In a previous post, I mused about the potential of an LLM that used both a publically trained model and a privately trained model. Schools could train assistants to mirror each of their teachers. Instead of one massivaly trained oracle of knowledge, schools could train assistants for specific grade levels. Students would receive the scaffolding they need for personalized assistance (see inferior and assistants below), but not an uber-assistant capable of answering all of their questions.
I see three ways students will use AI assistants:
Leverage AI. Tech-savvy students could write programs to integrate (embed) their own data to produce more robust responses (i.e., fine tuning) than prompt engineering
Inferior assistants. AI assistants could be created to help students take on the role of a teacher like an older sibling would do with a younger sibling. An example could be proofing and analyzing responses.
Peer assistants. AI assistants could be created to provide some automation of lesser cognitive skills to foster higher order thinking. An example could be providing response stems or aggregating research results.
Regargless of how AI assistants come to fruitition, chat interfaces have proven to be both engaging and excellent mechanisms to collect what students think academically and non-academically.
