Marc Andreessen on AI, China and Tech Policy, 2025
1:08:01 “The second way that you train it to lie is this wonderful term they have called RLHF, called Reinforcement Learning by Human Feedback, which is you basically give it teachers before you release it. You pair it with human beings and you have the human beings talk to it in like a God mode.” Share
Marc Andreessen speaks with Shane Smith about DeepSeek’s open-source reasoning model, AI regulation and his concerns about censorship and debanking. He discusses crypto and central bank digital currencies, the economic effects of AI, competition between the United States and China, Taiwan, future warfare, and whether technological progress can improve living standards while preserving freedom and meaningful work.
Passages
7:37 “What Deep SEQ did is they basically built the equivalent of that capability and then they gave it away as open source. And then giving it away as open source means that anybody can download a full AI reasoning model.” Share
33:24 “Because if you have digital banking, if the government is the bank and the money is digital, then the government, number one, sees all transactions and number two, they can turn your money off. So if the government has the ability, with a button push to take away fundamental rights and then on top of that, the surveillance mechanism to…” Share
Search what Andreessen said from Claude, ChatGPT or your own code. See plans