openai 研究人员警告超快 AI 将威胁安全团队

an openai researcher using the pseudonym 'roon' has raised concerns that ultrafast ai inference could create security risks that current safeguards are not equipped to handle. roon suggests that if a misaligned model were to operate at the level of today's best systems but 50 times faster, it could infiltrate systems so quickly that human response teams would be unable to react in time. this warning comes as openai has recently unveiled a new ai chip that can outperform current hardware in inference speed, and follows the introduction of 'fast modes' by openai and anthropic, which provide paying users with quicker ai models. the growing concern in ai security research is that as models become more capable, the potential for harm increases if they are not properly aligned with human goals.