Efficient mass creation of ismorphic physics problems assisted by GPT-3 and Wolfram Alpha

One of the most effective ways to deter the use of resource sharing websites such as Chegg is to create problem banks that contain large numbers of isomorphic problems. We have developed a method of efficiently and systematically creating many isomorphic numerical input problems base on a single template problem, assisted by multiple technologies. First, the creators drafts several problem scenarios that are considered equivalent, and under each scenario, several variations that are considered isomorphic such as changing the direction of forces or velocities. Second, a short text prompt that includes the key information is created for the template problem, and the prompt/problem pair is input into GPT-3, a natural language processing model. New problem body text can be generated by inputting new prompts and making small edits, and the knowns and unknowns can be rotated for each variation to create more problems. Third, problem figures are created using scalor vector graphics, which can be easily modified for new situations. Finally, a solution can be quickly generated using an open computational platform Wolfram alpha. The generated problem figure and problem text can be directly inserted into popular online systems such as Canvas, and significantly reduce the time to create new isomorophic problem variations.