The most clear-cut finding is fairly obvious and intuitive. If you tell people to write an essay with an LLM, they will obviously not remember what they “wrote” (or likely, wholesale copy/paste/edited from a chatbot) as well as someone *not* using an LLM.

The people recruited for this study are not only WEIRD (“Western, educated, industrialized, rich and democratic”) but are also all affiliated with Massachusetts universities. They likely understood the premise of the study and complied it:

The sample size was small and RIPE for p-hacking. This study measured MANY things (EEG results, survey responses, essay quality, N-grams, etc, etc… all across 4 sessions and many combinations of test conditions) and the results are p-hacked in every direction. like… this:

This study isn’t the first though, there was a study on Google effects on memory.