OpenAI believes that in the next 10 years, there may be artificial intelligence systems with capabilities surpassing those of humans, known as superintelligence. These highly advanced systems can bring tremendous benefits, but if not developed with caution, they may also pose potential risks.

In this context, OpenAI has partnered with Eric Schmidt to launch the "Superalignment Fast Grants" program, committing $10 million to research the alignment and safety of future AI systems with superhuman capabilities. These grants range from $100,000 to $2 million and are open to academic researchers, non-profit organizations, as well as individual engineers and scientists. There is also a $150,000 scholarship specifically for supporting graduate students.

OpenAI researchers believe that new technologies will be essential because humans will be unable to fully understand or regulate the complex behaviors of superhuman AI. For example, an advanced AI system may generate millions of lines of code in a new programming language that is difficult for humans to evaluate as safe or dangerous. The fundamental research challenge is to ensure that models much smarter than humans reliably follow the operator's intent and avoid unintended harmful behavior.

These funds will support research directions such as weak-to-strong generalization, interpretability within models, AI-assisted human supervision of AI systems, integrity and fidelity of model reasoning chains, and more. The overall goal is to achieve breakthrough progress in controlling superhuman systems, even with limited human understanding of their complex operations.

OpenAI hopes to accelerate progress in ensuring that advanced machines continue to benefit humanity by leveraging these generous new grants and mobilizing top talents to address the most important issues in the field of technology. The organization optimistically believes that the dedication of researchers from different fields can unlock transformative solutions before advanced AI itself changes society.