OpenAI Is Asking Contractors to Upload Work From Past Jobs to Evaluate the Performance of AI Agents

City: San Francisco: OpenAI is asking contractors to upload their real work tasks. This is to check how well their AI models can do the same jobs as humans. The company wants to know if they can create AI that is better than people at important tasks.

In September, OpenAI started a new way to measure how their AI does against humans in many fields. They think this will show how close they are to making an AI that can do most jobs better than people can.

A document from OpenAI says, “We’ve hired people to help us collect real work tasks similar to what you do in your job. This will help us see how AI works on these tasks.” Contractors must describe the work they have done and upload examples like Word docs or PDFs. They can also share fake examples to show how they would react in different situations.

OpenAI wants real-life tasks to include both the request (what a boss or coworker asked for) and the completed work (what they made in return). The company asks that these examples show “real, on-the-job work” that the contractors “actually did.”

One example given in a presentation is from a “Senior Lifestyle Manager” who needs to make a PDF plan for a family’s yacht trip. The contractor must upload the actual itinerary they created for a client.

OpenAI reminds contractors to remove any sensitive information before they upload anything. They explain that workers should not include personal data or secret business details.

An attorney mentions that sharing sensitive work information could lead to problems for contractors if they break previous job agreements. They warn that AI companies should be careful about what information they receive because it could include private trade secrets.