Jan Leike

OpenAI’s huge push to make superintelligence safe | Jan Leike

Jan Leike

Co-led by Jan Leike and Ilya Sutskever, OpenAI has launched the Super Alignment Project to resolve AI safety challenges within four years by dedicating 20% of its compute resources to developing scalable oversight and automated interpretability methods. The initiative explicitly targets the limitations of current human-feedback techniques by creating systems capable of evaluating superhuman intelligence, with a strategic roadmap prioritizing the alignment of human-level researchers before AGI arrives. While seeking to hire over ten new specialists to build these solutions, the project remains committed to publishing empirical evidence for external scrutiny and advocating for industry-wide safety standards if technical alignment lags behind capability growth.