Andrej Karpathy mentioned Permutation City by Greg Egan 4 times

5/5

— Andrej Karpathy

2019-12-20 on goodreads.com

Thanks for asking! I really enjoy reading sci-fi because it’s about predicting the future and a lot of success in research is critically predicated on making the right bets about the future. First I would highly recommend Arthur C. Clarke’s “Profiles of the Future” (see my review on Goodreads). It’s a wonderful study of the science and art of predicting the future. Second, on the topic of sci-fi’s I really like books written by scientists turned writers because I find the world building to be much more compelling, interesting and logically consistent. Recently I enjoyed: Stanislaw Lem’s Fiasco and His Master’s Voice Stories of Your Life and Others from Ted Chiang (especially Understand and Story of Your Life) Ready Player One, which paints a likely future where large portion of the population spends time in VR in a Second Life - like environment. This is another example of sci-fi that has repercussions for AI research. Suppose this were true, how amazing would that source of data be, of millions of people interacting in real time in virtual worlds, etc. What does it enable? What kinds of techniques would flourish? The Black Cloud from Hoyle Foundation by Asimov, especially the concepts of psychohistory and dark ages A Fire Upon the Deep (only chapter 1, describing a flowering Superintelligence, really awesome read. Later chapters went downhill) Permutation City In short, I like a lot of sci-fi that does compelling world building and I don’t care about beautiful language, story lines, or anything like that. I’ve also tried writing some sci-fi myself, one example is my recent Cognitive Discontinuity, which imagines a future where the way we make most progress is not by any new fancy algorithms but by brute-forcing our way through with supervised learning by recording huge amounts of teleoperation sessions of AMT-like workers on robots. I’m now also (slowly) writing a second short story set in a different possible future where unsupervised learning turns out to work very much as many people might hope. It’s very hard to write and amusingly, I’m finding that writing it and making it consistent dictates my beliefs over what research directions are interesting to pursue. i.e. it’s a form of research. Oh and link to my Goodreads profile if you’d like to stay on top of future good sci-fi I discover. Also always happy to hear suggestions.

— Andrej Karpathy

2016-11 on quorasessionwithandrejkarpathy.quora.com

there were parts of Permutation City that I didn't get / annoyed me. But I had audiobook & terrible narrator. Have to reread book

— Andrej Karpathy

Simulation. Artificial Life. Aliens. Highly inventive, enjoyable.

— Andrej Karpathy

karpathy.ai