OpenAI has unveiled a groundbreaking initiative, offering a whopping $10 million in grants to fuel technical research aimed at tackling the intricate challenge of controlling vastly intelligent artificial intelligence systems, surpassing the capabilities of the average human.
These grants, dubbed Superalignment Fast Grants, seek to unravel the mysteries surrounding the alignment of future superhuman AI systems. OpenAI envisions these grants as a beacon of insight into how robust models derive inference from minimal guidance, delving into the realms of weak supervision.
Among its ambitious goals, OpenAI aims to unravel the enigma of using AI systems to assess outputs generated by newer AI models, as well as constructing an AI lie detector. The company emphasizes that comprehending these superhuman AI systems might surpass human capability; faced with millions of lines of intricate code, evaluating its safety or potential hazards becomes an arduous task.
In a candid post on a designated platform, OpenAI articulated, “Solving the challenge of ensuring alignment and safety in future superhuman AI systems stands as one of the world’s most pressing technical puzzles. Despite its complexity, we firmly believe it’s a surmountable obstacle. There’s abundant potential waiting to be tapped, and fresh minds can make monumental strides!”
These grants primarily target academic institutions, non-profit organizations, and individual researchers. Additionally, OpenAI extends its support through the OpenAI Superalignment Fellowship—a one-year program offering $150,000 for graduate students diving into this critical arena.
Identifying seven key practices to uphold the safety and accountability of AI systems, OpenAI aims to channel its funding towards exploring lingering queries that arose from their research endeavors.
Part of this initiative involves Agentic AI Research Grants ranging from $10,000 to $100,000, aimed at empowering researchers to investigate the implications of superintelligent AI systems and devise strategies ensuring their safety.
