A Semafor report claims that OpenAI has hired roughly 1,000 contractors in the last six months to improve the company's AI coding and other capabilities.  Sources said the new hires are working on both labeling data and creating datasets to train the company's AI models. The contractors are based in regions such as Eastern Europe and Latin America. About 60% of the staffers were hired to label data, such as images and audio, for training OpenAI models. The remaining 40% are said to be programmers tasked with creating data to teach the models software engineering tasks. According to Semafor, these contractors may be working on a dataset featuring both lines of code and human-written natural language descriptions of that code. OpenAI's existing code-generating system, Codex, was already trained on open-source code from Microsoft's GitHub. Codex powers the GitHub Copilot coding assistant service that's become popular among programmers.