January 8, 2026
Code Thunderdome, fight!
Digital Red Queen: Adversarial Program Evolution in Core War with LLMs
Retro code gladiators spark AI hype vs OG skeptics
TLDR: Researchers used AI to evolve battling programs in retro game Core War, claiming smarter, adaptable strategies. The community is split between excitement and “seen it before,” with veterans demanding leaderboard results to prove it matters—turning a fun demo into a showdown over real-world relevance.
The 1984 cult game Core War just got an AI makeover and the commentariat is throwing popcorn. An author from Sakana AI/MIT parachuted in with, “We just released this paper… hooking up LLMs (large language models) to Core War,” promising evolving code gladiators and a “Red Queen” arms race where bots must adapt or die. The crowd? Split right down the middle. Old-school Core War vets rolled in with receipts, reminding everyone that evolution in Core War isn’t new, it’s just the LLM as mutation engine twist. Others demanded receipts of a different kind: rankings on the infamous “hill” leaderboards, asking, basically, scoreboard or it didn’t happen (example).
Fans cheered the demo’s chaotic self-modifying shenanigans and the claim of convergent strategies emerging from AI, calling it a safe playground for cybersecurity war games. Skeptics fired back with “cool toy, now beat the champs” energy. Jokes flew about “Skynet in a fishbowl,” and one wag dubbed it “AI Pokémon for assembly nerds.” The real drama: is this genuine progress or a shiny remix of 90s genetic algorithms? Either way, the comments turned into a ringside brawl—part nostalgia tour, part peer review, all deliciously spicy.
Key Points
- •Digital Red Queen (DRQ) uses LLMs to evolve Core War warriors in an adversarial loop against past champions.
- •The dynamic process yields increasingly general strategies and exhibits convergent evolution among code behaviors.
- •Core War’s Turing-complete, shared code-data memory leads to self-modifying and chaotic code dynamics.
- •Strategies discovered include targeted self-replication, data bombing, and multithreading, improving robustness.
- •Technical report and open-source code are provided, with interactive visualization of evolved warriors.