We consider the benefits and drawbacks of using generative AI to help teachers and students; Eran Malach argues that auto-regressive next-token predictors are powerful learners.
I appreciate the effort to push beyond more reporting about cheating scandals. Magic School is an interesting project. Its business model functions through teacher ambassadors. Teachers sign-up to get access to curriculum on the condition that they share that curriculum with other teachers at their home institutions. At the moment, Magic School has over 10000 teacher ambassadors in over 1000 schools around the US.
I also appreciate the commentary on access and equity. As all current forward-facing products for teachers and students run off GPT architecture, they bring to the playing field all of GPT's implicit biases, inaccuracies, limitations, etc. While it is exciting that teachers now have products to create quizzes for them, for instance, these quizzes will most certainly be tilted through a particular interpretive filter--not necessarily the teacher's own---despite best efforts through prompt engineering. I personally am suspicious of this big push towards teacher efficiency without any real analysis of what the costs are of that efficiency. Not many folks are pausing to have that conversation, so kudos to the Gradient for stopping to pause.
As to AI tutorial systems privileging particular learning styles and particular kinds of learners, there is a growing body of research to support this claim. Surprisingly, most tutorial systems are set up for neurotypical learners. Learners with intellectual and cognitive diversities usually have to wait for later editions of these systems to have their needs met, and by then, the neurotypical learners have already gained an advantage through the first-wave of support. In this way, the notorious technological gap trickles down through systems ostensibly designed to close the gap. Check out the Department of Education's Report on AI and the Future of Education for more info. https://www2.ed.gov/documents/ai-report/ai-report.pdf. Section on "Learning."
Thanks for sharing the report, and your points about privileging particular learning styles! Indeed, it’s worrying when even good-faith efforts to improve things seem likely to exacerbate existing inequalities or ignore important costs.
And yet forward we must move. Think of a world where every student gets individualized homework. We are only a stone’s throw away from that world.
I appreciate the effort to push beyond more reporting about cheating scandals. Magic School is an interesting project. Its business model functions through teacher ambassadors. Teachers sign-up to get access to curriculum on the condition that they share that curriculum with other teachers at their home institutions. At the moment, Magic School has over 10000 teacher ambassadors in over 1000 schools around the US.
I also appreciate the commentary on access and equity. As all current forward-facing products for teachers and students run off GPT architecture, they bring to the playing field all of GPT's implicit biases, inaccuracies, limitations, etc. While it is exciting that teachers now have products to create quizzes for them, for instance, these quizzes will most certainly be tilted through a particular interpretive filter--not necessarily the teacher's own---despite best efforts through prompt engineering. I personally am suspicious of this big push towards teacher efficiency without any real analysis of what the costs are of that efficiency. Not many folks are pausing to have that conversation, so kudos to the Gradient for stopping to pause.
As to AI tutorial systems privileging particular learning styles and particular kinds of learners, there is a growing body of research to support this claim. Surprisingly, most tutorial systems are set up for neurotypical learners. Learners with intellectual and cognitive diversities usually have to wait for later editions of these systems to have their needs met, and by then, the neurotypical learners have already gained an advantage through the first-wave of support. In this way, the notorious technological gap trickles down through systems ostensibly designed to close the gap. Check out the Department of Education's Report on AI and the Future of Education for more info. https://www2.ed.gov/documents/ai-report/ai-report.pdf. Section on "Learning."
Thanks for sharing the report, and your points about privileging particular learning styles! Indeed, it’s worrying when even good-faith efforts to improve things seem likely to exacerbate existing inequalities or ignore important costs.