OpenAI’s chief scientist says no lab has solved alignment well enough to keep scaling at full speed

Jakub Pachocki writes that he hopes voluntary slowdowns become commonplace, and admits the company’s main way of watching what its models are thinking is getting less reliable.




