Google’s AI-Generated Code Claims Under Fire: Engineers Call Out TabCompletion Hype
San Francisco, CA – Google’s recent boast about AIgenerating over 25% of its code has sparked controversy, with the company’s own engineers questioning the accuracy of the claim and highlighting the limitations of current AIcoding tools.
During the company’s third-quarter earnings call last week, CEO Sundar Pichai proudly announced, Over a quarter of our new code is generatedby AI, then reviewed and accepted by engineers. This statement, while intended to showcase Google’s AI prowess, has been met with skepticism and even ridicule from both the public and within Google itself.
One of the first to raise concerns wasHenry T Kirk, co-founder of Google’s iOS and Android product experience team, who commented, I bet most of that is prototype files or configuration files. Google’s codebase has a ton of this kind of ‘code’ thatis a pain to deal with. I know many L3 engineers whose only job is to fix these files.
However, the most damning critique came from within Google’s ranks. An anonymous engineer, identified as @asdfman123 on Hacker News, posted a direct rebuttal to Pichai’s claim. I work at Google and just finished a day of work. I’m writing what they call ‘AI-generated code.’ But the code auto-completion engine is basically good at finishing the code I’m already writing. If I write ‘function getAc…’, it will intelligently complete it as ‘function getActionHandler()’, maybe even suggest the correct parameters and a decent jsdoc comment. So basically, it’s a useful productivity tool, but it doesn’t do any engineering design at all. It’s probably about as good as Copilot, maybe slightly worse. (Although I haven’t used it recently).
Thissentiment was echoed by another anonymous engineer, @NotAnOtter, who commented, I also work at Google (until recently). The tools like Copilot don’t actually accelerate code development in any meaningful way.
These insider accounts paint a stark contrast to Google’s public pronouncements. While AI coding toolslike Copilot can undoubtedly enhance productivity by automating repetitive tasks and providing suggestions, they are not yet capable of replacing the core engineering skills required for designing and building complex software systems.
The controversy highlights the ongoing debate surrounding the role of AI in software development. While AI tools are becoming increasingly sophisticated, they remain far from achieving thelevel of autonomy and creativity that would allow them to fully replace human engineers.
This incident serves as a reminder that claims about AI breakthroughs should be scrutinized with a critical eye, particularly when they come from companies with vested interests in promoting their own technologies. As AI continues to evolve, it’s crucial to maintain abalanced perspective and recognize its limitations alongside its potential.
Views: 0