Seven minute read

A Take-Home Final Went In-Person and Scores Halved: The Real Lesson

A Brown University professor moved his final from take-home to in-person after suspecting AI cheating. The class average fell from 96% to 48.6%. Here's what actually happened, why it's a trend and not a one-off, and the honest takeaway for students on where AI tools help and where they'll sink you.

Short answer. A Brown University economics professor let his class take midterms and finals at home, got suspicious when the take-home midterm averaged 96%, and moved the final to in-person. The class average dropped to 48.6% — a record low for a course he'd taught for nearly two decades. The headline writes itself ("AI cheating exposed"), but the more useful lesson for students is quieter: the people who got hurt weren't hurt by the tool, they were hurt by outsourcing understanding they were later tested on in a room where no tool could reach them. That distinction is the whole game, and it's worth understanding before your next exam.

What actually happened

The facts, from reporting by Inside Higher Ed, Ars Technica, and others:

  • The professor is Roberto Serrano, who has taught economics at Brown for close to twenty years. The course was his welfare economics and social choice theory class (ECON 1170), long known as difficult.
  • After a December 2025 shooting on Brown's campus that killed two students and injured nine, Serrano moved his spring exams to a take-home format to ease students' anxiety about sitting in a room for a high-stakes test. A humane call.
  • Once he advertised take-home exams, enrollment nearly tripled — to 86 students, up from a typical ~30.
  • The take-home midterm averaged 96%. His historical average for the course was 65–80%, and he'd made the exam harder than usual. He ran the questions through ChatGPT and found answers "mirroring what his students had written" — "kind of correct, but very off and with a very convoluted style."
  • He voided the midterm and made the final in-person.
  • Then the numbers moved: 18 students dropped the course, and 9 more stayed enrolled but didn't show up for the final. Reporting indicates most of those students had aced the take-home midterm.
  • Of the 59 who sat the in-person final, the average was 48.6% — never previously below 65% in the course's history. Three scored zero. Only two students landed within 10% of their own midterm score, and just one of them did better in person.

That last line is the tell. If your take-home score and your in-person score describe two different people, one of them wasn't you.

The pattern hiding in the numbers

Strip away the "Ivy League" framing and the story is about a gap. A student who genuinely learned the material scores roughly the same whether the exam is at home or in a room. A student who didn't scores 96% at home and then either bombs the in-person version or doesn't show up at all.

The students who dropped or skipped after the format changed made a rational move given their situation — but the situation was one they built. They'd spent a semester producing scores that didn't correspond to anything they could reproduce. When the format shifted, there was nothing to fall back on.

That's the actual risk, and it has nothing to do with getting "caught." Most of them weren't formally caught. They were simply revealed the moment the tool wasn't available.

This is a trend, not a one-off

The specific circumstances here were unusual (a campus tragedy, a compassionate professor). The direction is not. Across higher ed, the response to AI is a steady move back toward in-person, proctored, and handwritten assessment — blue books, locked-down browsers, oral exams, in-class essays. The format that dominated 2020–2024 is retreating.

For students, the practical consequence is the one this site keeps hammering: the format of your exam decides everything, and the format is changing under you. A course that was take-home last year may be proctored this year. A weekly quiz you breeze through in Chrome tells you nothing about how the final will be administered. Assume the high-stakes assessment will be the one you can't shortcut, because increasingly, it is.

The honest lesson: know your format, and don't hollow out your understanding

Two things are true at once, and pretending otherwise is how students get burned:

  1. Unproctored online quizzes and homework still exist, and they're mostly low-stakes busywork. Using tools to move through them faster is a real, common workflow.
  2. The exams that decide your grade are increasingly in-person or proctored, and in those, you are on your own — your actual understanding is the only thing in the room.

The mistake the Brown students made was letting (1) erase their preparation for (2). They optimized the take-home number and skipped the learning, and the learning was the part that mattered.

Where ExamClutch fits — honestly

We build ExamClutch, so here's the straight version rather than a pitch that this story would make ridiculous anyway.

ExamClutch is a Chrome extension for unproctored online quizzes — the ones you take in ordinary Chrome, where the LMS only logs browser events. It reads questions from the page and helps you move through them without tab-switching or copy-paste. That's the layer it's built for.

It does not work in the situation this story is actually about:

So the honest positioning, using this story as the example: a tool like ExamClutch is for clearing the unproctored busywork so you have more time to actually learn the material — not for replacing the understanding that an in-person final will demand. The students who scored 96% and then fled the room did the opposite. Don't be the person whose two scores describe two different people.

If your quizzes are unproctored and you want to save time on them, see how ExamClutch works. If your finals are moving in-person, the smart use of that saved time is learning the material — because that's the exam that counts, and it's the one no extension can sit for you.

FAQ

What happened at Brown, briefly? A professor switched his final from take-home to in-person after a suspiciously high take-home midterm average (96%). The in-person average fell to 48.6%, a course record low, and many top-scoring students dropped or didn't show.

Did the professor prove students used AI? He didn't run a formal integrity case in the reporting. He ran the exam through ChatGPT and found its answers resembled students' submissions, and the score collapse when the format changed spoke for itself. The evidence is circumstantial but stark.

Does this mean online exams are going away? Not entirely, but high-stakes assessment is clearly shifting toward in-person and proctored formats specifically because of AI. Expect more of it, not less.

Can a tool like ExamClutch help on an in-person or proctored final? No. In-person exams have no browser, and proctored/lockdown exams don't load extensions. ExamClutch is only for unproctored online quizzes.

What's the actual takeaway for a student? Know which format each assessment uses, and never let an easy take-home score substitute for understanding you'll be tested on in a room. The gap between those two is exactly what got exposed here.

Sources

Ready to stop fighting your LMS?

Get ExamClutchSee pricing