Some suggested changes for dealing with Chat-GPT, in no particular order. Note that these are based on our discussions in (mathematics and natural sciences) teacher education, so they are not fully adjusted for pure/applied mathematics, so use your own judgment.

 1. Present a wrong proof or calculation provided by Chat-GPT and ask them to find an explain or fix the mistakes. (In general it is a good idea to vary the kinds of exercises you use; not only prove this and calculate that, but also other types.) Or you peer evaluation as an exercise.
 2. Have an exercise that requires use of both pictures and text, or where the answer is supposed to be a picture. Draw a comic about an epsilon-delta proof or explain what kind of argument a geometric drawing might contain the idea of. Easier in analysis, geometry or graph theory, but often visualizations are also useful elsewhere.
 3. Have longer exercises in several steps, where students deliver something, get feedback and adjust accordingly, delivering again.
 4. In university mathematics it is probably fairly hard to get writing exercises where one reflects on their own experiences, but if you manage it, Chat-GPT is of limited help. Self-reflection and similar skills are useful also in mathematics.