Superintelligent Ai

OpenAI Superalignment Team Faces Internal Strife, Leading to Key Resignations

In a universe not far from our own, OpenAI's Superalignment team, tasked with the formidable challenge of governing and steering "superintelligent" AI systems, has faced a critical juncture. Initially promised 20% of the company's compute resources, their requests for even a fraction of that capacity were frequently denied, obstructing their essential work.

This and other challenges culminated in the resignation of several key team members this week, including co-lead Jan Leike. A former DeepMind researcher, Leike contributed significantly to the development of pivotal AI models like ChatGPT, GPT-4, and its predecessor, InstructGPT, during his tenure at OpenAI.

Leike publicized his reasons for departing on Friday morning. "I have been disagreeing with OpenAI leadership about the company’s core priorities for quite some time, until we finally reached a breaking point," Leike wrote on X. He emphasized the need for increased focus on preparing for future AI models, highlighting areas such as security, monitoring, safety, and societal impact. "These problems are quite hard to get right, and I am concerned we aren’t on a trajectory to get there," he added.

    "Building smarter-than-human machines is an inherently dangerous endeavor. OpenAI is shouldering an enormous responsibility on behalf of all of humanity."

    — Jan Leike (@janleike) May 17, 2024

Despite these concerns, OpenAI did not immediately respond to inquiries regarding the allocation of resources to the Superalignment team.

Formed in July of the previous year, the Superalignment team, led by Leike and OpenAI co-founder Ilya Sutskever, aimed to address the core technical challenges of controlling superintelligent AI within four years. The team included scientists and engineers from OpenAI's alignment division and researchers from various other organizations within the company. Their mission extended to contributing safety research and collaborating with the broader AI industry through initiatives like a research grant program.

While the team managed to produce significant safety research and distribute substantial grants to external researchers, their efforts were hampered as OpenAI's leadership increasingly prioritized product launches. Consequently, the Superalignment team struggled to secure essential upfront investments necessary for the company’s mission of developing superintelligent AI for the benefit of humanity.

Leike's departure came amid a backdrop of internal conflict, notably Sutskever’s dispute with OpenAI CEO Sam Altman. Sutskever, along with OpenAI’s previous board of directors, attempted to oust Altman last year over transparency concerns. Under investor pressure, Altman was reinstated, resulting in board resignations and Sutskever’s departure.

Sutskever played a critical role in the Superalignment team, contributing research and facilitating communication between divisions within OpenAI. His departure leaves a significant gap in leadership and advocacy for the team’s mission.

Following Leike’s resignation, Altman acknowledged the need for further efforts in a post on X, hinting at a broader explanation provided by co-founder Greg Brockman. Brockman expressed gratitude for Leike's contributions and outlined OpenAI’s commitment to a balanced approach to safety and capabilities, emphasizing the importance of rigorous testing and world-class security.

In the wake of these departures, John Schulman, another OpenAI co-founder, has taken the helm of the work previously managed by the Superalignment team. Instead of maintaining a dedicated team, OpenAI plans to embed researchers focused on these issues across various divisions, a move described by an OpenAI spokesperson as “integrating [the team] more deeply.”

However, there are concerns that this restructuring might dilute the focus on AI safety, potentially compromising the company's ability to navigate the complexities of superintelligent AI development.