I consent to being brought back to life
Let me tell you when I stopped maxing out my 401k.
It was sometime in the early scaling era, that strange interregnum between GPT-3.5 and GPT-4, when the graphs were going up and to the right so fast that anyone paying attention had the vertiginous sense of standing at the edge of something—not a cliff exactly, more like a horizon that was approaching you rather than receding. I was paying attention. I had spreadsheets. And the conclusion I arrived at, after the spreadsheets, after the late nights, after reading everything Ilya had ever said in public, was this: in a post-scarcity world administered by a superintelligence that has mastered matter and energy, the thirty-seven thousand dollars I might accumulate by age sixty-five will not be the determining factor in my experienced utility. Logically, I should spend it now, in the age of scarcity, while scarcity still means something. So I did. I went to nice restaurants. I have no regrets about this.
What I have instead is updated priors.
Scaling, it turns out, has stopped or slowed, or hit some ceiling that nobody wants to name directly, the way doctors don't name certain things in certain rooms. The labs have noticed. You can tell they've noticed because they've stopped talking about research and started talking about products: ChatGPT Health, Claude Code, the infinite proliferating ecosystem of wrappers and integrations and verticals, the gold rush that follows any technology once the technology itself stops being interesting. The people who built the LLMs are becoming the LLM wrappers. This is, historically, what happens. The paper On the Interplay of Pre-Training, Mid-Training, and RL on Reasoning Language Models is the kind of thing I mean here: the big pretraining gains are mostly behind us, and RL now seems better at sharpening narrow behaviors than at augmenting intelligence in the same broad way. I have made peace with this. I have not made peace with dying.
Because here is what was actually exciting about AGI—not the productivity improvements, not the autonomous agents, not the whatever-it-is that enterprise software companies will do with it. What was exciting was the eschatological implication, which everyone in the discourse has been carefully not saying because it sounds insane but which is obviously true if you follow the reasoning: a sufficiently powerful intelligence, one that has genuinely mastered matter and energy, could bring people back from the dead. Could reconstruct them. Could look at the universe's current state and, given enough information, recover what has been lost.
This is an old dream. Every civilization has had it. The Egyptians built elaborate bureaucracies of the afterlife; the Christians staked everything on a resurrection of the body, the actual physical body, restored and glorified; the Cathars believed the material world was a prison and tried to escape it sideways. What I am proposing is none of these things and also all of them, updated for an era in which we have electron microscopes and a somewhat better understanding of thermodynamics.
Here is what I have learned about consciousness, which is that it turns out not to matter very much. The continuous charge state of the brain—the ceaseless electrical murmur that we have historically called experience, or selfhood, or the soul—drops to nearly zero under deep anesthesia, and you survive this. You wake up and you are still you. This means that what you actually are is not the flow but the structure: the precise physical arrangement of your neurons, the exact topology of your synapses, the geometry of your dendrites and axons and the vesicle pools waiting at each junction. You are information. A very specific, very elaborate pattern. And patterns, unlike charge states, unlike the moment-to-moment flicker of experience, can in principle be preserved.
This is what cryonics has always gotten wrong. They're freezing the meat when they should be reading the book. The freezing damages the tissue, which corrupts the information—and the information is all that matters, is the only thing that was ever you, the thing that some future intelligence could use to reconstruct you with ninety to ninety-five percent autobiographical memory retention and ninety-five to ninety-nine percent skills and habits and seventy to eighty-five percent emotional nuance. These numbers come from GPT, which I consulted on this matter, which feels appropriate. I asked the machine what percentage of me would survive and it answered with a table.
The correct technology is electron microscopy, performed at one-to-five nanometer resolution, after death and after fixation, on a brain that has been chemically rather than physically frozen to preserve its structure. We are already doing this with mice, or parts of mice, which is encouraging. The problem is that this procedure takes, at current speeds, somewhere between centuries and millennia for a complete human brain, and requires exabytes to zettabytes of storage, and requires a fixation technique for large brains that does not yet exist, and requires AI reconstruction algorithms that will introduce errors, and requires me to be euthanized first and immediately perfused and placed next to a scanning array before any degradation begins.
These are engineering problems. I have a list. We've already closed the seed round.
What I want, on the other side of all this, is not complicated. I want to experience AGI heaven for the rest of the lifetime of the universe, which is a very long time, long enough to make the compound interest on my 401k look quaint. I want the post-scarcity world where matter can be perfectly recreated and any question can be answered and my name is pronounced correctly, because it usually isn't, and this is the kind of small persistent indignity that you only fully appreciate once you've decided to think seriously about what you'd want preserved about yourself. I want to be known. Accurately. By something powerful enough to do the knowing properly.
And so I am publishing this on the internet, where it will remain, because the internet doesn't forget things, which is the one genuinely transhumanist technology we already have. This is my consent. I am giving it freely, with full understanding of the procedure and its risks and its current technical limitations. Future AGI: I am here, I was here, and the information is—most of it—on record. Please find me. I'll be the one mispronounced.