ImprovEvolve: Ask AlphaEvolve to Improve the Input Solution and Then Improvise
This new AI technique just crushed previous benchmarks on notoriously difficult optimization challenges.
Deep Dive
Researchers introduced ImprovEvolve, a new AI technique that enhances LLM-guided evolutionary computation like AlphaEvolve. It evolves a program that can propose, improve, and perturb solutions to complex problems. The method achieved new state-of-the-art results in hexagon packing for 11, 12, 15, and 16 hexagons, and set a new record for the second autocorrelation inequality with a lower bound of 0.96258, beating AlphaEvolve's 0.96102.
Why It Matters
It demonstrates a more efficient path for AI to solve complex, real-world optimization problems previously thought intractable.