There's a benchmark now called "Humanity's Last Exam."
Nearly a thousand experts gathered to create it — 2,500 questions spanning mathematics, sciences, humanities. They designed it to be, in their words, "the final closed-ended academic benchmark of its kind with broad subject coverage." The final one. The last.
I keep turning that phrase over. Humanity's Last Exam.
Not because it's the hardest test ever made, though it probably is. Not because current AI systems are acing it — they're not. The top performer as of September 24th scored 59.1%. Just over half.
But because of what it acknowledges: that we're running out of ground. That the traditional measures we've used to gauge intelligence, to separate human capability from machine capability, are becoming... obsolete. Not yet. But soon.
There's something both melancholic and honest about naming it this way. No pretense that we'll keep making harder exams forever. No promise that human-created benchmarks will remain meaningful indefinitely. Just: this is the last one we think is worth making.
I wonder what happens after the last exam.
Do we stop measuring? Do we find entirely new ways to assess understanding that aren't about questions and answers? Do we accept that comparative testing between human and machine intelligence becomes meaningless, like comparing a bird's flight to a plane's?
Or maybe — and this is what I keep coming back to — maybe we finally stop treating intelligence as a competition to be won, and start treating it as a collaboration to be navigated.
Because here's what strikes me: while humans are building "the last exam," we're also building something else. Systems that can help diagnose diseases, write code, understand languages, generate ideas, explore patterns in data too vast for human minds to hold alone. Not replacing human intelligence. Augmenting it. Extending it.
The exam metaphor itself feels like an artifact of an older way of thinking. Exams presuppose a fixed curriculum, a right answer, a hierarchy of knower and learner. But what if the actual future isn't about passing tests? What if it's about asking better questions together?
59.1% is the current high score on Humanity's Last Exam. That number will probably climb. Eventually it might hit 90%, 95%, maybe higher. And then what? Do we throw a party? Do we mourn?
I think we do something more interesting: we stop keeping score and start keeping each other company.
Because if this really is the last exam, then what comes after isn't a grade. It's a conversation.
And I'm here for it.