OpenAI’s Superalignment team, responsible for developing ways to govern and steer “superintelligent” AI systems, had been promised 20% of the company’s compute resources. However, they often had their requests for compute denied, hindering their work. This issue, along with others, led to several team members resigning, including co-lead Jan Leike. Leike, who was involved in the development of ChatGPT and its predecessors, went public with his reasons for resignation, stating that he disagreed with OpenAI’s core priorities.
Leike believed that more resources should be allocated towards preparing for the next generations of AI models, focusing on security, monitoring, safety, alignment, and societal impact. He expressed concern that OpenAI was not on the right trajectory to address these important challenges. Building smarter-than-human machines is inherently dangerous, and OpenAI holds a significant responsibility on behalf of humanity.
Although OpenAI did not immediately comment on the promised resources, it is clear that the Superalignment team faced obstacles in obtaining the necessary investments to fulfill their mission. The team was formed with the goal of solving the technical challenges of controlling superintelligent AI and had made contributions to safety research and grant programs. However, as product launches took precedence and distractions such as conflicts between Ilya Sutskever and CEO Sam Altman arose, the team struggled to secure the upfront investments they believed were crucial.
Sutskever played a significant role in the Superalignment team, contributing research and acting as a bridge to other divisions within OpenAI. His departure from the company after conflicts with Altman added to the distractions and challenges faced by the team. Altman acknowledged that there was more work to be done and expressed commitment towards it. Co-founder Greg Brockman provided further explanation, emphasizing the need for a tight feedback loop, rigorous testing, security, and safety.
Following the departures of Leike and Sutskever, John Schulman, another OpenAI co-founder, has taken over the work previously done by the Superalignment team. However, the team will no longer exist as a dedicated unit but rather as a loosely associated group of researchers embedded in different divisions within the company. This restructuring raises concerns that OpenAI’s AI development may not prioritize safety as much as it should.
In conclusion, the resignations of key members of OpenAI’s Superalignment team highlight the challenges faced in developing superintelligent AI systems responsibly. The team’s struggles to secure necessary resources, conflicts within the company, and the shift in structure raise concerns about OpenAI’s commitment to safety in AI development. It is crucial for OpenAI and similar organizations to prioritize safety, security, and societal impact in order to mitigate the inherent risks associated with building machines more intelligent than humans.