Sam Altman: OpenAI has paused RL training on some frontier models to ensure alignment, safety, and monitoring standards can keep pace

Odaily News: OpenAI CEO Sam Altman posted on X, saying that OpenAI has paused reinforcement learning (RL) training on some frontier models to ensure that its alignment, safety, and monitoring measures can keep pace with the new wave of model capability improvements. Altman said that model progress is currently "extremely rapid," and if the pace of model capability development exceeds the progress of safety and alignment efforts, OpenAI will take action.

Altman said that the future pace of AI development will increasingly depend on confidence in model safety, and that the entire industry will ultimately need to coordinate around common safety standards. Until then, OpenAI will take the lead in acting.