Building Gist
I’ve been working on something for the past few months now called Gist. Gist is an AI-powered data tool for teachers that ingests all student-submitted work and turns it into quantitative, trackable data, as well as qualitative, actionable insights.
The problem
Teachers face an extremely difficult task in deciding how and what to teach to their students. They only have students in their class for an average of 39min/day. Additionally, grading is a long, tedious process that sucks teachers’ time outside of the classroom. Because of this, the same lessons are often taught to students year after year, with little to no development or adjustment for any given class.The solution
AI has the potential to revolutionize this issue. I’ve been experimenting with LLMs for quite a while now (I remember GPT-2 releasing), and believe I’ve found a solution to the above problem(s): synthetic data. Gist will have access to all student work, as well as assignment data (instructions, rubrics, etc.). For each submission, it will come up with a suggested ‘grade’ for the teacher, as well as qualitative feedback on every aspect of the rubric/assignment details. Then, these grades and justification (which can be edited by the teacher) will be fed into another prompt, along with a batch of other students from the same class to identify commonalities between them. This tree (see below image) continues until eventually the teacher can view broad commonalities in errors made by their students on any assignment, and adjust their lecture to help those students.