John Mandrola, MD

40 out of 86 Brown students scored a perfect 100 on their midterm. Then the professor moved the final in person, and 22 of those perfect scorers never showed up again. He'd suspected AI cheating from the start. The take-home midterm was deliberately harder than usual, yet the class averaged 96 when the historical range is 65 to 80. Some answers contained odd phrasing that matched what ChatGPT produced when he ran the questions through it himself. Roberto Serrano has taught economics at Brown for 34 years. He filed no accusations. He announced the final would be in person, count for half the grade, and that if the two distributions didn't match, the final alone would determine grades. Then the exodus. 27 students never showed up. 22 of them had perfect midterms. Of the 59 who did show, 19 failed. Several signed the exam and turned it in blank. The average fell from 96 to 48, the lowest in the course's history. He never needed a plagiarism detector. The cheaters identified themselves by walking away. A grade distribution became a confession. Here's the part nobody's sitting with. Serrano proved it. He sent the distributions to Brown's dean and provost. The provost never responded. The academic committee's reply amounted to calling it "a wake-up call." The students who bailed before the final walked away clean. Every university in America is now grading two populations, students and students plus ChatGPT, on one curve. The honest kids in Serrano's class watched a 96 average get set by machines, then sat a real final against it. The cheaters lost nothing. That's the incentive structure now, and it grades itself.

Full disclosure: I believe this post by Dr. Kola. But just out of curiosity I asked Open Evidence: "Were clinical trial data manipulated for the drug avacopan (Tavneos) for ANCA vasculitis?" It's answer was a resounding "No" as shown below. I pasted Anish's tweets into OE and it told me many of the statements "could not be verified". If we are going to rely on AI like Open Evidence going forward, it's interesting that it does not acknowledge these claims. A practicing doc prescribing Tavenos would receive no feedback from OE that the drug's approval data was problematic.