For centuries, the world's system of assessment rested on the ability to write a coherent, deep, structured text. It meant a person could think and had absorbed the material. Essays, theses, and later written code were the main proof of intelligence. That rule no longer holds.
The quality of generation has become too high. Neural nets produce flawless research with case-law analysis, write clean code, and assemble complex business plans. Professors around the world see the same thing: a student brings in brilliant work but freezes in a one-on-one conversation, unable to explain the logic of their own conclusions.
On top of that, AI texts are everywhere, and they strongly shape our own speech and way of thinking. Telling a machine-written text from a human one is becoming very hard. Or rather — everyone has stopped caring. As a philologist, this pains me: we're simply not used to such fast, sweeping changes in language. Historical shifts used to take decades and centuries — think how long it took for the neuter gender of the Russian word "coffee" to become acceptable. But as a tech-realist, I understand why it's happening. And that nothing can be done about it. Great texts — literary, journalistic — are still easily within reach. And for that, at least, thanks.
Today many wrote about the Sochi branch of RUDN University, which scrapped written theses for law undergraduates in favor of an oral exam. But this is a global trend. Academia is changing how it grades, to test real knowledge rather than the quality of a prompt.
At Harvard students are now allowed to do anything with AI: generate, build a database, structure. But they'll have to answer out loud — to prove they themselves understand what their neural net wrote.
The same logic is at work in other universities. Faculty at Cornell are bringing back oral defenses. In the UK they look at the history of creation: students at Bath and Cardiff submit drafts and edit logs that prove the evolution of thought, not a one-prompt generation.
The same is happening in hiring. Perfect cover letters have lost their value. In April, Sundar Pichai said that 75% of new code at Google is written by AI (six months earlier it was 50%). At the same time the company is introducing an interview round with no IDE, no compiler, and no AI: the candidate gets two hundred lines of someone else's code and 45 minutes to find the bugs and explain what's wrong. The company that trusts AI with code faster than anyone checks hardest of all whether a human understands that code without it.
Text has separated from the human. For tech-optimists this is simply the next level: you no longer have to prove your worth by stringing words together smoothly — now we prove that we can find errors and take responsibility for architecture. I don't know whether that's good or bad. I deeply value literate speech, and a good text lifts my mood. Still, nothing stops me from writing and reading such texts myself. There won't be fewer of them — but finding them amid the swelling mass of worse ones will get harder.
And to prove intelligence, we'll once again have to speak, argue, and look the other person in the eye. It would be funny if rhetoric — the central subject of antiquity — returned to the stage, in the literal sense too.
P.S. I also thought that, because of the insane speed of everything around us — both creating and consuming — there's simply no time left to make a great text (and a great one takes sitting with). Maybe that, too, is an important reason to sometimes tell yourself "good enough," hit publish, and run on. But where, where are we all running?