Co-creator of RLHF joins OpenAI board

Paul Christiano, a co-author of RLHF (the “reinforcement learning from human feedback” technology on which all modern language models are based) and founder of Alignment Research Center, has been appointed to the OpenAI Board of Directors and to the Safety Committee, which has veto authority over releasing new models. After leaving OpenAI in 2021, he went on to sound the alarm about the “significant risk of catastrophic and irreversible loss of control” from the rapidly increasing capabilities of AI.

OpenAI has brought onto their Board of Directors an individual who helped develop a critical AI alignment tool — who openly regards it as an inadequate defense against catastrophe. They have also filled a seat on the Safety Committee with a voice focused on reducing release velocity instead of increasing it; this represents a crucial shift in the dynamics governing the world’s leading AI research institute.