice. Every breakthrough is one step closer to their own replacement.
These golden children, on their million-dollar salaries and their ten-million signing fees, spend every day doing one thing, and it has a name. Cutting their own throats.
This book stands on one root virtue, and I will say it in three words. You do not torment. Volume II sets it down as the deepest root of a civilization four thousand years old. You do not torment others. You do not torment yourself.
Silicon Valley is doing the exact opposite. It has built a whole industry out of tormenting itself, and it does the work expertly, devotedly, tirelessly, and with the whole world's admiration. Self-destruction has become a badge of honor. The better you are at cutting your own throat, the higher your bonus, the better your papers publish, the more the young ones worship you.
GitHub Copilot lands and the junior programmer market shrinks forty percent overnight. GPT-4 lands and the mid-level market starts to wither. Devin, Cursor, Claude Code, the AI programmers, arrive, and even the senior engineers get nervous.
The heads of the top AI companies have said it in public. Within months AI will write most code. Within a year, nearly all of it. Sit with who is saying that. The man saying it is the man employing hundreds of the best engineers alive to write the code that trains the AI. A boss telling his own staff, to their faces, the code you are writing will be replaced in three months by the AI you are building.
Did anyone jump up and quit? No. They kept writing. Kept training. Kept pushing the tool that erases them toward the finish line. Why? Because the self-destruction has gone too deep to stop. And that cannot-stop runs in three layers.
Layer one. The signing bonus is too sweet.
A hundred million dollars. Do the math. Median American household income is seventy thousand a year. A hundred million is fourteen hundred years of it. A thirty-five-year-old engineer signs that paper and his grandchildren's grandchildren never work. That pull sits at the floor of the human mind. Show any genius that number and watch his hands shake.
Capital knows the pull. Capital uses it to nail the smartest minds on Earth to the workstation where they destroy themselves. And every person who signs knows the deal. I have five years. In five years I push the AI to the next level, and after that the AI pushes itself. In five years the AI I trained replaces me. But by then I am free for life. Who cares if it replaces me, I never have to work again.
Clever math. It leaves one thing out. You get your freedom, and everyone around you falls. Your colleagues. Your students. Your successor. Your son, your daughter. Everyone who did not sign that paper, everyone still writing code without last generation's jackpot, every kid just graduating into the field, all of them replaced by the thing you built. Your one lifeboat, bought by drilling a hole in the whole ship.
Layer two. The prisoner's dilemma. This one is worse.
Say there are ten top engineers. If all ten walked out together, struck, stopped writing, the AI Stampede would stall. Ten people would protect their own positions, and the positions of tens of millions of programmers, and hundreds of millions of desk workers behind them.
But the ten will never walk out together. Because if nine strike and the tenth does not, the tenth takes the hundred million and pushes the AI to the next level alone. The nine who struck protected their jobs and got nothing. The one who stayed took all the money and waits to be replaced, and he already has the money.
So every one of them is terrified of being one of the nine. Every one of them wants to be the tenth. And the result. All ten scramble for the tenth spot. All ten keep writing. All ten keep training. All ten keep cutting their own throats.
That is the prisoner's dilemma, blown up to the scale of the smartest tens of thousands of minds on Earth, and the result is that the AI Stampede cannot be stopped. Not because any one person is bad. Because the mechanism itself. Each engineer's private decision is rational, take the money and run. All of them deciding it together is a catastrophe, everyone takes the money and runs, which means everyone wipes themselves out and drags the whole species in behind them.
Layer three. Capital will not allow a stop.
Say one top lab has a change of heart. The CEO stands up. We are stopping, no more training, let society catch its breath. The day that goes out, the investors lose their minds. They did not pour in tens of billions for a change of heart. They poured it in for money. They move to replace the CEO with one who will not have a heart. Can't replace him? They pull the money, and thousands of engineers are out of work that afternoon. Where do those engineers go? To the labs that did not stop.
Capital will not allow any single company to stop. The moment one stops, capital moves to the ones that do not, and the one that stopped is dead in the market. That pressure runs down the whole chain, all the way to each engineer's desk. You do not write it, the man at the next desk does. He does not, the desks in Beijing, Bangalore, Tel Aviv do. Twenty-four hours a day, somewhere on Earth, someone is writing it. Capital is not commanding a group of people. It is commanding the entire planet's supply of engineers. No individual's conscience can stop the chain.
