OpenAI's Superalignment project aimed to control AI smarter than humans, but collapsed due to conflicts

From Future PLC: 2024-07-05 14:23:21

July 5, 2024 marks what would have been the first anniversary of OpenAI’s project, Superalignment. The ambitious project aimed to control AI systems smarter than humans. However, the project fell apart less than a year later when co-founders Sutskever and Leike left the company due to internal conflicts.

During its short lifespan, Superalignment achieved key milestones. The group published a research paper on weak-to-strong generalization, exploring how humans can supervise superhuman AI. They also offered $10 million in fast grants to applicants. The project aimed to ensure AI systems followed human intent.

The question of supervising superintelligent AI remains unanswered as Superalignment’s future is uncertain. While the concept of human-safety-minded machines is groundbreaking, the potential risks are also significant. Sutskever and Leike may continue exploring this challenge in the future, but for now, Superalignment remains a dream unfulfilled.



Read more at Future PLC: A year ago today, OpenAI launched a bold four-year plan that didn’t even last a year