Teaching assistants (TAs) often have the task to compile exercise sheets for the students and need to come up with new problems for such sheets.

Let's assume that in the lecture, two methods for a certain use case were presented, and a TA is supposed to write a problem that should make the students use the second method. The first one should be cumbersome to use, so that the students can figure out themselves that the second method is the one to use in this case. Also, the problem should be somewhat motivated by an arbitrary problem from practice, so that solving the exercise sheet problem also teaches the students that the second method has its use in practice.

Now some TAs will be naturally good at such tasks, and others do not know where to start, despite knowing all of the relevant scientific results. How to teach TAs the creative skills to performing such tasks?

Note: This is a follow-up question to an earlier question on a related topic, which went without answer ideas. This new question may be of more universal interest, and I suspect that if good answers can be given, they are the same in both cases anyway.

  • Is the student eager to learn? If so, you could try giving examples or detailed instructions; after iterating with you a couple of times, they'll start to internalize the goal. But in my experience, most people who don't "naturally" have this skill have very little interest in learning it, they just do things as quickly as possible and never improve -- in this case, I would just do it myself and not waste time on them.
    – cag51
    Apr 28, 2019 at 16:48

3 Answers 3


I'm not sure why you find this to be a special problem. We basically know how to teach people things. Just do it in the ordinary way.

If no framework for understanding how to do a task exists then you need to develop one. If you know how to do it, but are unique, then you need to capture what you do in some way that can be transmitted.

Then you "teach" it by presenting the framework and giving some exercises so that it can be practiced and giving some feedback on how people do.

It is a standard teaching problem.

If no one knows how to do it, then you need to do something like research to discover how it can be done and then operationalize that research. Again, a standard problem, with a standard solution.

But, one thing I would suggest is that you use some sort of team learning so that those who do it naturally can pass skills to those who struggle with it. Pairing is a good approach and one I often recommend for such tasks.

Beyond that, it will likely take the examination of good and bad examples, using discussion (by the TAs) of why some things work and others don't.

But it is practice and feedback that are the key to learning almost anything, hence key to teaching it.

Don't take the word "framework" too literally.

Also, the task itself is ill defined here. Expect that no one will be perfect at it and that the "examples" created will sometimes fail. But there are ways, such as pre-testing or other evaluative methods, to deal with that as well.

  • Ok, thanks - the team-based approach is definitely a good idea. About the suggestion to capture in a framework what I am doing myself: I wish I knew how to do that. Puzzling together examples until they work is to me like walking - there is no known answer how exactly it works (witnessed by the fact there is a whole research community in robotics about this) even though many people are doing it. The practice and feedback approach can of course be applied again, as you wrote. I wish there was teaching material on this.
    – DCTLib
    May 3, 2019 at 15:59

A bit of background, so you can see where I am coming from: I recently created exercises for two courses on applied physics:

  • one about a subject I was mostly very familiar, where I had to create exercises from scratch,
  • one about a subject that was mostly new to me, but for which I had with an existing collection of exercises targeted at a different audience for inspiration.

Almost all the exercises I created came into being via one of the following:

  • I simplified the hell out of a prominent scientific application of the topic.

  • I took an existing exercise and modified it for the target audience.

  • Through my knowledge of the first subject, I could almost immediately recall a bunch of examples for a given topic, and could choose something fitting for an exercise. Note that I would consider my knowledge here to go beyond “knowing all of the relevant scientific results”, since it also involved being familiar with applications, not so prominent results, etc.

    Not seldom these were cases where I had already thought that the case would make for a good exercise years before. Also, some of these examples directly came from my own research. Fun fact: Some of these cases were problems I had first encountered in SE questions and in one case debunking a wrong SE answer.

  • Somebody experienced with the subject suggested a topic to me.

  • Discussing the topic in general, rejected exercise ideas, etc. with somebody knowledgeable about the topic.

Note that thinking very hard about the topic is not on this list. If it were in the case you are outlining (relevant applications), one could easily come up with a new research topics just like that. Instead, my standard workflow for the second course would be to delve into the topic at hand, reading the lecture notes, textbooks, online material, papers, etc. – for the first course I had already done this and at worst had to jog my memory. In particular, creating exercises for the second course was considerably more difficult, even though I had more material to work from.

Once I had recognised a good topic for an exercise as such, fleshing it out to an actual exercise was mostly legwork.

So, in your case, the following may help:

  • Consider if it is subject knowledge that your TA is lacking. If yes, encourage them to spend more time on studying the subject.

  • Suggest topics to the TA and task them with finding out whether the topic is suitable for an exercise and, if yes, fleshing it out. This relieves your TA of the problem of finding suitable topics and trains their skills in recognising good topics as such and doing the legwork.

  • Discuss the topic with your TA. If there are no exercises to discuss about, talk about what skills you want the students to learn, etc.

Finally beware that creating exercises mostly from scratch is a considerable time sink and may cost more time than your advisees can reasonably spend on teaching.


Creating an exercise is a harder task than solving it. Creating an exercise with restrictions on how a solution should go is even harder. The skill comes only after you get really proficient in solving problems and that requires a few years, IMHO, so it is hopeless to try to teach it in one semester or so. When you look at a typical medium level exercise, you see in a split second what techniques should be applied and in another split second which approach is most promising (though it may take you some noticeable time to actually carry it out). Some graduate students are like that too and then they can play the problem design meta-game of going from an approach or a combination of approaches to some particular setup, but some others are still at the level of "I can figure out how to use this method of solution on this problem, but my only way to figure out which method to use is to randomly try all methods I remember until I either get through with one of them, or get stuck on everything". Those are just not ready for problem composition and, IMHO, won't be ready for it until their problem solving skills improve dramatically. So for them I would assign a simpler task of choosing an appropriate problem from a set of textbook exercises (which requires just solving the problems and seeing whether the method you used is the one that was specified). In the worst cases, just compose the exercises yourself and give up on the hope to teach in a few weeks the skill that takes a few years to develop.

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .