Greg Brockman, co-founder of OpenAI, has revealed that artificial intelligence tools generate up to 80% of the code in the company's internal projects. This data not only marks a milestone in development automation but also raises questions about the future role of human programmers in an increasingly machine-assisted environment.
How code assistants work in production environments 🤖
Language models, such as GPT-4, are integrated into workflows through plugins that suggest functions, fix errors, and generate complete blocks of code in real time. At OpenAI, these assistants are trained on internal codebases, achieving 90% accuracy in refactoring tasks. However, human review remains necessary to avoid logical errors and security vulnerabilities, especially in critical systems.
Programmers, from creators to correctors of AI 🧑💻
Now it turns out that developers spend their time reviewing code written by a machine. It's like having an intern who writes 80% of the work, but without asking for coffee or complaining about the schedule. The problem is that when the AI makes a mistake, it does so with astonishing confidence, and you have to untangle the mess while it asks for more training data. The future of the programmer: a style corrector with shortcut keys.