“Find something that you can't build yet and the only bottleneck is the AI model itself… then start building that and just bet that in a year you're going to have GPT-5 capable of doing that, and you have one full year as a head start.”

Bet on the model curve — build for the capability that is coming next

Pick an idea that the current models can almost — but not quite — handle. By the time the next model lands, the product is ready and you have a year-long head start on everyone who waited for the capability to arrive.