OpenAI cannot make AI safe on its own [pdf] (mikitabalesni.com)

Three former safety and alignment employees of OpenAI—Tomek Korbak, Jasmine Wang, and Mikita Balesni—have publicly voiced concerns about their firings and the implications for AI safety practices within the organization. They argue that the abrupt dismissal has created a climate of fear that stifles open dialogue and collaboration, which are essential for addressing the unique risks associated with AI development. Their experience emphasizes the importance of trusting relationships with external safety organizations and a culture that encourages safety discourse, which they believe is critical for navigating the challenges of superintelligent AI. This situation is particularly significant for the AI/ML community as it raises alarms about shifts in operational norms at OpenAI, a key player in AI safety. The trio calls on OpenAI to maintain commitments to third-party safety audits, uphold the monitorability of AI models, and foster an open environment where safety concerns can be addressed freely. As AI technology progresses, embracing transparency and collaboration with external experts becomes crucial to prevent potential safety disasters, stressing that the internal culture can significantly influence the industry's approach to AI risks.