Marc Andreessen

By the way, Google invented the transformer. That was in 2017. But the transformer itself was a derivation of the idea of neural networks. Neural networks are an idea that goes back literally to 1943. So 82 years ago is actually when the original paper on neural networks came out.

By the way, Google invented the transformer. That was in 2017. But the transformer itself was a derivation of the idea of neural networks. Neural networks are an idea that goes back literally to 1943. So 82 years ago is actually when the original paper on neural networks came out.

And the transformer built on 70 years of research and development, much of it funded by the federal government and by European governments. at research universities for a very long time. And so this is one of those things where there's this incredibly long lineage of intellectual thought and development that's gone into it.

And the transformer built on 70 years of research and development, much of it funded by the federal government and by European governments. at research universities for a very long time. And so this is one of those things where there's this incredibly long lineage of intellectual thought and development that's gone into it.

The vast majority of the ideas that go into all these systems were not developed by the companies that are currently building the systems. Nobody sitting here, including none of our companies, have any like special moral claim that somehow we did like de novo build that we should have complete control over. Like it's just not true. So

The vast majority of the ideas that go into all these systems were not developed by the companies that are currently building the systems. Nobody sitting here, including none of our companies, have any like special moral claim that somehow we did like de novo build that we should have complete control over. Like it's just not true. So

I would describe arguments like that as frustration in the moment. And then, by the way, there are also kind of moot point arguments, which is China went ahead and did it. It's out. It's done. Now, there is an argument around copyright. If you talk to experts in the space, a lot of people have been trying to understand why DeepSeek is as good as it is.

I would describe arguments like that as frustration in the moment. And then, by the way, there are also kind of moot point arguments, which is China went ahead and did it. It's out. It's done. Now, there is an argument around copyright. If you talk to experts in the space, a lot of people have been trying to understand why DeepSeek is as good as it is.

And one of the theories, and this is an unproven theory, but one of the things the experts believe is China probably trained on data that the U.S. companies don't train on. And in particular, one of the really surprising things about DeepSeek is DeepSeek is really, really good at creative writing. DeepSeek is probably the best creative writing AI in the world right now in English.

And one of the theories, and this is an unproven theory, but one of the things the experts believe is China probably trained on data that the U.S. companies don't train on. And in particular, one of the really surprising things about DeepSeek is DeepSeek is really, really good at creative writing. DeepSeek is probably the best creative writing AI in the world right now in English.

And this is fairly weird because China's a different language. There's some very good Chinese novelists in English. Generally speaking, you'd expect the best creative writing to be coming out of the West in English. And DeepSeek is today probably the best. And it's shockingly good.

And this is fairly weird because China's a different language. There's some very good Chinese novelists in English. Generally speaking, you'd expect the best creative writing to be coming out of the West in English. And DeepSeek is today probably the best. And it's shockingly good.

And so one of the theories is DeepSeek trained, for example, there's these websites with names like LibGen that are basically giant internet repositories of pirated books. I myself would never use LibGen, but I have a friend who uses it all the time. It's like a superset of the Kindle store. It's got like every digital book and it's up there as a PDF and you can download it for free.

And so one of the theories is DeepSeek trained, for example, there's these websites with names like LibGen that are basically giant internet repositories of pirated books. I myself would never use LibGen, but I have a friend who uses it all the time. It's like a superset of the Kindle store. It's got like every digital book and it's up there as a PDF and you can download it for free.

or something, the US labs might not feel like they can just basically download all the books from LibGen and train on it, but maybe the Chinese labs feel like they can. So there's potentially a differential advantage there. That said, there is this looming copyright fight.

or something, the US labs might not feel like they can just basically download all the books from LibGen and train on it, but maybe the Chinese labs feel like they can. So there's potentially a differential advantage there. That said, there is this looming copyright fight.

People need to be careful on how they think about this because there's this looming copyright fight where certain publishing companies that would basically like to prevent generative AI companies like OpenAI, Anthropic, and DeepSeek from being able to use their content. And there's one argument that says that that material is copyrighted and can't just be used willy-nilly.

People need to be careful on how they think about this because there's this looming copyright fight where certain publishing companies that would basically like to prevent generative AI companies like OpenAI, Anthropic, and DeepSeek from being able to use their content. And there's one argument that says that that material is copyrighted and can't just be used willy-nilly.

There is another argument that basically says an AI training on a book, you're not copying the book. You're reading the book. It's the AI equivalent of reading a book. And you and I are allowed to read a book. By the way, we're allowed to check a book out from the library. We're allowed to pick a book up off the street. We're allowed to read a friend's copy of a book. Those are all legal.

There is another argument that basically says an AI training on a book, you're not copying the book. You're reading the book. It's the AI equivalent of reading a book. And you and I are allowed to read a book. By the way, we're allowed to check a book out from the library. We're allowed to pick a book up off the street. We're allowed to read a friend's copy of a book. Those are all legal.