![OpenAI Dissolves Superalignment Team](https://dt2sdf0db8zob.cloudfront.net/wp-content/uploads/2024/06/OpenAI-Dissolves-Superalignment-Team-850x563.webp)
OpenAI Dissolves Superalignment Team
OpenAI’s Superalignment team has been dissolved, and the remaining researchers will be integrated into various teams throughout the company.
OpenAI announced the creation of the Superalignment team in July 2023. It was designed to focus on research that would evaluate the risks and ensure the safe construction of AI models that exhibit Artificial General Intelligence (AGI), that is computer systems that could be smarter than humans.
The team was co-led by Ilya Sutskever and Jan Leike, who were promised 20% of OpenAI’s computing power to carry out this research. During the past few months, the team has published various research studies on AI safety and, in December, launched its Fast Grants program, which offered up to $10 million in research grants.
This team was not focusing on the deployment of products available today, where existing teams are ensuring their safety, but on identifying the risks of much more powerful models that could be built in the years to come.
OpenAI believed this form of superintelligence could arise in the next decade and that these “AI systems would have vast capabilities—they could be hugely beneficial, but also potentially pose large risks.”
While OpenAI has not officially announced the organizational changes, it has announced the departure of Ilya Sutskever, one of its co-founders, in a blog post.
The second co-lead of the team, Jan Leike, resigned on the same day. In a series of posts on X, he explains that he is no longer aligned with the company’s core priorities.
Leike posted, “I believe much more of our bandwidth should be spent getting ready for the next generations of models on security, monitoring, preparedness, safety, adversarial robustness, (super)alignment, confidentiality, societal impact, and related topics. These problems are quite hard to get right, and I am concerned we aren’t on a trajectory to get there.”
John Schulman, another OpenAI co-founder, has taken over the responsibility of the work done by the Superalignment team even though the dedicated team no longer exists. The researchers will be “embedded in divisions throughout the company.”