I’m reading Greg Egan’s Permutation City right now. It was published in 1994. The internet was barely a thing. GPUs were for video games. “The cloud” meant weather.
Three things stuck with me in the first few chapters.
Buried in a scene about a researcher running bacterial simulations, Egan wrote this:
“SNV regrets to inform you that the computing power you require has been allocated to a higher-bidding user. A memory dump of your calculations has been stored in mass storage and will be available on your next access.”
In 1994, Egan describes compute as a tradeable commodity. This is like the rate limits on tokens. This is the entire cloud GPU market in 2026.
In a Lex Fridman interview, Sam Altman said:
“Compute is going to be the currency of the future. It will maybe be the most precious commodity in the world.”
Thomas insists on speaking English himself — even though his office software could translate for him using the exact same words:
“Thomas insisted, whenever possible, on producing the language that existed inside his own skull.”
The software can do it just as well. But the act of producing it yourself is the point.
Why write when the machine writes fine? Why code when coding agents write it?
Experts run a Turing test on uploaded human minds. None of them can tell who’s real and who’s a copy. The result?
“Some philosophers and psychologists argued that this proved nothing more than that consciousness could be ‘simulated’ — ergo, copies were programs that brazenly faked a nonexistent inner life.”
The test was passed. It didn’t matter. Similar to what happened when LLMs passed various benchmarks including the Turing test.
In 2025, GPT-4.5 was judged to be human 73% of the time in a three-party Turing test — more often than the actual humans. Emily Bender called LLMs “stochastic parrots” — stitching together linguistic forms without any reference to meaning.
Note: I’m reading the German translation (“Cyber-City”). The book quotes above are my translations back to English.