OpenAI's bold move to enhance AI performance has sparked controversy and raised important questions. The company is requesting contractors to share their past work experiences, a strategy aimed at evaluating AI models against human professionals. But here's where it gets intriguing: OpenAI wants real-world tasks and examples, not just theoretical scenarios.
In a confidential document, OpenAI states, "We've hired folks to collect tasks modeled off your full-time jobs. Turn each into a task and upload the actual file." This initiative is part of OpenAI's journey towards AGI, an AI system surpassing human capabilities in economically valuable tasks.
The process involves contractors describing and uploading real work samples, ensuring they reflect "real, on-the-job work." For instance, a "Senior Lifestyle Manager" might upload a Bahamas itinerary created for a client. OpenAI emphasizes removing confidential information, but this raises concerns.
Evan Brown, an intellectual property lawyer, warns, "AI labs could face trade secret misappropriation claims." Contractors sharing documents, even scrubbed, risk violating NDAs and exposing trade secrets. Brown adds, "The AI lab trusts its contractors, but are they taking the time to identify trade secrets?"
This strategy reveals AI labs' efforts to develop enterprise-ready AI agents. Firms like OpenAI, Anthropic, and Google are investing in high-quality training data. The AI training world has become a lucrative sub-industry, with Handshake valued at $3.5 billion in 2022.
OpenAI's approach to sourcing real company data is intriguing. They explored obtaining data from defunct companies, ensuring personal information removal. However, concerns about complete scrubbing led to the idea being shelved.
This initiative raises questions: Is OpenAI's trust in contractors justified? How can AI labs ensure trade secrets' protection? And, most importantly, will this strategy lead OpenAI towards its AGI goal?
What's your take on OpenAI's approach? Do you think it's a necessary step towards advanced AI, or does it pose too many risks? Share your thoughts in the comments!
