OpenAI’s ambitious “Superalignment” team has disbanded less than a year after its inception. This team was created with the goal of controlling future AI systems that could potentially pose existential risks to humanity.
Despite initial promises to dedicate 20 per cent of the company’s computing power over the next four years to this critical issue, internal tensions and resource struggles have reportedly led to its dissolution.
Jan Leike, one of the leaders of the Superalignment team, recently resigned, citing increasing difficulties in obtaining the necessary resources for their research in post on X (formerly Twitter). Leike expressed concerns that safety protocols had taken a backseat to the development of high-profile products. His departure highlights a growing rift within OpenAI over the company’s core priorities. This internal discord was further exacerbated by the resignation of chief scientist and co-founder Ilya Sutskever, who also played a significant role in the Superalignment initiative.
Sutskever’s departure followed his involvement in the brief and controversial firing of CEO Sam Altman, which led to significant unrest within the company. Nearly 800 employees threatened to quit if Altman was not reinstated, leading to his return as CEO just five days later. This episode, along with other departures and firings within the Superalignment team, has raised questions about the stability and direction of OpenAI’s leadership.
In response to these developments, OpenAI told Bloomberg that it would integrate the efforts of the Superalignment team into its broader research activities. The company has appointed John Schulman, another co-founder, to lead future safety initiatives. Schulman’s research focuses on large language models, and Jakub Pachocki, known for his role in developing GPT-4, will replace Sutskever as chief scientist.