AI 'AlphaCode' developed by DeepMind wins 'average' rating in programming contest

DeepMind, which handles the world's strongest Go AI ``

AlphaGo '', has developed an AI `` AlphaCode '' that can handle programming. The code written by AlphaCode scored average points in competitions, proving that it can write human-level code.

Competition-level code generation with AlphaCode | Science

Competitive programming with AlphaCode

DeepMind's AlphaCode attains 'average' rating in programming competition

AlphaCode is an AI that enables competitive programming level programming, and can successfully generate code to create the desired program or application. AlphaCode has won within the top 54.3% of the contest 'Codeforces' in which more than 5000 people participated, and has earned a rating of 'average' for the content of the code.

DeepMind announces 'AlphaCode' of automatic programming AI, it is possible to generate competition programming level code - GIGAZINE

In contests, questions that cannot be predicted in advance are submitted, and AI has to read the questions and submit the correct answer. AlphaCode is designed to read and think about what humans want AI to do and how it can be achieved.

AlphaCode doesn't teach 'rules' about how code works by structure, but rather simply trains and observes a lot of code. AlphaCode trained by this approach generates millions of different codes and filters them to find the optimal solution.

In the actual contest, AlphaCode creates a large amount of C ++ and Python programs for each problem, and finally leaves 10 candidates after the process of filtering, clustering, and ranking. As a result of submitting this for evaluation, it seems that the top 54.3% evaluation was obtained.

Codeforces founder said, 'AlphaCode's results far exceeded my expectations. AlphaCode was able to perform at the same level as a promising human competitor.I'm looking forward to the future.' I was.

in Software, Posted by log1p_kr