Kosmyna and colleagues at MIT Media Lab, in a recent preprint:
"The use of LLM had a measurable impact on participants, and while the benefits were initially apparent, as we demonstrated over the course of 4 sessions, which took place over 4 months, the LLM group's participants performed worse than their counterparts in the Brain-only group at all levels: neural, linguistic, scoring."
The study gave 54 college students EEG headsets and had them write essays over four months. One group used only ChatGPT, another only Google, and a third used nothing but their own brains. The brain-only group showed 55% stronger neural connectivity in regions tied to memory encoding and executive function. The ChatGPT group? Weakest across the board.
Immediately after writing, 83% of ChatGPT users couldn't quote a single sentence from their own essay. Zero provided a correct quote. In the other groups, only 11% failed this test.
The researchers call this "cognitive debt":
"As the educational impact of LLM use only begins to settle with the general population, in this study we demonstrate the pressing matter of a likely decrease in learning skills."
When some ChatGPT users later wrote without the tool, their brains didn't bounce back. They even reused ChatGPT's vocabulary patterns. Meanwhile, brain-only students given ChatGPT for the first time showed increased neural activity. They had built something to integrate with.
The lesson: struggle first, then integrate. Not the reverse.