Sign In

Communications of the ACM

ACM TechNews

DeepMind's AlphaCode Can Outcompete Human Coders

View as: Print Mobile App Share:

AlphaCode’s coding overall performance, according to a research paper, corresponds to that of a novice programmer with a few months to a year of training.

Credit: Joe Raedle/Getty Images

DeepMind's AlphaCode model performed well against human coders in a programming competition, with a paper describing its overall performance as similar to that of a "novice programmer" with up to a year of training

AlphaCode achieved "approximately human-level performance" and solved previously unseen natural language problems by forecasting code segments and generating millions of potential solutions.

The model then winnowed them down to a maximum of 10 solutions, which the researchers said were produced "without any built-in knowledge about the structure of computer code."

Carnegie Mellon University's J. Zico Kolter wrote, "Ultimately, AlphaCode performs remarkably well on previously unseen coding challenges, regardless of the degree to which it 'truly' understands the task."

From Gizmodo
View Full Article - May Require Paid Subscription


Abstracts Copyright © 2022 SmithBucklin, Washington, DC, USA


No entries found

Sign In for Full Access
» Forgot Password? » Create an ACM Web Account