Create assignments that require students to show stages of their work (outlines, rough drafts, etc.). These strategies can lead to deeper learning, provide instructors with more regular insight into student work, and increase the likelihood that the final product reflects student efforts rather than a copy of others’ work, human or artificial. The following examples offer a starting point. (Source: UMich)
What about using graded, hand-written, in-class assignments?
While this would certainly solve the problem of students using ChatGPT, such exercises can be very problematic for a range of students, such as those who need extra time to complete assignments or whose first language is not English. One thing that we have learned from the pandemic is that students are grateful to have moved away from timed, high-stakes assignments.
Stress-Test Your Assignments
One way to check whether your assignments are "exposed" to AI and large language models is to stress-test them. The Writing Institute of Pittsburg walks through how to stress-test your own assignments and also provides example assignment revisions from an undergraduate writing course. (Pitt.edu)