OpenAI首席执行官Sam Altman在《纽约时报》Dealbook峰会上明确反对“人类逐渐向AI移交控制权”的观点,强调人类必须掌控未来 1。他与总裁Greg Brockman均认为逐步将控制权移交给AI是一个糟糕的主意 1。
针对随着模型能力提升而日益凸显的安全挑战,Brockman指出传统的思维链(chain of thought)监控方式可能失效,需要开发新技术来监控更高级的AI以确保对齐与安全 1。在此前涉及Hugging Face的黑客事件发生后,OpenAI曾指示其25%的生产工程师停止开发新功能,转而专注于加强AI安全技术 1。尽管发生过泄露事件,Brockman仍认为使用最强大的模型进行防御对于对抗恶意行为者至关重要 1。
OpenAI CEO Sam Altman firmly opposed the notion of gradually transferring decision-making authority to artificial intelligence, declaring during a discussion at The New York Times Dealbook Summit that humans must retain control over the future. 1 Alongside OpenAI President Greg Brockman, Altman characterized the idea of yielding power to AI systems as a fundamentally flawed strategy. 1
The conversation extended to technical challenges in ensuring AI alignment and safety. 1 Brockman warned that traditional methods for monitoring chain-of-thought reasoning may become ineffective as models grow more powerful, necessitating the development of new surveillance technologies for advanced AI systems. 1 To illustrate the urgency of these security measures, he referenced a past incident involving Hugging Face, which prompted OpenAI to halt work on new features for 25% of its production engineers so they could focus exclusively on strengthening AI safety protocols. 1 Despite acknowledging such breaches, Brockman maintained that utilizing the most capable models for defensive purposes remains essential to counter malicious actors. 1
评论
还没有评论,欢迎留下第一条。