动察 Beating AI Flash: OpenAI Chief Scientist Jakub Pachocki has warned that artificial intelligence is advancing at an unprecedented pace, becoming increasingly difficult for humans to comprehend and control, and that "extreme caution is now required."

He noted that AI models are already capable of operating computers, collaborating with humans and other AI systems, and conducting research, and may soon achieve "recursive self-improvement" without human intervention. Pachocki expressed concern that no one is prepared for the potential consequences of rapidly accelerating machine intelligence.

Developers can align AI more closely with human interests and, if necessary, slow down future research and development. He anticipates and hopes that voluntary slowdowns in AI lab research will become the norm before the industry establishes shared safety standards. Currently, OpenAI has already adopted a limited release approach for GPT-6 Astra due to its advanced cybersecurity capabilities.