Permitted by Physics | Christian Engel<br>Post<br>Cancel<br>Sometime in the last two years the price of intelligence fell by a factor of two hundred and eighty, and almost nobody noticed.<br>In November 2022 it cost about twenty dollars to get a million tokens of GPT-3.5-level reasoning out of a machine. By October 2024 the same quality of thinking cost seven cents. A 280-fold collapse in the price of the thing our entire civilization is supposedly built on, in twenty-three months.1 If the price of oil had done that, it would have been the only story on Earth.
Nov 2022 (GPT-3.5)<br>$ 20.00 / Mtok
Oct 2024 (Flash 8B)<br>$ 0.07 / Mtok
2025 bonus: GPT-4 class<br>$ 0.18 / Mtok
Blended input+output cost (3:1 weighting) per million tokens for the cheapest model at GPT-3.5 quality. Endpoint Oct 2024 / Gemini 1.5 Flash 8B is the AI Index's verified GPT-3.5-equivalent. The price floor for that quality has held since. What changed in 2025 is the quality at the same price tier: GPT-4.1 nano (Apr 2025) and Gemini 2.5 Flash-Lite (Jul 2025) are GPT-4 class and priced at $0.10 input / $0.40 output (~$0.18 blended).
It wasn’t. It was barely a story at all. And the reason it wasn’t, I have come to think, is the same reason I have come to believe the future is going to be bright.<br>I have never understood cyberpunk. Not the aesthetic. The aesthetic is fine. I like rain and neon as much as anyone. The premise. The premise is always the same: breathtaking technology, flying cars, brain implants, holograms in the street, and underneath it a ruined planet, a starving underclass, and three megacorporations running everything. Half the future is turned up to eleven and the other half is left to rot.<br>It is not a prediction. It is a contradiction wearing a prediction’s clothes.<br>The argument I want to make about that comes from David Deutsch. Problems are inevitable. Problems are soluble. The human ability to transform nature is limited only by the laws of physics. Anything not forbidden by physics is achievable, given the right knowledge. The only thing standing between us and any particular transformation is knowing how.<br>Apply that to the flying car. To build one that ordinary people use, you need four things. Energy so cheap that lifting a ton of metal into the air all day is economically trivial. Materials light and strong enough that the thing flies at all. Artificial intelligence good enough to manage millions of autonomous vehicles in shared airspace without killing anyone. And an economy rich enough to pay for the other three.<br>But look at what each of those four things also does, the moment you have it. Cheap energy is the entire problem of carbon removal. Pulling CO₂ out of the atmosphere is not forbidden by physics. It is expensive, and expensive is a synonym for needs cheap energy. A civilization with flying-car energy prices removes its carbon backlog the way we pave roads. Boring, scheduled, unremarkable.<br>Materials science good enough for flying cars is materials science good enough for cheap desalination, engineered crops, and ecological restoration at scale. The AI that safely flies a million cars is the AI that runs a personal doctor and a personal tutor for every human being. And the economic surplus that pays for all of it is, by definition, the opposite of the starving underclass the story insists on.<br>So the cyberpunk future asks you to believe something genuinely strange. That a civilization solved the hard problem, limitless clean energy, advanced AI, abundant materials, and then sat in the rubble for a century declining to solve the easy ones that the hard solutions hand it for free. That isn’t dark. It is incoherent. It is a society that built the ladder and then refused, on principle, to climb the last rung.<br>The honest extrapolation runs the other way. If you genuinely believe in the flying cars, you are already committed to the regreened Earth. They come from the same knowledge. The solarpunk future, clean and abundant and alive, is not a softer alternative to the cyberpunk one. It is the cyberpunk one with the arithmetic finished.<br>The doom is the part that doesn’t follow. When I say the optimistic case is the logical one, people assume I’m doing vibes. So here is the opposite of vibes. Three things that happened in 2025, and a note on the address.<br>The first AI-designed drug reached patients. A treatment for idiopathic pulmonary fibrosis, a disease that for as long as we have had a name for it has killed people on a steady schedule, where the machine did both halves of the work: picking the target and designing the molecule. The lab is in Hong Kong. The Phase IIa results were published in New York.<br>A second cure moved from clinic to clinic. By late 2025 nearly three hundred sickle-cell patients had been referred for the first gene-editing medicine ever approved. Sickle cell, a thing that for the entire history of medicine you simply had, is now, for some people, a thing you had and no longer have. The science came out of...