Google DeepMind

Google DeepMind reported that it was able to develop improved sorting algorithms thanks to its AlphaDev AI reinforcement learning system. The company notes that, unlike conventional approaches, the development was done from scratch using an assembly-code-based game.

When making a move, AlhpaDev evaluated the algorithm it generated and the information in the processor, and then chose the next instruction to add on a new move.

DeepMind says that the game was incredibly complicated due to the huge number of possible combinations, the number of which is similar to the number of particles in the universe or the number of possible chess combinations (10^120). And just one wrong move could cause the whole algorithm to fail.