Zuck’s manifesto argues for decentralized AI

Dating in San Francisco now involves getting an earful about Mark Zuckerberg’s 6,500-word manifesto on AI published yesterday. The Meta boss painted a rosy picture of the tech’s potential to improve lives by turbocharging scientific advances and empowering people. He also expanded on how he thinks AI should develop in a decentralized manner.AI for allZuckerberg argued that AI should align with individuals’ values, rather than those of AI labs, have strong privacy controls, and be widely available for free. He thinks that top AI labs shouldn’t single-handedly control AI development, arguing for open-source models that individual developers can modify:Zuck says these models broaden access to the best AI capabilities and are key to ensuring Silicon Valley can compete with popular Chinese open-source AI.He believes US labs should be able to use each other’s AI for model training, a process called distillation that Meta’s rivals and the US government oppose.Some AI safety advocates say open-source models can be easier to exploit for malicious ends. Addressing widespread safety anxieties, Zuck also called for government oversight of AI models as they are being trained. And he announced Meta earmarked $1 billion to invest in communities near data centers.Big picture: Zuck’s essay comes as Meta races to keep up with models from OpenAI and Anthropic. Meta announced yesterday that it will develop open models that can run locally on laptops and will release an open version of its most powerful model, Muse Spark 1.2.—SK