> At a Glance
> – OpenAI is asking contractors to upload actual workplace tasks to benchmark AI models
> – Contractors must strip personal and proprietary data but submit real deliverables
> – The effort aims to set a “human baseline” for evaluating progress toward AGI
> > Why it matters: The project shows how AI labs are racing to prove their models can match or beat humans at economically valuable work.
OpenAI wants contractors’ real work files-spreadsheets, slide decks, code repos-to see how its next models stack up against human professionals, according to confidential documents viewed by News Of Fort Worth.
What OpenAI Wants
The company is recruiting across occupations and instructing workers to:
- Pick complex, multi-hour assignments they’ve actually completed
- Upload the original deliverable, not a summary
- Remove any personal data, trade secrets, or non-public information
One example task: a 2-page Bahamas yacht itinerary drafted for a luxury-concierge client. Contractors can also submit fabricated but realistic samples if they lack usable past work.
Data Scrubbing Risks
OpenAI provides a ChatGPT tool dubbed “Superstar Scrubbing” to help anonymize files. Yet intellectual-property lawyer Evan Brown warns:
> “The AI lab is putting a lot of trust in its contractors to decide what is and isn’t confidential. If they do let something slip through, are the AI labs really taking the time to determine what is and isn’t a trade secret?”
Brown notes that both contractors and OpenAI could face trade-secret misappropriation claims if sensitive data leaks.
Why Real Tasks Matter
| Component | Description |
|---|---|
| Task Request | What a manager or colleague asked the person to do |
| Task Deliverable | The actual file the worker produced in response |
OpenAI says pairing real requests with real outputs is critical for measuring how close its models are to AGI-systems that outperform humans at most economically valuable tasks.
Key Takeaways
- OpenAI will pay contractors for authentic workplace samples across industries
- All files must be stripped of personal and proprietary information
- The push highlights the premium AI labs now place on high-quality, real-world training data
The program underscores the urgency among AI developers to benchmark their systems against genuine human output before claiming human-level competence.
