The AI ​​revolution seems to be here, but so far these systems have not been used, for example, to discover some extraordinary drugs or some materials with special properties. Real practical applications seemed limited to the creative realm, but DeepMind has proven time and time again that these systems can go even further.

AlphaZero’s heir. The people in charge of DeepMind had created AlphaZero from scratch to learn how to play chess, but now they have created a derivative of this system, AlphaDev, which has a different purpose.

ranking lists. The exact purpose of AlphaDev was to try to discover the best way to sort lists of items. This AI system has managed to discover a method that is 70% faster than traditionally used algorithms, which is very important for many fields, but has an immediate impact in one field in particular.

Programmers took advantage. Although news of this discovery is now published in the prestigious Nature, the technique of sorting items into a list has been available for months. In January 2022, DeepMind submitted its solution to the organization that manages the C++ language, one of the world’s most popular languages, and after two months of rigorous analysis, this algorithm was incorporated into the language: it was the first change made to the algorithm. in C++ for over a decade.

Also improves encryption. DeepMind has also added new discoveries to Abseil, a set of C++ algorithms used in cryptography that allow working with hashes, which are the unique IDs of all types of data. At DeepMind, they believe their algorithms are used “billions of times a day” and speed up their processing by 30%, which definitely represents a remarkable improvement in this department.

The algorithm created by AlphaDev made it possible to save an assembly instruction when sorting lists of three items. For those with five, the gain was even greater. Source: Nature.

Assembly language returns. To achieve their goal at DeepMind, they worked as a foundation with assembly language, a very low-level programming language that allows algorithms to be broken down into small steps that are then useful for finding “shortcuts” and new solutions to similar problems. You found DeepMind. As with AlphaZero and chess, AlphaDev found the solution by “playing a million games” or, in his case, trying millions of combinations of assembly instructions to generate even a single new command for his algorithm.

Fewer instructions, better. DeepMind’s Daniel Mankowitz explained in the MIT Technology Review how they work with algorithms to organize short lists of three or five items commonly used in all types of development. His team explained that the best human version of the algorithm for lists with three elements is the 18 instructions in the assembler. AlphaDev managed to build an algorithm with 17 instructions. Sorting lists of fives requires 46 instructions in the best human algorithm, and AlphaDev cut that down to 42. It doesn’t sound like much, but running the algorithm on an Intel Skylake chip took 6.91 nanoseconds to sort a list of five. 2.01 nanoseconds, 70% faster.

limitations. Still, AlphaDev has its limits. The longest algorithm you can write is 130 instructions because when you try to combine more, the number of possible combinations is even greater than the possible number of moves in chess (10^120) or atoms in the universe (about 10^80). ). However, since DeepMind works directly with other algorithms or languages ​​such as C++ as its creators stated, we are likely to see new developments in this regard. Shortcuts may not be as flashy, but because they are higher level, these improvements can be applied to other scenarios.

Image | Midjourney and Xataka

on Xataka | AlphaZero has a solution to make chess more fun: no castling