Lex Fridman: belief

Transcript context

…… the early copyright lawsuits have fallen in the favor of AI training. I would say that the long tail of use is going to go inside of AI, which is if you scrape trillions of tokens of data, you’re not looking and saying, “This one New York Times article is so important to me.” But if you’re doing a audio generation for music or image generation, and you say, “Make it in the style of X person,” that’s a reasonable case where you could figure out what is their profit margin on inference. I don’t know if it’s going to be the 50/50 of YouTube Creator Program or something, but I would opt into that program as a writer, please. It’s going to be a rough journey, but there will be some solutions like that that makes sense. But there’s a long tail where it’s just on the internet. I think one of the other aspects of that Financial Times article implied, and so that leads to a more general question. Do you think there’s… How difficult is spying, espionage, and stealing of actual secret code and data from inside of companies? How much of that is being attempted? Code and data is hard, but ideas is easy. Silicon Valley operates on the way that top employees get bought out by other companies for a pay raise, and a large reason why these companies do this is to bring ideas with them. And there’s no… I mean, in California, there’s rules that certain non-competes or whatever are illegal in California. And whether or not there’s NDAs and things, that is how a lot of it happens. Recently, there was somebody from Gemini who helped make this 1 million context length. And everyone is saying the next Llama who, he went to the Meta team, is going to have 1 million context length. And that’s kind of how the world works.…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.