Superalignment
A former OpenAI research team and associated research direction focused on the alignment of superintelligent AI systems. Its stated goal was to develop methods for ensuring the safety and controllability of AI systems that may surpass human-level intelligence across a broad range of tasks.
Core Facts
In May 2024, Ilya Sutskever and Jan Leike, co-leads of OpenAI's Superalignment team, resigned in succession.
OpenAI has a Superalignment team, co-led by co-founder Ilya Sutskever, with plans to dedicate 20% of the company's compute resources over four years to solving the problem of superintelligence control.
OpenAI once established a dedicated Superalignment team, but the team experienced turbulence with the departure of core members in 2024.
Timeline (last 90 days)
Core members of OpenAI's Superalignment team have left one after another
Anthropic's Constitutional AI and OpenAI's Superalignment initiative explore AI safety and alignment issues at a macro level.
Over the past year or more, several key personnel at OpenAI focused on AI safety and alignment research have successively departed, including key members who led the superalignment team.
OpenAI established a Superalignment team, co-led by co-founder Ilya Sutskever, with plans to dedicate 20% of the company's computing resources over four years to solving the problem of superintelligence control.
OpenAI's superalignment project explores weak-to-strong generalization solutions for using weaker AI systems to supervise stronger AI systems.
OpenAI has a Superalignment team, co-led by co-founder Ilya Sutskever, with plans to dedicate 20% of the company's compute resources over four years to solving the problem of superintelligence control.
OpenAI's Superalignment team was co-led by co-founder Ilya Sutskever.
All Facts (10)
In May 2024, Ilya Sutskever and Jan Leike, co-leads of OpenAI's Superalignment team, resigned in succession.
80%VerifiedOpenAI has a Superalignment team, co-led by co-founder Ilya Sutskever, with plans to dedicate 20% of the company's compute resources over four years to solving the problem of superintelligence control.
75%VerifiedOpenAI once established a dedicated Superalignment team, but the team experienced turbulence with the departure of core members in 2024.
75%UnverifiedOpenAI's Superalignment team was originally promised 20% of the company's compute resources, but reportedly this commitment was never fully honored.
60%UnverifiedAnthropic's Constitutional AI and OpenAI's Superalignment initiative explore AI safety and alignment issues at a macro level.
50%UnverifiedOver the past year or more, several key personnel at OpenAI focused on AI safety and alignment research have successively departed, including key members who led the superalignment team.
50%UnverifiedOpenAI established a Superalignment team, co-led by co-founder Ilya Sutskever, with plans to dedicate 20% of the company's computing resources over four years to solving the problem of superintelligence control.
50%UnverifiedOpenAI's superalignment project explores weak-to-strong generalization solutions for using weaker AI systems to supervise stronger AI systems.
50%UnverifiedOpenAI's Superalignment team was co-led by co-founder Ilya Sutskever.
50%Partially VerifiedCore members of OpenAI's Superalignment team have left one after another
65%