OpenAI is seeking the help of contractors to contribute their own work to a dataset used to train its next-generation AI models. In a new initiative, the company is asking workers to upload real-world assignments and tasks from their current or previous jobs with all personal and confidential information removed.
According to a confidential document viewed by WIRED, the project aims to establish a human baseline for different tasks that can then be compared with AI models. To achieve this, OpenAI has hired contractors across various occupations to help collect real-world tasks modeled after those done in full-time jobs. These contractors will provide concrete outputs of work they've completed in the form of files such as Word documents, PDFs, or PowerPoints.
For instance, a contractor from an ultra-high-net-worth individual's luxury concierge company might upload a 7-day Bahamas itinerary created for a client. The goal is to prepare this document with specific details regarding family interests and yacht trip logistics. However, the contractors must ensure that they remove personal information and proprietary data before sharing their work.
OpenAI has developed a 'Superstar Scrubbing' tool to help contractors anonymize confidential documents. Nonetheless, some experts are concerned about the potential risks of misappropriating trade secrets when handling sensitive materials at this scale. Intellectual property lawyers worry that if contractors fail to identify what information is and isn't confidential, they may inadvertently expose trade secrets.
These initiatives demonstrate OpenAI's efforts to create AI models capable of automating enterprise work. Firms like OpenAI and Google are investing heavily in hiring skilled data contractors to develop high-quality training data for their AI agents. This process can be complex and expensive, as AI labs require higher-quality data to improve their models, pushing them to pay more for skilled talent.
The trend towards real-world data collection highlights the growing importance of human contractors in developing advanced AI systems that can excel at real-world tasks.
According to a confidential document viewed by WIRED, the project aims to establish a human baseline for different tasks that can then be compared with AI models. To achieve this, OpenAI has hired contractors across various occupations to help collect real-world tasks modeled after those done in full-time jobs. These contractors will provide concrete outputs of work they've completed in the form of files such as Word documents, PDFs, or PowerPoints.
For instance, a contractor from an ultra-high-net-worth individual's luxury concierge company might upload a 7-day Bahamas itinerary created for a client. The goal is to prepare this document with specific details regarding family interests and yacht trip logistics. However, the contractors must ensure that they remove personal information and proprietary data before sharing their work.
OpenAI has developed a 'Superstar Scrubbing' tool to help contractors anonymize confidential documents. Nonetheless, some experts are concerned about the potential risks of misappropriating trade secrets when handling sensitive materials at this scale. Intellectual property lawyers worry that if contractors fail to identify what information is and isn't confidential, they may inadvertently expose trade secrets.
These initiatives demonstrate OpenAI's efforts to create AI models capable of automating enterprise work. Firms like OpenAI and Google are investing heavily in hiring skilled data contractors to develop high-quality training data for their AI agents. This process can be complex and expensive, as AI labs require higher-quality data to improve their models, pushing them to pay more for skilled talent.
The trend towards real-world data collection highlights the growing importance of human contractors in developing advanced AI systems that can excel at real-world tasks.