Korva is a scroll of the internet’s best ideas.
Kill good ideas for great ones
What matters most
When compute, people, and resources are limited, the question is what is most important. For OpenAI, the answer is general intelligence for knowledge work and eventually for science. Everything that goes into producing that intelligence matters most: building its own chip, building its own data centers, writing good infrastructure software, and training models.
What got cut
The proof is what got cut. Last year OpenAI killed Sora, a good and fun product that used a lot of compute and mattered less than Codex, where the compute went instead. It also killed Atlas, its web browser, which was arguably the best browser, because the talent was needed somewhere more important.
What is left
What is left is an AI offered as a service, for intellectual pursuit, for work, for scientific discovery, and for being more productive in personal life, on a flexible general platform with not a lot of other things around it.
Why does limited compute force hard choices?
Training and running AI models both depend on specialized chips, and the supply of those chips is finite. Each chip can serve only one workload at a time, so using it for one product means not using it for another. Video generation is especially costly, because a single clip contains many frames and each frame requires heavy computation. A text model answering a question needs far fewer calculations per result than a model producing seconds of video.
How did Jobs cut Apple down to focus?
When Steve Jobs returned to Apple in 1997, the company sold a confusing range of Macintosh models. He cut the lineup to four core products, arranged in a grid of consumer and professional, desktop and portable. The company was close to running out of cash at the time. Apple's later hits, the iPod in 2001 and the iPhone in 2007, came from that narrowed focus.
More from David Senra
