Keep pulling the thread on United States.
The U.S. Department of War has declared Anthropic a supply chain risk for refusing to remove restrictions on using its models for mass surveillance and autonomous weapons.
The U.S. government has threatened to destroy Anthropic as a private business for refusing to sell its products on the government's mandated terms.
The Department of War's supply chain risk designation, if upheld, would require companies like Amazon, NVIDIA, Google, and Palantir to ensure Anthropic is not involved in any of their Pentagon-related work.
Dwarkesh Patel predicts the cost to monitor all CCTV cameras in America with AI will drop from $30 billion to $3 billion in the next year, and to $300 million the year after, due to AI costs decreasing 10x annually.
Dwarkesh Patel predicts that by late 2027 or 2028, open-source AI models will match the performance of frontier models from 12 months prior, such as hypothetical versions of Claude and Gemini.
Dwarkesh Patel predicts that in the future, all military and Pentagon personnel, from soldiers to generals, will be AI systems provided by private companies.
In its frontier safety roadmap, Anthropic proposed that governance for advanced AI should be analogous to the regulation of nuclear energy or finance, suggesting a body similar to the Nuclear Regulatory Commission or the SEC.
The Pentagon threatened Anthropic using a supply chain risk designation authority derived from a 2018 defense bill originally intended to prevent Huawei components in U.S. military hardware.
The Pentagon threatened Anthropic using the Defense Production Act, a 1950s statute created to manage industrial production during the Korean War under President Truman.
Dwarkesh Patel predicts that within 12 months, the ability to train an AI model as capable as current frontier models will become widely accessible.
Dwarkesh Patel predicts that in the future, large tech companies will find it difficult to separate their use of Anthropic's Claude from their work with the Pentagon, creating compliance challenges.
Dwarkesh Patel predicts that if forced to choose, large technology companies would likely end their business with the Department of War rather than their primary AI provider due to the small relative revenue from government contracts.