1. Resuming Open Source Approach
Meta will resume releasing open weighted models. After restructuring its AI business last year, Meta suspended the release of such models. Zuckerberg also advocated allowing "distillation." He wrote, "I think we should protect the principle that you have the right to learn from anything you can observe."
2. Investing More in Data Center Communities
Zuckerberg stated, "Sustainable infrastructure means that every project must provide significant benefits to the local community." Meta will also launch a $1 billion "Future Is For Everyone Fund" to directly support communities surrounding its data centers.
3. Hoping for Earlier Government Intervention in New Model Release Process
The Trump administration is implementing an executive order that includes a voluntary 30-day review period for new models. Zuckerberg opposes this, arguing that 30 days is "quite a long time." His solution is: "A close, proactive partnership should be established between cutting-edge AI labs and governments, rather than mechanically following a rigid process and fixed review schedule in all cases." He also stated that AI labs "should not wait until the model training is fully completed before being reviewed, but should provide intermediate training checkpoints to governments during the training process for government use and evaluation."
4. Strengthening Internal Model Release Oversight Within Companies
Zuckerberg stated that Meta is adopting a new governance structure that empowers the company's board of directors to determine whether new models meet established safety standards, which will also be approved by the board. He wrote, "I don't believe it's in my best interest, Meta's, or the world's best interest to make me or anyone else the sole decision-maker on how superintelligence is deployed." He also called on other companies to adopt a similar structure and believes the entire industry should establish an "industry-wide unified version" of governance.
5. Ultimately, Users Should Decide What AI Should Be
Zuckerberg has previously used the concept of "personal superintelligence" to describe Meta's vision for AI. In the article, he further elaborated on his ideas. He believes that users, not model developers, should determine the values followed by AI systems. He also suggested that personal AI agents should have a “completely private mode” where Meta or any other service provider cannot see the user’s information or grant access to such information to third parties.
