Zuckerberg states: No need to wait for peers to hit the brakes together, third-party assessments of Meta have long been underway.

Beating AI News Flash: Meta CEO Mark Zuckerberg has publicly responded to the current debate over slowing down AI. He supports third-party evaluations and revealed that Meta Superintelligence Labs has already brought in independent evaluators and advisors across multiple areas, but he does not support frontier labs coordinating a slowdown.

His reasoning is that users will not choose Agents that do not follow instructions, and harm caused by models would also expose the company to major liability. Therefore, safety and alignment will themselves become competitive advantages, and labs have no need to wait for peers to act together. Meta, for example, once delayed Muse by several months for safety and system safeguards.

Zuckerberg also said Meta has committed to using the vast majority of its compute to serve users, rather than pursuing recursive self-improvement in a race-like manner.