OpenAI may use subcontractor files from previous projects to train AI
OpenAI, together with Handshake AI, which specializes in data preparation for AI, allegedly approached third-party contractors with a request to provide examples of real tasks completed in previous and current work. This is reported by Wired.
What is known
According to the source, contractors are advised to describe the tasks they performed and upload genuine "source files." This does not refer to resumes, but to original documents: Word, PDF, Excel, presentations, images, or even codes from repositories. The probable goal is to create quality datasets that will help AI models better simulate the work of office employees.
Contractors are advised to pre-clean the files of confidential and personal information using the ChatGPT tool called Superstar Scrubbing. At the same time, intellectual property lawyer Evan Brown warns: such practice is risky and requires excessive trust in contractors who decide for themselves what is confidential.
OpenAI refused to comment.
Source: Wired