The sorting algorithms that DeepMind’s AlphaDev discovered through reinforcement learning were contributed to the LLVM libc++ standard sorting library. DeepMind described it as the first change to that part of the library in over a decade, and the first such addition designed using reinforcement learning.
Because sorting is one of the most common operations in software, the impact is large: DeepMind estimated the routines are executed trillions of times every day. AlphaDev found them by treating the search for fast code as a single-player game played in low-level assembly instructions, building on the AlphaZero lineage.