threat of nuclear retaliation. Maybe that sounds kind

The use of LLM outputs for training LLMs, known as synthetic data, is still controversial, but it appears to complement human training data to some extent.207 Training a successor system requires better chips, data, and algorithms—domains where existing AI has already made some progress.208

Lastly, prior bubbles didn’t have the most profitable companies that humanity has ever created investing in them. They were debt financed instead. AI is not debt financed yet.

Take self-driving cars, for instance. Take a self-driving car operating in the Bay Area. Do you think you could just drop it in New York City, or drop it in London, where people drive on the left? No, it’s going to fail. Not only can it not generalize to a change in driving rules, you cannot even make it generalize to a new city. It needs to be

To get real creativity, you need to search through spaces of possibilities and find these hidden gems. That’s what creativity is.

Evolution doesn’t have foresight. What gets more surviving offspring and grandchildren in this generation is the thing that becomes more common. Evolution doesn’t think, “If you do this, then in a million years, you’ll have a lot of descendants.” It’s about what survives and reproduces now.

We spend more compute by having a larger brain than other animals—more than three times as large as chimpanzees—and by having a longer childhood. We’re spending more compute in a way that is analogous to having a bigger model and training it for longer.

One of the first things [OpenAI cofounder and former chief scientist] Ilya Sutskever said to me was, “Look. The models just want to learn. You have to understand this. The models just want to learn.” It was a bit like a Zen kōan. I listened to this and I became enlightened.

It’s like having a brilliant but somewhat deluded and amnesic coworker.