AI Framework GenOS Evolves Algorithms to Solve NP-Hard Problem

A proprietary multi-agent framework named GenOS has been developed to autonomously evolve algorithms by having sub-agents write, compile, benchmark, and iteratively refine Rust code. This framework is designed to tackle extremely complex algorithmic challenges, with agents sharing knowledge and competing to improve their architectures over multiple generations.

In a recent experiment, GenOS was tasked with solving the "Reverse Game of Life," a notoriously NP-Hard problem that involves finding the initial state of a cellular automaton that leads to a specific target state. The framework successfully evolved three distinct peak architectures, each representing a fundamental paradigm of computer science optimization: a causal optimizer, a genetic algorithm, and a reinforcement learning agent. The causal optimizer, named Epsilon, employed a deterministic approach by mapping causal light-cones to calculate local gradients, demonstrating a sophisticated analytical strategy.