The most important question nobody's asking about AI.
- Source
- youtube.com
- Author
- Dwarkesh Patel
- Date
Context
In this narrated essay, Dwarkesh Patel uses the Pentagon's dispute with Anthropic to pose a question the industry rarely addresses: when AI systems do most of the work in government, business and the military, to whom should they defer? The Department of War had designated Anthropic a supply chain risk after the company kept limits on mass surveillance and autonomous weapons. Anthropic said it received the letter on March 4, 2026, considers the action not legally sound, will challenge it in court, and reads the designation's scope as narrow.
Patel's view is that the Pentagon was entitled to decline Anthropic's terms, but that threatening the company's business went further. Patel argues AI removes the labor bottleneck on mass surveillance, offering a rough calculation: a frame every ten seconds from about 100 million US cameras, at 1,000 a frame and 10 cents per million tokens with an , would cost about 30 billion dollars, a cost Patel expects to fall tenfold each year. That is a back-of-envelope estimate under stated assumptions. Patel says the question, whether AIs should defer to the company, the user, the law or their own morality, is still unanswered.
- Token — A piece of text, such as a word, part of a word or a punctuation mark, that a language model represents as numbers and handles as a unit.
- Open-weight model — A model whose trained weights can be downloaded and run outside the maker's hosted API.
- alignment — The work of making AI systems actually pursue what their builders and users intend, rather than something subtly or dangerously different.
Checking sign-in…
Loading comments…





