Susan Ferrari

When Humanity’s Last Exam launched in January of last year, the best AI models answered only a small fraction of its questions correctly. Today, just 18 months later, the leading model is past 50%. That fact is at odds with our instinct to find the best AI tool and commit. Learn it deeply. Build everything around it. That instinct made sense five years ago, when tools changed slowly enough to reward that kind of loyalty. It doesn’t work that way anymore. That’s good news once you see it the right way.