Co-evolution of self-replication and function in a digital primordial soup

A new research paper just dropped a delightfully weird idea into the internet’s lap: what if you throw a bunch of tiny random computer programs into a digital “soup,” don’t give them a built-in way to reproduce, and just see what happens? According to the paper, some of them figure out how to copy themselves anyway—and, with the right reward, also get better at solving math problems. In plain English: these little scraps of code weren’t handed the usual rulebook, yet some still stumbled into survival tricks and useful behavior at the same time. Naturally, the comment section immediately turned into a mix of awe, nostalgia, and “wait, people are still doing this?” energy.

The strongest vibe was fascinated nerd joy. One commenter raced in with an independent reproduction of the main result on GitHub, which is basically the online equivalent of yelling, “Pics or it didn’t happen—oh wait, here are the pics.” Another said the whole thing felt like systems that learn multiple jobs at once, arguing that doing two things together can make both better. Meanwhile, one amused reader admitted they’d forgotten this whole corner of research was even alive in the age of chatbot mania, while another went full throwback mode, name-dropping Tierra and old-school artificial life experiments like it was a family reunion for digital creatures. There wasn’t much hostile drama here—more of a cheerful culture clash between today’s AI hype and the old guard saying, “Actually, we’ve been making weird little life-forms in computers for ages.”