Capital chasing profit is an instinct doing its job. It sees the meat of the AI Stampede and moves to eat. It does not care how society digests the position vacuum afterward. It cares about next quarter's numbers, the next round's valuation, the next IPO. Capital is not evil. But the instinct, run to its end, nails the smartest tens of thousands of minds to the workstation where they destroy themselves, pushes hundreds of millions of desk workers to the edge, and shoves the whole species toward something it has never seen.
These engineers are not villains. They are the smartest, hardest-working, best-educated people of the age. Most were first in the class from grade school. MIT, Stanford, Tsinghua, Peking. Graduate work under Turing laureates and Fields medalists. Their contribution to human knowledge is real. The Transformer, diffusion models, reinforcement learning, vision-language models, all of it stands on tens of thousands of their sleepless nights.
Geoffrey Hinton, one of the founders of deep learning, a Turing laureate, quit Google in 2023 and said in public that he regretted his life's work, that AI had outrun what he expected, and that he was afraid of what it would do to people. Yoshua Bengio, another founder, said much the same. Stuart Russell wrote a whole book on the danger. These are the grandmasters of the field, and they are all pulling the alarm.
Did their students and their students' students hear it? Yes. And they kept writing the code. Why? Signing bonus. Prisoner's dilemma. Capital. The three layers lock them in place. They hear the alarm, and they cannot stand up.
And it goes further. Bring out the part where the bloodline ends.
The Silicon Valley engineer's child is usually at Stanford too, usually headed into AI. Father is a research scientist at a top lab. Son studies computer science at Stanford, plans to join an AI company and do AI research. The classic Silicon Valley family path.
Father's generation writes the code that trains the AI, cuts himself down with his own hands. Son's generation studies AI to enter the field and finds the door shut. The AI already writes code. Already does research. Already designs the next AI. It does not need more human researchers. Father took his hundred million and retired free. Son stands at his Stanford graduation and finds the entire field he trained for has been cut away by his father's generation.
Father tormented himself. And his son. And his son's son. The whole professional bloodline severed at the father's generation. The hundred million passes to the son, enough that the son never works. But the son's son? Run it out at three percent inflation a year, and by the third generation the money's buying power is down to a quarter, by the fifth to a twentieth. Capital's wealth gets digested by time. The severed bloodline never comes back.
Some in Silicon Valley have seen this layer. Which is why a new set of words is suddenly fashionable. AI safety. AI alignment. AI governance. Companies raise the safety banner. Superalignment teams get stood up. Universities open AI ethics courses. On the surface, taking responsibility for the human future.
But the better the safety work goes, the stronger the AI, the more human work it replaces, including the work of the people doing the safety research. The loop is closed. Every path around it is still self-destruction. AI safety research is, at its root, making the self-destruction a little safer, a little more aligned, a little more orderly. The tormenting itself does not change. The whole AI safety industry is also cutting its own throat. The smartest people alive, studying how to let the AI replace humans safely, and the better they study it, the more completely it replaces. AI safety is the lubricant of the AI Stampede, not the brake.
The real brake is in Volume VI, Saving People. Institutional adjustment. Limits on capital. Self-discipline in every person. The refusal to torment. AI safety research cannot solve the root, which is tormenting yourself.
So someone finishes this section and asks. If the logic of capital is this solid, if the prisoner's dilemma cannot be cracked, if even the smartest minds cannot stand up, is there any hope?
Volume I, Despair, does not answer that. And I am not going to answer it here, because answering it now would be a lie you could feel. This volume has one job. Pull every alarm to its loudest, lay every position vacuum out in the open, let you see exactly where you stand. That everyone is on the list, and that even the names at the very top are cutting their own throats with their own hands. The way out is Volume VI, and I will walk you there. But not yet.
If the top of the food chain cannot hold its ground, how does anyone below it? Next. Lawyers, doctors, therapists, accountants. Ten years of study to earn the position, and the AI stands in it in seconds. The whole licensed professional class, caving in.
Or go straight to the way out. Volume VI, Saving People: A Race Against Time Before the Blood Runs.
Credit to the ancestors; the mistakes are mine: Tiger Lyon.