20260809
#225
“When you’re young you think that time moves forward. At 80 you know that it doesn’t, it stands still. We’re the ones that move.”
— Bob Dylan, via workfutures.io“Technology is legislation. Introducing a new technology is equivalent to imposing new global economic, legal, or societal policies with major consequences — and we don’t know in advance what that policy is going to be.”
— Karl Schroeder, Unapocalyptic
People: Fatherhood rewires the brain. Capitalism’s approval below 50%. Summer penis, wtf. Plastic surgery nightmare.
AI strat: China’s Kimi K3 matches frontier models in some benchmarks. GenAI drives retail traffic: nearly 2% of Walmart and Target web visits now come from AI agents. US export controls force Anthropic to restrict model access.
Security: Agents find real bugs. GRU intelligence on the marketplace. Drone killing a moth, soon mosquitoes?
Society: AI drafts manager objectives. Africa’s infrastructure gap. Chalance. Christian interior design.
Tech: GPU demand rebounds. Solar’s cost paradox: PV panels remain on an aggressive learning curve.
Foresight: Technology as legislation.
Random: Human curation in crisis (he says ^). An unlikely time capsule (Paul Lukas).
AI: At Brown University, 56 of 59 students failed — a professor suspected AI use on papers, ran a closed-book final; scores collapsed by up to 100% for 56 students; three performed as expected. The coming divide.
The Closed Book
The professor had taught economics for twenty-two years and had never felt the need to specify, in writing, that the final examination would be closed book. It had always been closed book. That was what a final was.
He specified it this year.
He did this quietly, in a footnote of the syllabus distributed at the start of term, and again in the email with the revision guidelines, and a third time in the exam hall briefing. He was not trying to hide it. He just did not particularly want to discuss it.
Fifty-nine students. He had been suspicious of the papers for six months — not because they were wrong, but because they were right in a way that felt practiced. A smoothness he recognised from grant applications written under pressure and not from second-year undergraduates working on their first extended economic analysis. He said nothing at the time. He noted it. He designed a final that could not be assisted, and he waited.
The results came back in three days.
Fifty-six scores collapsed. Some by a third. Several by sixty percent. Four by a hundred. Three students performed exactly as their previous work had predicted: competently, without drama. One of them had submitted a midterm paper he suspected an AI had partly written. She still, it appeared, knew the material.
He sat with this for longer than was professionally necessary.
He had a colleague at the medical school, Isabelle, who had recently published work on neuroplasticity and fatherhood — the finding that becoming a father physically reorganised the male brain, thickening the cortex in regions associated with empathy and sustained attention. The change was not metaphorical. It was anatomical. You were a different person on the other side of the experience than before it: measurably so, in tissue.
She came by his office that afternoon. He showed her the exam results without explanation.
She looked at them for a moment. “What did they do instead,” she said. “Instead of learning the economics.”
“They delegated the understanding,” he said. “And kept the transcript.”
She turned the data over. “The brain changes in response to what we actually do. Not what we intend to do, not what we simulate doing. What we practice, we become.” She set the paper down. “These fifty-six students practised the performance of economics. They’re now very capable at that. It’s just a different skill than the one they enrolled to learn.”
He asked whether the brain could recover.
She considered this. “You’d have to give it something to do,” she said. “Something it couldn’t delegate.”
Outside his window it was a summer evening of the kind that seems to believe it will last forever, which it won’t. A percentage of Americans read for pleasure, he had recently read, on any given day. Sixteen percent, as of 2023. Down from twenty-eight in 2004. The trend had begun before AI arrived. Something had already been changing before anyone named it.2
He thought about the three students who had passed. He did not know what they had done differently — whether it was stubbornness or discipline or some third thing that had no module attached to it. He could not publish this. There was no control group, no clean methodology, no IRB approval for what amounted to an accidental experiment. What he had was an observation.
He wrote on the inside cover of his notes: A grade proves you can pass the test. The test proves nothing about what happens in the absence of one.
He thought about the student who had submitted a partly AI-assisted midterm and still passed the closed-book final. She had used the tool without becoming dependent on it. He wasn’t sure whether that was wisdom or habit or luck. He intended to find out.3
He did not change the syllabus the following year. He changed the assessment. He did not announce in advance that it would be closed book. He specified it, briefly, in a footnote.
He had been teaching for twenty-three years. He had always believed that what you learned was a function of what you did with your attention over time. He still believed this. He was less certain what to do with it.
The summer evening continued, implausibly, outside the window. A bird was singing something repetitive and confident in the campus trees. It did not appear to have revised.
1 She would pass every subsequent closed-book exam without difficulty and graduate near the top of her class. The midterm AI use had apparently been the equivalent of checking her working rather than replacing it — a shortcut that left the understanding intact. This is not the typical outcome. It is, however, what made her interesting.
2 The 28% figure from 2004 was also, in retrospect, not good. Cognitive engagement researchers had been worried for thirty years. The concern had been legible for a long time before anyone started writing newsletters about it.
3 He never taught a student quite like her again. Or possibly he did, and the assessment design made it impossible to tell.