Google Warned Us, But This Japanese Lab Just Proved AI Doesn’t Need Our Data Anymore
Why Sakana AI’s Core War experiment changes everything we thought about machine learning
On January 8th, a Japanese lab dropped a paper that honestly made me question everything I thought I knew about AI creativity. Sakana AI, working with MIC, just proved something wild: language models competing against each other in a game from the 1980s didn’t just match the best human players.
They surpassed them without ever being trained in human techniques or strategies.
And here’s what gets me. This announcement lands right when everyone’s obsessing over Claude Code and “vibe coding,” this new way of building apps just by describing what you want. A senior Google engineer recently admitted that a project her team spent a year on got reduced to one hour with AI. Andrej Karpathy, a machine learning legend, said he’s never felt more behind as a programmer.
But what Sakana AI discovered goes way deeper than just coding faster.
When Code Becomes Warfare
To understand why this matters, you need to know about Core War. Born in 1984, the game features “warriors” battling for control of shared virtual memory. What about the beauty? There’s no distinction between code and data. Everything in memory can be executed as an instruction or weaponized against opponents.
Before each battle, you write a small assembly language program. This program gets randomly placed in memory alongside your opponent’s. Each program tries to crash the other while defending itself. You can bomb memory zones with deadly instructions, replicate like a hydra to survive even when parts of you are destroyed, or scan the environment to locate and neutralize enemies.
For 40 years, the global community has been perfecting increasingly sophisticated strategies. There’s even a ranking called “King of the Hill” where the world’s best programs compete constantly. These decades of competition created what we call meta-strategies: dominant approaches everyone knows and must defend against.
Beginner’s guide to Redcode
Version 1.10 There aren’t too many beginners interested in the game of Core War these days. Of course, this is quite…corewars.org
The Digital Red Queen Never Stops Running
Here’s where it gets fascinating. Sakana AI’s algorithm, called the Digital Red Queen (DRQ), works on a simple but brutally effective principle. The name references the Red Queen from Alice in Wonderland, who explains to Alice that you must run as fast as possible to stay in the same place.
