August 11th.Meta CEO MARKZuckerbergYesterday, in a long article entitled " The future belongs to everyone " , he argued that super intelligence should be given widely to individuals, not to single companies, governments or AI The system determines the goals and values of all. He listed "personal empowerment, invention and power checks and balances" as an AI route for Meta and, in an interview with Axios, stated that he was more concerned about concentration of power than about a single risk in industry discussions。
In his view, the smart body was responsive to each user ' s own goals and helped different people to pursue different choices within legal and security borders. Meta plans to provide a completely private model of personal intelligence that prevents companies and service providers from viewing or authorizing access to user information; free or low-cost versions are available to a wide range of users, and those who need more calculus purchase resources through dynamic competitive bidding。
At the same time, Zuckerberg suggested that forward laboratories should provide the United States Government with model training intermediate checkpoints and technicians in advance for the reinforcement of critical infrastructure, rather than for start-up review when the model is released. In his view, such cooperation would allow the Government to acquire security capabilities ahead of time, while avoiding the slow release of the harmonized review cycle model。
With regard to open sources, he argued that the United States should reduce restrictions on training data and model distillation, arguing that model learning from other models was part of the open source ecology and that the principle of "learning from anything that can be observed" should be protected. Meta will resume the release of the partial open model and the model will be reviewed by an independent commission to issue safety standards。