On Tuesday, Google’s CEO revealed that AI techniques now generate greater than 1 / 4 of recent code for its merchandise, with human programmers overseeing the computer-generated contributions. The assertion, made throughout Google’s Q3 2024 earnings call, exhibits how AI instruments are already having a large affect on software program improvement.
“We’re additionally utilizing AI internally to enhance our coding processes, which is boosting productiveness and effectivity,” Pichai said throughout the name. “As we speak, greater than 1 / 4 of all new code at Google is generated by AI, then reviewed and accepted by engineers. This helps our engineers do extra and transfer sooner.”
Google builders aren’t the one programmers utilizing AI to help with coding duties. It is tough to get laborious numbers, however in line with Stack Overflow’s 2024 Developer Survey, over 76 p.c of all respondents “are utilizing or are planning to make use of AI instruments of their improvement course of this yr,” with 62 p.c actively utilizing them. A 2023 GitHub survey discovered that 92 p.c of US-based software program builders are “already utilizing AI coding instruments each in and out of doors of labor.”
AI-assisted coding first emerged in a giant approach with GitHub Copilot in 2021, and the characteristic noticed a large launch in June 2022. It used a particular coding AI mannequin from OpenAI known as Codex, which was skilled to each counsel continuations to present code and create new code from scratch from English directions. Since then, AI-based coding has expanded in a giant approach, with ever-improving options from Anthropic, Meta, Google, OpenAI, and Replit.
GitHub Copilot has expanded in functionality as nicely. Simply yesterday, the Microsoft-owned subsidiary announced that builders will be capable to use non-OpenAI fashions akin to Anthropic’s Claude 3.5 and Google’s Gemini 1.5 Professional to generate code throughout the utility for the primary time.
Whereas some tout the advantages of AI use in coding, the observe has additionally attracted criticism from those that fear that future software program generated partially or largely by AI may turn out to be riddled with difficult-to-detect bugs and errors.
Based on a 2023 study by Stanford College, builders utilizing AI coding assistants tended to incorporate extra bugs whereas paradoxically believing that their code is safer. This discovering was highlighted by Talia Ringer, a professor on the College of Illinois at Urbana-Champaign, who told Wired that “there are in all probability each advantages and dangers concerned” with AI-assisted coding, emphasizing that “extra code is not higher code.”