The news of September 17thMeta CEO MARK ZUCKERBERGMark ZuckerbergThis post is part of our special coverage Syria Protests 2011AI Laboratories do not need to slow down through industry coordination. In his view, market competition, legal responsibility and reputational risks already provided sufficient security incentives for companies, and it would be more feasible for independent evaluation agencies to continuously test front-line models。
This statement responds to the industry slowdown proposed by Anthropic CEO Dario Amodei. According to Zuckerberg, the Meta super-smart laboratory has used a third-party assessment and will be delayed for release when risks are discovered; previously, the company had delayed the Muse model on security grounds. At the same time, he stressed that much of Meta's calculations were still being used to improve existing products rather than to allow the model to evolve itself。
His differences focused on those responsible. The collective commitment of the industry can easily be coordinated by a small number of laboratories, while independent assessments allow different agencies to compare the capabilities, controllability and risk of abuse with a unified approach, with regulation and the market holding individual companies accountable。
Zuckerberg also distinguishes between current product development and model incremental self-improvement. He said that much of Meta ' s calculations were used to improve the products that users were using; even if companies continued to accelerate their research and development, forward models should be released through external assessments and internal security thresholds。